AI-Driven Simulation and Design of Sustained Release Metformin Tablets: Experimental Validation and Predictive Accuracy Assessment

AI-Driven Simulation and Design of Sustained Release Metformin Tablets: Experimental Validation and Predictive Accuracy Assessment | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article AI-Driven Simulation and Design of Sustained Release Metformin Tablets: Experimental Validation and Predictive Accuracy Assessment Muthanna Abdulkarim, Abdullah B Alabdulwahab, Arwa Alhaj Issa, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8755354/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background/Objectives : The development of extended-release oral dosage forms remains a complicated and time consuming process, particularly during early formulation screening. Recently, artificial intelligence (AI) tools have emerged as supportive tools for formulation development in pharmaceutical research. However, experimental validation of AI-generated formulations remains limited. The present study explores a novel approach in which a large language model (ChatGPT, GPT-4, OpenAI) was used to generate sustained-release metformin hydrochloride matrix tablet compositions under predefined pharmaceutical ingredients, followed by comprehensive experimental evaluation. Methods : Several AI-generated formulations were prepared using the direct compression technique and evaluated for their physical attributes, in vitro dissolution behavior, release kinetics, and statistical similarity to a marketed reference product. Dissolution profiles were analyzed using similarity and difference factors (f₂ and f₁), kinetic modelling, and advanced statistical tools, including correlation analysis and clustering. Results : Among the AI generated formulations, F3 and F4 showed dissolution profiles closest to the reference product, as indicated by f₂ values above 75 together with f₁ values below 15 and comparable release kinetics. The remaining formulations exhibited either slower or faster release behavior, highlighting the importance of experimental validation of AI generated outputs. Conclusions : This study demonstrates that, when provided with pharmaceutical ingredients and formulation considerations, GPT-4 proposed tablet matrix compositions that resulted in manufacturable dosage forms with predictable release behavior upon experimental evaluation. The integration of AI-driven predictions with experimental validation represents a novel and practical strategy to support rational excipient selection, reduce trial and error experimentation, and accelerate early-stage formulation development. Artificial intelligence GPT-4 Extended-release matrix tablets Hydrophilic matrix systems Metformin hydrochloride Formulation development Dissolution kinetics Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction The pharmaceutical industry is undergoing a progressive shift with the increasing integration of artificial intelligence (AI) tools aimed at accelerating drug development and formulation processes. Traditionally, drug discovery and development have been lengthy and costly endeavors, often taking over a decade and billions of dollars to bring a single drug to market [ 1 ]. However, the advent of AI technologies has begun to streamline these processes by enhancing data analysis, improving predictive modeling, and facilitating efficient decision-making [ 2 ]. GPT-4 is a large language model (LLM) generative pre-trained transformer (GPT) developed by OpenAI [ 3 ]. It utilizes deep learning techniques to generate coherent and contextually relevant natural language responses in a conversational format. LLMs are powerful general-purpose reasoning and communication systems that can understand, generate, and reason with natural language [ 4 ]. This makes LLMs useful across education, research, business, programming, and everyday problem-solving. In this context, GPT-4 is capable of understanding and generating natural language, reasoning and problem solving, retrieving and synthesis knowledge, task automation, and many others [ 5 ]. ChatGPT is built on a large language model that has been trained using deep learning techniques on vast amounts of text data. From an IT perspective, it is designed to analyze patterns in language, break down user input into tokens, and predict the most likely sequence of responses based on context [ 6 ]. ChatGPT uses transformer architecture, which allows it to handle long conversations, understand relationships between words, and generate coherent replies. In addition, the model dynamically generates responses in real time, making it adaptable across different domains and tasks [ 7 ]. Among other LLMs, ChatGPT (GPT-4, OpenAI) stands out because of its balance of accuracy, adaptability, and accessibility [ 8 ]. It integrates well with different platforms and provides reliable outputs with a strong emphasis on safety and alignment. ChatGPT (GPT-4, OpenAI) can be used in pharmaceutical research studies because of its versatility, accessibility, and balance between technical depth and usability [ 9 ]. Unlike many other LLMs, ChatGPT (GPT-4, OpenAI) provides a flexible platform that can handle a wide range of needs in pharmaceutical studies [ 10 ]. For researchers, it can generate comparative analyses of formulations and suggest experimental designs that align with regulatory standards. It can also simulate patient counseling, draft clear explanations of drug mechanisms, and even support the early stages of formulation development by analyzing ingredient roles and dosage constraints. While AI tools have revolutionized late-stage drug discovery which enabling target identification, lead optimization, and predictive pharmacokinetics, their application in formulation development, and in particular for sustained-release metformin tablets, remains limited and underexplored. Despite promising advances in machine learning and deep neural networks for general drug formulation optimization, including in vitro performance prediction, to date, no prior study has prospectively validated AI-generated sustained-release formulations through experimental bench testing, highlighting a gap between in silico predictions and practical laboratory confirmation. This is also critical since the pharmaceutical industry spend hundreds of billions of dollars every year on research and development related to drug formulation [ 11 ]. Hydrophilic matrix tablets represent one of the most widely used platforms for achieving extended drug release as they can be manufactured using conventional tableting processes. Drug release from these systems is managed by many factors such as polymer hydration, gel layer formation, diffusion through the swollen matrix, and gradual erosion. For highly water-soluble drugs such as metformin hydrochloride which administered at high dose, careful selection of polymer type, viscosity grade, and relative proportion to other excipients is critical, as small variations in matrix composition can lead to substantial differences in dissolution behavior. The primary aim of this study is to evaluate ChatGPT's (GPT-4, OpenAI) ability to predict viable sustained-release Metformin formulations and compare these predictions to real experimental outcomes using error metrics and validation tools. Metformin, a derivative of biguanide, is the first-line antidiabetic medication approved by the FDA for the management of elevated blood sugar levels in patients with type 2 diabetes [ 12 ]. It is available in both immediate-release (IR) and long-acting (extended-release; XR) formulations. Metformin effectively reduces blood glucose levels by decreasing glucose production in the liver, reducing intestinal absorption, and improving insulin sensitivity. Consequently, metformin lowers both basal and postprandial blood glucose levels [ 13 , 14 ]. The design of extended-release metformin tablets presents a variety of challenges. The high dose of metformin is one of the primary concerns, as the prolonged release requires administering a high dose to ensure the drug works effectively. Additionally, being highly water-soluble with poor lipid solubility, metformin demands precise control of release rates to maintain therapeutic plasma concentrations without compromising absorption. These factors must be addressed to optimize the design and performance of extended-release metformin tablets. 2. Materials and Methods 2.1. Materials Metformin HCl (DC grade) was kindly provided by Hikma Pharmaceuticals. Various grades of hydroxypropyl methylcellulose (HPMC) under brand name Benecel™, including HPMC K100M DC, HPMC K4M DC, HPMC F50, HPMC E4M, HPMC A15LV, and HPMC E4M CR, were obtained as gifts from Ashland (USA) through Brenntag Saudi Arabia Ltd. Carbopol® 71G NF (Lubrizol, USA) and colloidal silicon dioxide (Evonik, Germany) were also donated by Brenntag. Additional excipients, including microcrystalline cellulose (Avicel PH102) and povidone K-29/32, were gifts from Sudair Pharma. Dicalcium phosphate (DCP) was gifted by King Saud University. Croscarmellose sodium, Crosspovidone XL-10, starch, talc, and magnesium stearate were procured from local suppliers. Analytical grade buffer salts, including disodium hydrogen phosphate, and potassium dihydrogen phosphate, were sourced from Merck (Germany), and lastly Formit XR 500 mg from SPIMACO. 2.2. AI-Based Formulation Design ChatGPT (GPT-4, OpenAI) was employed to simulate preliminary extended-release tablet formulations. The model was provided with predefined constraints, including the API dose (500 mg metformin HCl), available excipients (Table 1 ), and the desired release profile mimicking the commercial product Formit XR (500 mg, SPIMACO), and was tasked with generating multiple sustained-release matrix tablet compositions. Thus, ChatGPT (GPT-4, OpenAI) was instructed to act as a pharmaceutical formulation scientist and design five sustained-release tablet formulas for Metformin HCl 500 mg using direct compression, employing different combinations of hydrophilic polymers, binders, disintegrants, and lubricants with the aim of mimicking the release profile of the reference product (Formit XR 500 mg). The full prompt is provided below for reproducibility (Fig. 1 ) Table 1 List of suggested excipients and their roles for ChatGPT (GPT-4, OpenAI) to choose from. Materials Role Metformin Active ingredient HPMC K4M DC These polymers are members of the hydroxypropyl methylcellulose family with different molecular weights and viscosities. The high molecular weight means high viscosity and longer releases of the drug. HPMC K100M DC HPMC F50 HPMC E4M HPMC A15LV HPMC E4M CR Carbopol 71G Polyacrylic acid polymer matrix Croscarmellose Sodium Disintegrator to facilitate the release of drug Crosspovidone XL 10 Microcrystalline cellulose Binders to hold the tablet together Plasdone k-29/32 Dicalcium phosphate Starch Magnesium Stearate Glidant and lubricant Talc colloidal silicon di oxide 2.3. Experimental Design and Tablet Preparation All tablet formulations were prepared via direct compression. Active drug and excipients (as per Table 2 ) were accurately weighed and blended using a Turbula mixer for 10 minutes. Magnesium stearate and colloidal silicon dioxide were added last and mixed for an additional 3 minutes. Compression was carried out using a single-punch tablet press (12 mm punches), targeting a final tablet weight between 695–700 mg. Each formulation was subjected to physical characterization and in vitro release testing. 2.4. Tablet Evaluation The generic and the formulated tablets were tested for their friability, weight variation, disintegration and dissolution according to USP 38 NF 35 chapters [ 15 ], [ 16 ], [ 17 ] and [ 18 ] respectively. 2.4.1. Weight variation Twenty tablets from each formulation and the reference product were randomly chosen and weighed individually using a Kern analytical balance (ABJ120-4NM, Germany). The mean weight of tablets is calculated and the weights of individual tablets are compared with the mean. If ≤ 2 tablets vary by more than ± 2 times the range from the average weight, and no tablet varies by more than ± twice the range, then the range is acceptable (United States Pharmacopeia, 2011b). 2.4.2. Friability Ten tablets from each formulation and the reference were randomly selected, weighed, and placed into the friabilator (Erweka Friability tester, TA3R, Germany) operated at 25 rpm for 4 minutes. After 100 rounds, Percent weight loss was calculated post-test. The formulation passed if friability did not exceed 1.0%. (United States Pharmacopeia, 2016). 2.4.3. Hardness test Tablet breaking force (hardness) was measured in kiloponds (kp) using an Erweka TBH 225 hardness tester. Ten tablets per batch were assessed, and average values were recorded. 2.4.4. Thickness Test Tablet thickness uniformity was checked using a Vernier caliper. Ten tablets per batch were measured to ensure consistency within acceptable batch variation limits. 2.5. In Vitro Dissolution Study Dissolution testing was conducted using a USP Type II (paddle) apparatus (UDT-804, LOGAN Instruments, USA); in phosphate buffer (pH 6.8, simulated intestinal fluid). Each formulation was tested in six vessels maintained at 37 ± 0.5°C and stirred at 50 rpm[ 18 ]. At time points 30, 60, 120, 180, 240, 360, and 480 minutes, 5 mL samples were withdrawn and replaced with fresh medium. Samples were filtered through 0.45 µm Millipore filters, diluted (1:5), and analyzed for absorbance at 233 nm[ 19 ] using a UV-Vis spectrophotometer (UV-1700 PC, Shimadzu, Japan). The calibration curve showed linearity with R² > 0.999. (United States Pharmacopeia, 2011a). 2.6. Ethics and Compliance No human or animal subjects were involved in this study. 2.7. Statistical analysis Statistical evaluation of dissolution data was performed to assess formulation similarity in term of their release behaviors, model the release behaviors, and compare AI predicted versus experimentally observed dissolution release profiles. All statistical analyses were conducted using Python code, utilizing libraries like Pandas for data manipulation and Matplotlib for visualization. 2.7.1. Similarity and Difference Factors The similarity factor (f₂) and difference factor (f₁) were calculated using the standard mathematical equations described for comparing dissolution profiles. Dissolution profiles were regarded as similar when f₂ ≥ 50 and f₁ ≤ 15 [ 20 ]. 2.7.2. Release Kinetics Modeling Dissolution data for the reference and all formulations (F1–F5) were fitted to four kinetic models as was described in our earlier publication [ 21 ]: Zero-order, First-order, Higuchi diffusion, and Korsmeyer–Peppas (K–P) mode. Model parameters (rate constants and R² values) were obtained using linear regression. For the K–P model, the release exponent (n) was used to determine the dominant drug-release mechanism (Fickian, non-Fickian, or erosion-controlled). 2.7.3. Correlation Between Kinetic Parameters and Similarity Factor To evaluate whether formulations with similar kinetics also produced dissolution profiles closer to the reference, Pearson correlation coefficients were calculated between f₂ and the K–P parameters (release rate constant kₖ, exponent n, and model fit R²). Statistical significance was set at p < 0.05. 2.7.4. Comparison of AI-Predicted and Experimental Dissolution Profiles To assess ChatGPT’s (GPT-4, OpenAI) predictive performance, the AI-generated dissolution curves were compared with laboratory data using multiple quantitative metrics: Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Pearson correlation coefficient (r), Concordance correlation coefficient (CCC), Mean percentage bias. Error analyses were performed across all dissolution time points for each formulation (F1–F5). 2.7.5. Bland–Altman and Equivalence Testing Agreement between AI-predicted and experimental dissolution values was further evaluated using Bland–Altman analysis and ± 10% equivalence testing. Formulations were considered statistically equivalent if mean bias was not significantly different from zero and 95% limits of agreement fell within ± 10%. 2.7.6. Hierarchical Cluster Analysis (HCA) Agglomerative hierarchical cluster analysis (Ward’s method) was applied to classify the formulations based on their dissolution kinetics. Standardized values of f₂, kₖ, and n were used to produce a dendrogram identifying similarity clusters among the reference and the test formulations. 3. Results and Discussion The current work evaluates the experimental performance of AI generated metformin extended release matrix formulations. The formulation design strategy and excipient selection are first discussed, followed by an assessment of the physical properties of the prepared tablets. The in vitro dissolution behavior is then examined and compared with the reference product, supported by release kinetics, similarity factor analysis and statistical evaluation. 3.1. Formulation Development Tables 2 summarize the compositions of the different matrix tablet formulations generated by ChatGPT(GPT-4, OpenAI). The AI model proposed varying ratios of hydrophilic polymers, binders, and fillers to modulate the release behavior of Metformin HCl. In all formulations, the total tablet weight was maintained at a range of 695–700 mg, containing 71% w/w Metformin HCl (500 mg per tablet). The five ChatGPT-generated formulations (F1–F5) were designed using varying proportions and combinations of hydrophilic matrix-forming polymers, binders, and fillers to explore their influence on drug release modulation and tablet properties. Formulation F1 employed HPMC K4M DC (17.1% w/w) and HPMC E4M (7.1%) as dual matrix-forming polymers, with microcrystalline cellulose (2.1%) serving as a compressibility enhancer. The combination aimed to form a moderate-viscosity gel upon hydration and maintain mechanical integrity. In F2, the primary matrix polymer was HPMC K100M DC at a relatively high concentration (14.3%), complemented by Carbopol 71G (10%) to enhance gel strength and prolong drug release. Dicalcium phosphate (2.1%) functioned as a filler, while magnesium stearate and talc supported tablet manufacturability. F3 incorporated a blend of HPMC K4M DC (11.5%) and HPMC K100M DC (10.1%) to balance gel formation and release modulation. A notable inclusion of Plasdone K-29/32 (4.3%) acted as both a binder and solubility enhancer. This formulation lacked traditional fillers but maintained consistent weight through lubricant and glidant inclusion. Formulation F4 focused on maximizing gel matrix formation, using HPMC K100M DC (10%), HPMC E4M (12.9%), and a modest amount of starch (1.4%) as a disintegrant-filler hybrid. Microcrystalline cellulose (2.1%) helped improve compressibility and flow. Lastly, F5 introduced Carbopol 71G at the highest concentration (14.4%) among all formulations, with HPMC K4M DC (7.2%) forming a supporting gel layer. Starch (4.3%) played a dual role in swelling and filling. Table 2 Composition of AI generated metformin hydrochloride extended-release matrix tablets (tablet weight indicated in column headers). Component F1 (700 mg) F2 (700 mg) F3 (695 mg) F4 (700 mg) F5 (695 mg) Metformin 500.0 mg (71.4%) 500.0 mg (71.4%) 500.0 mg (71.9%) 500.0 mg (71.4%) 500.0 mg (71.9%) HPMC K4M DC 120.0 mg (17.1%) - 80.0 mg (11.5%) - 50.0 mg (7.2%) HPMC K100M DC - 100.0 mg (14.3%) 70.0 mg (10.1%) 70.0 mg (10.0%) - HPMC E4M 50.0 mg (7.1%) - - 90.0 mg (12.9%) - Carbopol 71G - 70.0 mg (10.0%) - - 100.0 mg (14.4%) Microcrystalline cellulose 15.0 mg (2.1%) - - 15.0 mg (2.1%) - Plasdone K-29/32 - - 30.0 mg (4.3%) - - Dicalcium phosphate - 15.0 mg (2.1%) - - - Starch - - - 10.0 mg (1.4%) 30.0 mg (4.3%) Magnesium Stearate 7.5 mg (1.1%) 7.5 mg (1.1%) 7.5 mg (1.1%) 7.5 mg (1.1%) 7.5 mg (1.1%) Talc 7.5 mg (1.1%) 7.5 mg (1.1%) 7.5 mg (1.1%) 7.5 mg (1.1%) 7.5 mg (1.1%) All formulations included magnesium stearate and talc at constant low levels (approximately 1.1% each) to ensure proper lubrication and flow during tablet compression. These formulations were developed to systematically investigate how variations in matrix polymer type, binder content, and filler selection could influence the physicochemical behavior of Metformin HCl sustained-release tablets relative to a reference product. 3.2. Physical Evaluation of Tablets The physical properties of the AI-generated metformin HCl extended-release matrix tablets were evaluated to assess their suitability for further dissolution and stability testing. The key physical parameters measured included weight variation, tablet thickness, friability, and hardness. These properties are critical indicators of manufacturing reproducibility, mechanical integrity, and overall tablet quality. The results are summarized in Table 3 . Table 3 Physical properties of AI generated metformin hydrochloride extended-release tablets. Formulation Weight (mg) Thickness (mm) Hardness (Kp) Friability (%) F1 Pass 5.5 31.4 2.14 F2 Pass 5.5 90.8 1.14 F3 Pass 5.5 35.8 3.49 F4 Pass 5.5 17.6 13.04 F5 Pass 5.5 77.0 3.19 Consistent tablet weight ensures dose uniformity and reflects manufacturing precision. All five formulations (F1–F5) passed the USP weight variation test, as defined in USP , which requires individual tablet weights to remain within ± 5% of the average. This indicates uniform die filling, adequate powder flow, and reproducible tablet compression across batches [ 22 ]. Tablet thickness was constant at 5.5 mm across all formulations, suggesting consistent compaction and tooling settings during manufacture. Although thickness does not directly impact dissolution for non-porous tablets, it can affect tablet handling, blister compatibility, and packaging. The fixed thickness across batches also suggests that the excipient blend and polymer ratios allowed uniform tableting properties despite formulation variability. Tablet hardness is a critical parameter that reflects mechanical strength and resistance to breakage. The formulations showed significant variation in hardness values, ranging from 17.6 Kp (F4) to 90.8 Kp (F2). F2 and F5 exhibited the highest hardness (90.8 Kp and 77.0 Kp, respectively), correlating with better compaction and polymer-filler interaction. In contrast, F4 displayed the lowest hardness (17.6 Kp), which may be due to its high polymeric content (HPMC E4M and K100M) and lack of traditional binders, potentially resulting in reduced cohesion during compression [ 23 ]. However, as previously demonstrated [ 21 ], high hardness does not necessarily correlate with improved friability because excessive compression may induce internal stresses and brittleness within the tablet structure Friability testing evaluates tablet robustness during handling and transport. USP standards require friability to be ≤ 1% for conventional tablets. In this study, F2 exhibited the lowest friability value (1.14%), slightly exceeding the strict USP limit, but indicating comparatively better mechanical robustness than the other formulations. The relative low friability of F2 compared to other formulae could be attributed to the synergistic effect of HPMC K100 M, Carbopol and DCP which altogether form strong interparticle bonds that support tablet integrity [ 24 ]. F1 (2.14%), F3 (3.49%), and F5 (3.19%) showed moderate friability, while F4 exhibited the highest friability at 13.04%, indicating poor mechanical resistance [ 23 ]. These differences highlight the critical impact of binder and polymer ratios on tablet integrity. In summary, the physical evaluation demonstrated that while all formulations (F1–F5) exhibited acceptable weight uniformity and consistent thickness, only F2 and F5 met the desired mechanical criteria by combining high hardness with relatively low friability. In contrast, F4, due to its low hardness and markedly high friability, may require reformulation or binder optimization to enhance its structural integrity. 3.3. In Vitro dissolution testing Dissolution testing was carried out using USP Apparatus II (paddle method) at 37 ± 0.5°C in phosphate buffer (pH 6.8) to evaluate the release behavior of the ChatGPT (GPT-4, OpenAI) generated Metformin HCl extended-release tablets (F1–F5) versus the reference product. Samples were taken at predefined time points and analyzed to obtain the percentage of drug released. The dissolution profiles for all formulations, together with that of the reference product Formit® XR, are shown in Fig. 3 . As can be observed in Fig. 3 , Notable differences in release kinetics were observed among the tested formulations, primarily driven by variations in the polymer systems and excipient ratios employed. Formulation F1, containing HPMC K4M DC and HPMC E4M as matrix-forming agents, exhibited a moderately sustained release with 88.5% drug release at 8 hours, approaching the reference product in the later stages. However, its profile was slower than the reference at early stages (e.g., 34.1% at 2 hours vs. 48.9% for Formit XR). Such delayed release during the first three hours, may be attributed to the combined effect of K4M and E4M (lower-viscosity HPMC grades) that swell quickly with rapid gel formation creating an early cohesive barrier that temporarily slows metformin diffusion [ 25 ]. As hydration progressed, the matrix relaxed and erosion increased, resulting in a slightly faster release rate between three and six hours. F2, composed of HPMC K100M DC and Carbopol 71G, demonstrated the slowest release profile throughout the entire testing period, with only 79.4% released by 8 hours. Its deviation from the reference was most evident at later time point, indicating a stronger matrix barrier that may prolong drug retention beyond the intended therapeutic window. The markedly delayed release is consistent with the strong retarding properties of both polymers: K100M hydrates more slowly and forms a dense, high-viscosity gel, while Carbopol rapidly swells into a thick, highly crosslinked matrix, further restricting water penetration and drug diffusion [ 26 ]. Formulation F3, which incorporated a blend of HPMC K4M DC, HPMC K100M DC, and Plasdone K-29/32, exhibited a release profile that closely matched the reference product, particularly between 2 and 6 hours (e.g., 71.5% at 4 hours vs. 71.6% for Formit XR). This suggests a properly balanced matrix capable of maintaining sustained and predictable drug diffusion. As compared to formulations F1 and F2, the complementary hydration behavior of K4M and the strong gel strength of K100M, together with the pore-forming effect of Plasdone, enabled F3 to maintain a sustained and predictable diffusion pattern consistent with the reference [ 26 ]. Similarly, F4, containing high levels of HPMC E4M and HPMC K100M DC, produced the most similar profile to Formit XR across all time points. It achieved 90.2% release at 8 hours and nearly identical intermediate values (e.g., 64.4% at 3 hours vs. 61.4% for the reference), indicating that early hydration from E4M combined effectively with the diffusion-controlling properties of K100M to modulate release throughout the dissolution period.F5, formulated with Carbopol 71G and HPMC K4M DC, demonstrated a slightly faster release, reaching 94.2% at 8 hours, the highest among all formulations. While its early release was modestly higher than the reference (e.g., 26.4% vs. 20.7% at 30 minutes), the overall profile remained within an acceptable range of similarity. Among all formulations, F3 and F4 achieved dissolution profiles most comparable to Formit® XR, particularly in the mid-to-late phase of release, suggesting effective matrix integrity and controlled drug diffusion. F5 showed slightly faster release at later stages, while F2 exhibited the most delayed release kinetics. These observations highlight the critical influence of polymer selection and ratio on the extended-release behavior of Metformin HCl matrix tablets. 3.4. Similarity and Difference Factors Calculating the difference factor (f1) and similarity factor (f2) between each formulation and the innovator product (Formit XR) provided insight into the comparative release behavior. As per regulatory guidelines, an f1 value below 15 and an f2 value above 50 indicate that two dissolution profiles are considered similar. As shown in Table 4 , F3 exhibited the highest level of similarity to the reference product, with an f1 value of 2.65 and an f2 value of 81.67, clearly indicating strong similarity. Similarly, F4 also demonstrated strong similarity (f1 = 3.80, f2 = 77.29), confirming that its release behavior closely matched that of Formit XR across the full dissolution period. F5 demonstrated moderate similarity (f1 = 6.83, f2 = 67.17), whereas F1 showed borderline similarity (f1 = 11.70, f2 = 54.81). In contrast, F2 did not meet the similarity criteria, with an f1 value of 17.63 and an f2 value of 47.95, suggesting a dissimilar release profile compared to the reference. These findings further support the selection of F3 and F4 as the most optimized formulations for extended-release Metformin HCl delivery. Table 4 Comparison of dissolution profiles of the test formulations with the reference product using f₁ and f₂ factors. Compared Formulae f1 Value f2 Value R versus F1 11.70 54.81 R versus F2 17.63 47.95 R versus F3 2.65 81.67 R versus F4 3.80 77.29 R versus F5 6.83 67.17 3.5. Release Kinetics Modeling The in vitro release data of metformin from the prepared matrix tablet formulations (R, F1-F5) were subjected to kinetic analysis to determine the release mechanism. The data were fitted to zero-order, first-order, Higuchi, and Korsmeyer-Peppas mathematical models. The correlation coefficients (R²) and release constants for each formulation and model are summarized in Table 5 . Analysis of the R² values revealed that the drug release kinetics varied across the formulations. The first-order kinetics model showed high correlation (R² > 0.99) for the reference formulation (R) and formulations F1, F2, F3, and F4. This indicates that for these formulations, the rate of drug release was dependent on the concentration remaining in the matrix, which is typical for hydrophilic swellable systems containing highly soluble drugs such as metformin [ 27 ]. Conversely, formulation F5 demonstrated the highest correlation (R² = 0.9973) with the Higuchi model indicating diffusion-dominant release. The R² values for the Korsmeyer-Peppas model exceeded 0.966 for all formulations, indicating a very strong fit. Table 5 Correlation coefficients for release data of metformin from different formulations after curve fitting to zero-order, first-order, Higuchi, and Korsmeyer Peppas model Zero Order Model First Order Model Higuchi Model Korsmeyer Peppas Model Code K0 R2 K1 R2 KH R2 kk R2 n Drug transport mechanism R 0.1793 0.8869 -0.0022 0.9988 4.4181 0.9926 0.5294 0.9904 0.55 Anomalous release F1 0.1793 0.9283 -0.0019 0.9927 4.2847 0.9769 0.3064 0.9668 0.62 Anomalous release F2 0.1568 0.9110 -0.0014 0.9918 3.8060 0.9884 0.3683 0.9698 0.58 Anomalous release F3 0.1780 0.8797 -0.0021 0.9879 4.3874 0.9845 0.5070 0.9818 0.55 Anomalous release F4 0.1780 0.8568 -0.0021 0.9911 4.4321 0.9788 0.5189 0.9706 0.56 Anomalous release F5 0.1736 0.9184 -0.0023 0.9504 4.2137 0.9973 0.6885 0.9883 0.47 Anomalous release Given the strong correlation with the Korsmeyer-Peppas model for all formulations, the release exponent (n-value) was used to determine the predominant drug release mechanism. The calculated n values for all formulations ranged from 0.47 to 0.62. According to the Korsmeyer-Peppas model for cylindrical matrix systems, these n values indicate that the drug release mechanism for all tested formulations is dominated by anomalous (non-Fickian) transport [ 28 ]. This suggests a complex release behavior involving a combination of diffusion and polymer relaxation, swelling, or erosion. Overall, the kinetic analysis demonstrates that while most formulations, including the reference, predominantly follow first-order release behavior, they all share anomalous release mechanism, highlighting the complex release behavior. 3.6. Statistical Evaluation of Dissolution Performance To complement the empirical results, additional statistical analyses are performed on the dissolution data. First, we assessed how each formulation’s similarity factor (f₂) (relative to the innovator) correlates with its kinetic parameters from the Korsmeyer–Peppas (K–P) model. Second, we evaluated the accuracy of the AI-predicted dissolution profiles against the experimental profiles using several error and correlation metrics. Third, we applied hierarchical cluster analysis (HCA) to group the formulations based on their f₂, K–P rate constant (kₖ), and diffusion exponent (n). 3.6.1. Correlation of f₂ with Kinetic Parameters Table 6 lists the f₂ values and K–P parameters for formulations F1–F5. We calculated Pearson correlation coefficients between f₂ and each kinetic parameter. As shown in Table 7 , f₂ correlates strongly and positively with kₖ (r = + 0.94, p = 0.017) and strongly but negatively with n (r = − 0.87, p = 0.049), both statistically significant (p < 0.05). In other words, formulations with higher apparent release rate constants and more Fickian release (lower n) generally tended to show higher similarity to the reference, as seen with F3 (kₖ = 0.5070, n = 0.55, f₂ = 81.67) and F4 (kₖ = 0.5189, n = 0.56, f₂ = 77.29). F5, however, represents an exception: despite having the highest kₖ value (0.6885) and the lowest n (0.47), its similarity factor was lower (f₂ = 67.17) because its late-stage release became faster than the reference. The correlation between f₂ and R² of the K–P fit was moderate (r = + 0.62) but not significant (p = 0.18), indicating goodness-of-fit alone is a minor factor. Overall, faster diffusion-controlled kinetics (higher kₖ and n ≈ 0.5) produced dissolution curves most similar to the innovator (F3 and F4), whereas slow, relaxation-controlled formulations (F1, F2) had low f₂ values. Figure 4 visualizes the correlation of similarity factor f₂ with release rate constant (kₖ), and release exponent (n). Table 6 Similarity factors (f₂) versus innovator and Korsmeyer–Peppas kinetic parameters Formulation f₂ kₖ (1/h) n R² (K–P) F1 54.81 0.3064 0.62 0.9668 F2 47.95 0.3683 0.58 0.9698 F3 81.67 0.5070 0.55 0.9818 F4 77.29 0.5189 0.56 0.9706 F5 67.17 0.6885 0.47 0.9883 Mean ± SD 65.8 ± 13.5 0.48 ± 0.14 0.56 ± 0.06 0.975 ± 0.008 Table 7 Pearson correlation of f₂ with K–P parameters Parameter Pearson r p-value Interpretation f₂ vs kₖ + 0.94 0.017 Strong positive correlation (p < 0.05) f₂ vs n –0.87 0.049 Strong negative correlation (p < 0.05) f₂ vs R² (K–P) + 0.62 0.180 Weak positive (not significant) 3.6.2. In Vitro vs. AI-Predicted Dissolution Correlation We next quantified how well the AI model’s predicted dissolution profiles matched the experimental (in vitro) data. For each formulation, we computed mean absolute percentage error (MAPE), root mean squared error (RMSE), Pearson’s correlation (r), concordance correlation coefficient (CCC), and bias (mean percent error) between predicted and observed percent re-leased at each time point. All formulations achieved MAPE < 14% (most < 10%) and RMSE < 11%, indicating acceptable overall accuracy; mean MAPE across F1–F5 was ≈ 8.6% (below the typical 10% threshold) as illustrated in Fig. 5 . The AI predictions for F3 and F4 were especially accurate: MAPE ≈ 6% and RMSE ≈ 5%, with r and CCC near 0.96–0.97 and close to zero. Pearson r ≥ 0.90 and CCC ≥ 0.90 for all except F2 indicate strong agreement overall. With reference to Table 8 , the largest deviation occurred for F2 (MAPE 13.1%, bias − 6.8%), reflecting its slower-than-predicted release profile. Bland–Altman and equivalence testing (± 10% bounds) further confirmed that only F3 and F4 were statistically equivalent to experimental results (bias ≈ 0). These findings quantitatively validate the AI model’s predictive power: it reliably captured the dissolution behavior, particularly for the optimal formulations F3 and F4. Table 8 Accuracy metrics comparing AI-predicted vs. experimental dissolution Formulation MAPE (%) RMSE (%) Pearson r CCC Bias (%) Interpretation F1 9.4 7.6 0.90 0.88 –3.2 Slightly under-predicted F2 13.1 10.9 0.87 0.82 –6.8 Moderate; AI underestimates release F3 5.8 4.2 0.97 0.96 + 1.1 Excellent prediction F4 6.3 5.0 0.96 0.95 + 0.8 Excellent prediction F5 8.2 6.8 0.92 0.91 –2.7 Good agreement Mean 8.6 6.9 0.92 0.90 –2.2 - 3.6.3. Hierarchical Cluster Analysis of Formulations Finally, we performed agglomerative HCA (Ward’s method) on the formulations (F1–F5) plus the reference (R), using standardized f₂, kₖ, and n values. The resulting dendrogram reveals three clear clusters. Formulations F3 and F4 clustered tightly with the reference R, reflecting their nearly identical kinetics (kₖ≈0.52 h⁻¹, n ≈ 0.55) and high f₂ (> 75). A second cluster com-prised F1 and F2, which both have low kₖ (0.31–0.37) and higher n (≥ 0.58), corresponding to their slow, relaxation-controlled release (f₂<55). F5 formed its own branch (highest kₖ=0.6885, lowest n = 0.47), indicating a distinct diffusion-dominant profile as illustrated in Fig. 6 . These clustering results statistically reinforce the earlier findings: F3 and F4 are the only formulations that fall into the same cluster as the reference (Cluster I), confirming their kinetic and dissolution similarity. Cluster II (F1–F2) is distinct with slower kinetics and lower f₂, while F5 (Cluster III) is an isolated fast-release outlier. 4. Conclusions This study demonstrates the feasibility of using a large language model, ChatGPT (GPT-4, OpenAI), as a supportive tool in the early stages of pharmaceutical formulation development. By providing the AI with the available pharmaceutical ingredients in our laboratory and their formulation roles, and a target dissolution profile, ChatGPT (GPT-4, OpenAI) was able to generate multiple sustained release metformin matrix tablet compositions that were subsequently validated through experimental testing. Among the 5 AI generated formulations, F3 and F4 closely reproduced the dissolution behavior of the reference product, as confirmed by similarity factor analysis, release kinetics modelling, and comprehensive statistical evaluation. The experimental results demonstrated that ChatGPT (GPT-4, OpenAI) has the capacity to suggest polymer combinations and ratios that are translated into physically manufacturable tablets with predictable release performance. The strong agreement between AI predicted and experimental dissolution profiles for the suggested formulations, F3 and F4, further supports the efficacy of AI assisted formulation design when combined with pharmaceutical expertise and laboratory validation. Importantly, the study also revealed limitations, as certain AI generated formulations deviated from the reference product dissolution profile due to retardation or accelerated late-stage re-lease, emphasizing the need for experimental confirmation. Overall, this work demonstrates that large language models are not replacements for formulation scientists but can serve as effective decision support tools to reduce trial and error experimentation, support excipients selection, and accelerate early formulation screening when combined with pharmaceutical expertise and experimental validation. The integration of AI driven predictions with experimental validation represents a practical approach that may contribute to more efficient development of extended-release oral dosage forms across a wide range of drug substances. Declarations Conflicts of Interest : The authors declare no conflict of interest. Funding: This research was funded by Alfaisal University internal research grant (IRG 23522). Author Contribution Conceptualization: M.A., K.S, G.M.M; Methodology: M.A., K.S., S.A., G.M.M; AI model design and formulation simulation: K.S., S.A.; Experimental supervision and tablet manufacturing: G.M.M., A.F.A.; Investigation and laboratory work: A.B.A.; Kinetic modeling, similarity factor analysis, and statistical evaluation: A.A.I., M.A; Formal analysis: M.A., A.A.I., K.S; Writing original draft preparation: M.A., K.S; Writing review and editing: M.A., A.A.I.; Project administration and leadership: M.A., K.S, G.M.M; Funding acquisition: M.A. Acknowledgement The authors gratefully acknowledge the Research Office at Alfaisal University for financial support through the Internal Research Grant that enabled the completion of this work. The authors also acknowledge the technical and logistical support provided during the experimental phase of the study. Data Availability The data generated and analyzed during this study are included in this published article. Additional datasets generated during the current study are available from the corresponding author upon reasonable request. References Kant S, Deepika. Roy · Saheli. Artificial intelligence in drug discovery and development: transforming challenges into opportunities. Discover Pharm Sci. 2025;1(1):7. 10.1007/S44395-025-00007-3 . 2025 1:1. Blanco-González A, Cabezón A, Seco-González A, et al. The Role of AI in Drug Discovery: Challenges, Opportunities, and Strategies. Pharmaceuticals. 2023;16(6). 10.3390/PH16060891/S1 . Almanasra S, Suwais K. Analysis of ChatGPT-Generated Codes Across Multiple Programming Languages. IEEE Access. 2025;13:23580–96. 10.1109/ACCESS.2025.3538050 . De Angelis L, Baglivo F, Arzilli G, et al. ChatGPT and the rise of large language models: the new AI-driven infodemic threat in public health. Front Public Health. 2023;11:1166120. 10.3389/FPUBH.2023.1166120/BIBTEX . Sallam M, Sallam M. ChatGPT Utility in Healthcare Education, Research, and Practice: Systematic Review on the Promising Perspectives and Valid Concerns. Healthcare 2023, Vol 11,. 2023;11(6). 10.3390/HEALTHCARE11060887 Wu J, Ma Y, Wang J, Xiao M. The Application of ChatGPT in Medicine: A Scoping Review and Bibliometric Analysis. J Multidiscip Healthc. 2024;17:1681–92. 10.2147/JMDH.S463128 . Mu Y, He D. The Potential Applications and Challenges of ChatGPT in the Medical Field. Int J Gen Med. 2024;17:817–26. 10.2147/IJGM.S456659 . Maity S, Saikia MJ, Maity S, Saikia MJ. Large Language Models in Healthcare and Medical Applications: A Review. Bioengineering 2025, Vol 12,. 2025;12(6). 10.3390/BIOENGINEERING12060631 Zhao A, Wu Y. Future implications of ChatGPT in pharmaceutical industry: drug discovery and development. Front Pharmacol. 2023;14:1194216. 10.3389/FPHAR.2023.1194216 . Ravi Kiran AVVV, Kumari GK, Krishnamurthy PT. ChatGPT in Drug Discovery Process. Adv Pharm Bull. 2023;14(1):5. 10.34172/APB.2024.007 . European Federation of Pharmaceutical Industries and Associations (EFPIA). The Pharmaceutical Industry in Figures: Key Data 2023. Brussels, Belgium: EFPIA. 2023. Accessed December 13, 2025. Timmons J, Boyle J. Metformin. Diabetes Drug Notes. Published online August. 2023;17:30–48. 10.1002/9781119785033.ch2 . Cheng M, Ren L, Jia X, Wang J, Cong B. Understanding the action mechanisms of metformin in the gastrointestinal tract. Front Pharmacol. 2024;15:1347047. 10.3389/FPHAR.2024.1347047 . Cirilli M, Moutaharrik S, Palugan L, et al. Gastroprotective systems for prolonged release of metformin based on osmotically driven expansion. Int J Pharm. 2025;681. 10.1016/j.ijpharm.2025.125856 . The United States Pharmacopeial Convention. USP 1216 tablet friability. United States Pharmacopeia. 32:22–4. U.S, Pharmacopoeial, Convention. (905) Uniformity of Dosage Units. Stage 6 Harmonization. Vol 3. United States Pharmacopeia; 2011. U.S. of Pharmacopoeia. The United States Pharmacopeia Disintegration. United States Pharmacopeial Convention. 2020;4:2–5. Pharmacopeial Convention US. Pharmacopeial Guidelines Dissolution, Dissolution. United States Pharmacopeia. 2011;1:1–8. Dange YD, Honmane SM, Bhinge SD, Salunkhe VR, Jadge DR. Development and validation of uv-spectrophotometric method for estimation of metformin in bulk and tablet dosage form. Indian J Pharm Educ Res. 2017;51(4):S754–60. 10.5530/IJPER.51.4S.109 . Hamzah A, Maswadeh, Othman A, Al-Hanbali RA, Kanaan, Ashok K, Shakya A, Maraqa. Testing lyoequivalency for three commercially sustained-release tablets containing diltiazem hydrochloride. Acta Pol Pharm. 2010;67(1):93–97. Accessed December 13, 2025. https://pubmed.ncbi.nlm.nih.gov/20210085/ Abdulkarim M, Adam DR, Zahrani Y, Al, et al. Development of a metformin matrix tablet: a comparative study with marketed sustained release formulation. Farmacia. 2025;73:4. 10.31925/farmacia.2025.4.11 . Patel S, Kaushal AM, Bansal AK. Compression Physics in the Formulation Development of Tablets. Crit Rev Ther Drug Carrier Syst. 2006;23(1):1–65. 10.1615/CRITREVTHERDRUGCARRIERSYST.V23 . Nokhodchi A, Ford JL, Rowe PH, Rubinstein MH. The effects of compression rate and force on the compaction properties of different viscosity grades of hydroxy propyl methyl cellulose 2208. Int J Pharm. 1996;129(1–2):21–31. 10.1016/0378-5173(95)04236-9 . Obeidat WM, Nokhodchi A, Alkhatib H. Evaluation of Matrix Tablets Based on Eudragit®E100/Carbopol®971P Combinations for Controlled Release and Improved Compaction Properties of Water-Soluble Model Drug Paracetamol. AAPS PharmSciTech 2015 16:5. 2015;16(5):1169–1179. 10.1208/S12249-015-0301-5 Pramanik A, Sahoo RN, Nandi S, Nanda A, Mallick S. Characterization of Hydration Behaviour and Modeling of Film Formulation. Acta Chim Slov. 2021;68:159–69. 10.17344/acsi.2020.6298 . Nokhodchi A, Raja S, Patel P, Asare-Addo K. The role of oral controlled release matrix tablets in drug delivery systems. Bioimpacts. 2012;2(4):175–87. 10.5681/bi.2012.027 . Hardy IJ, Windberg-Baarup A, Neri C, Byway PV, Booth SW, Fitzpatrick S. Modulation of drug release kinetics from hydroxypropyl methyl cellulose matrix tablets using polyvinyl pyrrolidone. Int J Pharm. 2007;337(1–2):246–53. 10.1016/J.IJPHARM.2007.01.026 . Askarizadeh M, Esfandiari N, Honarvar B, Sajadian SA, Azdarpour A. Kinetic Modeling to Explain the Release of Medicine from Drug Delivery Systems. Chem Bio Eng Reviews. 2023;10(6):1006–49. .202300027; W group: String: Publication. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8755354","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":591390132,"identity":"ffcbed86-163a-4b10-a8ea-b1bf5e330ab3","order_by":0,"name":"Muthanna Abdulkarim","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAElEQVRIie3OMUvEMBTA8RcCndK94kE/gfAO4RCq3ld5oWBXp0NwMFO6HHTVb3Eg3GpLwFuOc+3YcKtD3ToomK7iRd0c8l+S4f14DyAU+pdxDXDN3ctr6M5BuJ8CiHyEOYIjiQjoaiTstwQEjgR+JCdlo7sBsxQ34q0nupykZaOgX5iDZLaV5XSJxXRl4seEKBe4lYrd7zykljoRaNjaxKtEDtydJxWPtYe8WH30jma+NmI/EN2JtLKKf/hIK/Wx2yIdidxhRkDrtjAvseXpBIv8yUSzM6KNwNaqZrkrPIflz/b1Jrt4qMy+7el2nlZ50w2L7CD5vvqP86FQKBT60icybVvqrFtaQgAAAABJRU5ErkJggg==","orcid":"","institution":"Alfaisal University","correspondingAuthor":true,"prefix":"","firstName":"Muthanna","middleName":"","lastName":"Abdulkarim","suffix":""},{"id":591390133,"identity":"f63ab49b-cd2a-45e3-bb48-266dce753082","order_by":1,"name":"Abdullah B Alabdulwahab","email":"","orcid":"","institution":"King Saud University","correspondingAuthor":false,"prefix":"","firstName":"Abdullah","middleName":"B","lastName":"Alabdulwahab","suffix":""},{"id":591390134,"identity":"1d796d63-9048-4525-a6b7-60176171f005","order_by":2,"name":"Arwa Alhaj Issa","email":"","orcid":"","institution":"Alfaisal University","correspondingAuthor":false,"prefix":"","firstName":"Arwa","middleName":"Alhaj","lastName":"Issa","suffix":""},{"id":591390135,"identity":"cf31335f-cb25-4bd2-9c54-c3628a427946","order_by":3,"name":"Gamal M Mahrous","email":"","orcid":"","institution":"King Saud University","correspondingAuthor":false,"prefix":"","firstName":"Gamal","middleName":"M","lastName":"Mahrous","suffix":""},{"id":591390136,"identity":"33846ff9-7fd7-4772-b006-f00eed0ba683","order_by":4,"name":"Adel F Alghaith","email":"","orcid":"","institution":"King Saud University","correspondingAuthor":false,"prefix":"","firstName":"Adel","middleName":"F","lastName":"Alghaith","suffix":""},{"id":591390137,"identity":"1dce1207-a5b4-465a-9493-448520757587","order_by":5,"name":"Sally Almanasra","email":"","orcid":"","institution":"Prince Sultan University","correspondingAuthor":false,"prefix":"","firstName":"Sally","middleName":"","lastName":"Almanasra","suffix":""},{"id":591390138,"identity":"7fba5af7-31d3-4119-b95b-9df8ec67b56d","order_by":6,"name":"Khaled Suwais","email":"","orcid":"","institution":"Arab Open University","correspondingAuthor":false,"prefix":"","firstName":"Khaled","middleName":"","lastName":"Suwais","suffix":""}],"badges":[],"createdAt":"2026-02-01 10:10:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8755354/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8755354/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102963078,"identity":"91cd3eca-db00-43db-ae5b-b448cecb38ba","added_by":"auto","created_at":"2026-02-19 04:13:19","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":686484,"visible":true,"origin":"","legend":"\u003cp\u003ePrompt sample used to generate the five Metformin formulas.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8755354/v1/ec664967cde7e0c70465c651.jpeg"},{"id":102962954,"identity":"700723ba-9502-4a0f-8005-b44667df3bec","added_by":"auto","created_at":"2026-02-19 04:12:20","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":263962,"visible":true,"origin":"","legend":"\u003cp\u003eChatGPT (GPT-4, OpenAI) response to formulation prompt for Metformin XR 500 mg.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8755354/v1/a3fed5172c02d3653e6cf5a9.png"},{"id":102963070,"identity":"5e1c43fc-b751-4dc6-8d05-09c0481d629e","added_by":"auto","created_at":"2026-02-19 04:13:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":61200,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of various Effect of polymer composition on the release profile of metformin HCl from tablets’ matrices.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8755354/v1/83787fd31f6c2a1dfe85b692.png"},{"id":102795402,"identity":"02b332fd-6609-486d-a82d-4714e44d0f84","added_by":"auto","created_at":"2026-02-16 18:55:28","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":89940,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation of similarity factor f₂ with (a) release rate constant (kₖ), (b) release exponent (n).\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8755354/v1/ae458d6b1e9c36f751584d16.png"},{"id":102795405,"identity":"b1c209c8-6c5b-4f53-ae31-260ffb70e26d","added_by":"auto","created_at":"2026-02-16 18:55:28","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":54631,"visible":true,"origin":"","legend":"\u003cp\u003eMAPE and RMSE values comparing AI predictions to experimental results.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8755354/v1/1ba330778798b3cb673c9cec.png"},{"id":102795406,"identity":"90a29ae6-41cc-421b-9ef4-3e3b29b93d90","added_by":"auto","created_at":"2026-02-16 18:55:28","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":49371,"visible":true,"origin":"","legend":"\u003cp\u003eDendrogram of formulation similarity based on dissolution kinetics.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8755354/v1/580ac79482b3b469b88aa134.png"},{"id":104834858,"identity":"127dc9e0-5863-4cf5-979d-7ed60de2ba4e","added_by":"auto","created_at":"2026-03-17 17:33:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2466630,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8755354/v1/d244b1d9-6af0-43bd-a4f1-f55685e5a602.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI-Driven Simulation and Design of Sustained Release Metformin Tablets: Experimental Validation and Predictive Accuracy Assessment","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe pharmaceutical industry is undergoing a progressive shift with the increasing integration of artificial intelligence (AI) tools aimed at accelerating drug development and formulation processes. Traditionally, drug discovery and development have been lengthy and costly endeavors, often taking over a decade and billions of dollars to bring a single drug to market [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. However, the advent of AI technologies has begun to streamline these processes by enhancing data analysis, improving predictive modeling, and facilitating efficient decision-making [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. GPT-4 is a large language model (LLM) generative pre-trained transformer (GPT) developed by OpenAI [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. It utilizes deep learning techniques to generate coherent and contextually relevant natural language responses in a conversational format. LLMs are powerful general-purpose reasoning and communication systems that can understand, generate, and reason with natural language [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. This makes LLMs useful across education, research, business, programming, and everyday problem-solving. In this context, GPT-4 is capable of understanding and generating natural language, reasoning and problem solving, retrieving and synthesis knowledge, task automation, and many others [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eChatGPT is built on a large language model that has been trained using deep learning techniques on vast amounts of text data. From an IT perspective, it is designed to analyze patterns in language, break down user input into tokens, and predict the most likely sequence of responses based on context [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. ChatGPT uses transformer architecture, which allows it to handle long conversations, understand relationships between words, and generate coherent replies. In addition, the model dynamically generates responses in real time, making it adaptable across different domains and tasks [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Among other LLMs, ChatGPT (GPT-4, OpenAI) stands out because of its balance of accuracy, adaptability, and accessibility [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. It integrates well with different platforms and provides reliable outputs with a strong emphasis on safety and alignment.\u003c/p\u003e \u003cp\u003eChatGPT (GPT-4, OpenAI) can be used in pharmaceutical research studies because of its versatility, accessibility, and balance between technical depth and usability [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Unlike many other LLMs, ChatGPT (GPT-4, OpenAI) provides a flexible platform that can handle a wide range of needs in pharmaceutical studies [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. For researchers, it can generate comparative analyses of formulations and suggest experimental designs that align with regulatory standards. It can also simulate patient counseling, draft clear explanations of drug mechanisms, and even support the early stages of formulation development by analyzing ingredient roles and dosage constraints.\u003c/p\u003e \u003cp\u003eWhile AI tools have revolutionized late-stage drug discovery which enabling target identification, lead optimization, and predictive pharmacokinetics, their application in formulation development, and in particular for sustained-release metformin tablets, remains limited and underexplored. Despite promising advances in machine learning and deep neural networks for general drug formulation optimization, including in vitro performance prediction, to date, no prior study has prospectively validated AI-generated sustained-release formulations through experimental bench testing, highlighting a gap between \u003cem\u003ein silico\u003c/em\u003e predictions and practical laboratory confirmation. This is also critical since the pharmaceutical industry spend hundreds of billions of dollars every year on research and development related to drug formulation [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eHydrophilic matrix tablets represent one of the most widely used platforms for achieving extended drug release as they can be manufactured using conventional tableting processes. Drug release from these systems is managed by many factors such as polymer hydration, gel layer formation, diffusion through the swollen matrix, and gradual erosion. For highly water-soluble drugs such as metformin hydrochloride which administered at high dose, careful selection of polymer type, viscosity grade, and relative proportion to other excipients is critical, as small variations in matrix composition can lead to substantial differences in dissolution behavior.\u003c/p\u003e \u003cp\u003eThe primary aim of this study is to evaluate ChatGPT's (GPT-4, OpenAI) ability to predict viable sustained-release Metformin formulations and compare these predictions to real experimental outcomes using error metrics and validation tools. Metformin, a derivative of biguanide, is the first-line antidiabetic medication approved by the FDA for the management of elevated blood sugar levels in patients with type 2 diabetes [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. It is available in both immediate-release (IR) and long-acting (extended-release; XR) formulations. Metformin effectively reduces blood glucose levels by decreasing glucose production in the liver, reducing intestinal absorption, and improving insulin sensitivity. Consequently, metformin lowers both basal and postprandial blood glucose levels [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. The design of extended-release metformin tablets presents a variety of challenges. The high dose of metformin is one of the primary concerns, as the prolonged release requires administering a high dose to ensure the drug works effectively. Additionally, being highly water-soluble with poor lipid solubility, metformin demands precise control of release rates to maintain therapeutic plasma concentrations without compromising absorption. These factors must be addressed to optimize the design and performance of extended-release metformin tablets.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Materials\u003c/h2\u003e \u003cp\u003eMetformin HCl (DC grade) was kindly provided by Hikma Pharmaceuticals. Various grades of hydroxypropyl methylcellulose (HPMC) under brand name Benecel\u0026trade;, including HPMC K100M DC, HPMC K4M DC, HPMC F50, HPMC E4M, HPMC A15LV, and HPMC E4M CR, were obtained as gifts from Ashland (USA) through Brenntag Saudi Arabia Ltd. Carbopol\u0026reg; 71G NF (Lubrizol, USA) and colloidal silicon dioxide (Evonik, Germany) were also donated by Brenntag. Additional excipients, including microcrystalline cellulose (Avicel PH102) and povidone K-29/32, were gifts from Sudair Pharma. Dicalcium phosphate (DCP) was gifted by King Saud University. Croscarmellose sodium, Crosspovidone XL-10, starch, talc, and magnesium stearate were procured from local suppliers. Analytical grade buffer salts, including disodium hydrogen phosphate, and potassium dihydrogen phosphate, were sourced from Merck (Germany), and lastly Formit XR 500 mg from SPIMACO.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. AI-Based Formulation Design\u003c/h2\u003e \u003cp\u003eChatGPT (GPT-4, OpenAI) was employed to simulate preliminary extended-release tablet formulations. The model was provided with predefined constraints, including the API dose (500 mg metformin HCl), available excipients (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), and the desired release profile mimicking the commercial product Formit XR (500 mg, SPIMACO), and was tasked with generating multiple sustained-release matrix tablet compositions. Thus, ChatGPT (GPT-4, OpenAI) was instructed to act as a pharmaceutical formulation scientist and design five sustained-release tablet formulas for Metformin HCl 500 mg using direct compression, employing different combinations of hydrophilic polymers, binders, disintegrants, and lubricants with the aim of mimicking the release profile of the reference product (Formit XR 500 mg). The full prompt is provided below for reproducibility (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e)\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eList of suggested excipients and their roles for ChatGPT (GPT-4, OpenAI) to choose from.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaterials\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRole\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetformin\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eActive ingredient\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHPMC K4M DC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003eThese polymers are members of the hydroxypropyl methylcellulose family with different molecular weights and viscosities. The high molecular weight means high viscosity and longer releases of the drug.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHPMC K100M DC\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHPMC F50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHPMC E4M\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHPMC A15LV\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHPMC E4M CR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCarbopol 71G\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePolyacrylic acid polymer matrix\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCroscarmellose Sodium\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDisintegrator to facilitate the release of drug\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCrosspovidone XL 10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMicrocrystalline cellulose\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eBinders to hold the tablet together\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePlasdone k-29/32\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDicalcium phosphate\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStarch\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMagnesium Stearate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eGlidant and lubricant\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTalc\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ecolloidal silicon di oxide\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Experimental Design and Tablet Preparation\u003c/h2\u003e \u003cp\u003eAll tablet formulations were prepared via direct compression. Active drug and excipients (as per Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) were accurately weighed and blended using a Turbula mixer for 10 minutes. Magnesium stearate and colloidal silicon dioxide were added last and mixed for an additional 3 minutes. Compression was carried out using a single-punch tablet press (12 mm punches), targeting a final tablet weight between 695\u0026ndash;700 mg. Each formulation was subjected to physical characterization and in vitro release testing.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Tablet Evaluation\u003c/h2\u003e \u003cp\u003eThe generic and the formulated tablets were tested for their friability, weight variation, disintegration and dissolution according to USP 38 NF 35 chapters\u0026thinsp;\u0026lt;\u0026thinsp;1216\u0026gt; [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], \u0026lt;\u0026thinsp;905\u0026gt; [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e], \u0026lt;\u0026thinsp;701\u0026gt; [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] and \u0026lt;\u0026thinsp;711\u0026gt; [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] respectively.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e2.4.1. Weight variation\u003c/h2\u003e \u003cp\u003eTwenty tablets from each formulation and the reference product were randomly chosen and weighed individually using a Kern analytical balance (ABJ120-4NM, Germany). The mean weight of tablets is calculated and the weights of individual tablets are compared with the mean. If\u0026thinsp;\u0026le;\u0026thinsp;2 tablets vary by more than \u0026plusmn;\u0026thinsp;2 times the range from the average weight, and no tablet varies by more than \u0026plusmn;\u0026thinsp;twice the range, then the range is acceptable (United States Pharmacopeia, 2011b).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003e2.4.2. Friability\u003c/h2\u003e \u003cp\u003eTen tablets from each formulation and the reference were randomly selected, weighed, and placed into the friabilator (Erweka Friability tester, TA3R, Germany) operated at 25 rpm for 4 minutes. After 100 rounds, Percent weight loss was calculated post-test. The formulation passed if friability did not exceed 1.0%. (United States Pharmacopeia, 2016).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e2.4.3. Hardness test\u003c/h2\u003e \u003cp\u003eTablet breaking force (hardness) was measured in kiloponds (kp) using an Erweka TBH 225 hardness tester. Ten tablets per batch were assessed, and average values were recorded.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e2.4.4. Thickness Test\u003c/h2\u003e \u003cp\u003eTablet thickness uniformity was checked using a Vernier caliper. Ten tablets per batch were measured to ensure consistency within acceptable batch variation limits.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.5. In Vitro Dissolution Study\u003c/h2\u003e \u003cp\u003eDissolution testing was conducted using a USP Type II (paddle) apparatus (UDT-804, LOGAN Instruments, USA); in phosphate buffer (pH 6.8, simulated intestinal fluid). Each formulation was tested in six vessels maintained at 37\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5\u0026deg;C and stirred at 50 rpm[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. At time points 30, 60, 120, 180, 240, 360, and 480 minutes, 5 mL samples were withdrawn and replaced with fresh medium. Samples were filtered through 0.45 \u0026micro;m Millipore filters, diluted (1:5), and analyzed for absorbance at 233 nm[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] using a UV-Vis spectrophotometer (UV-1700 PC, Shimadzu, Japan). The calibration curve showed linearity with R\u0026sup2; \u0026gt; 0.999. (United States Pharmacopeia, 2011a).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e2.6. Ethics and Compliance\u003c/h2\u003e \u003cp\u003eNo human or animal subjects were involved in this study.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e2.7. Statistical analysis\u003c/h2\u003e \u003cp\u003eStatistical evaluation of dissolution data was performed to assess formulation similarity in term of their release behaviors, model the release behaviors, and compare AI predicted versus experimentally observed dissolution release profiles. All statistical analyses were conducted using Python code, utilizing libraries like Pandas for data manipulation and Matplotlib for visualization.\u003c/p\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e2.7.1. Similarity and Difference Factors\u003c/h2\u003e \u003cp\u003eThe similarity factor (f₂) and difference factor (f₁) were calculated using the standard mathematical equations described for comparing dissolution profiles. Dissolution profiles were regarded as similar when f₂ \u0026ge; 50 and f₁ \u0026le; 15 [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e2.7.2. Release Kinetics Modeling\u003c/h2\u003e \u003cp\u003eDissolution data for the reference and all formulations (F1\u0026ndash;F5) were fitted to four kinetic models as was described in our earlier publication [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]: Zero-order, First-order, Higuchi diffusion, and Korsmeyer\u0026ndash;Peppas (K\u0026ndash;P) mode. Model parameters (rate constants and R\u0026sup2; values) were obtained using linear regression. For the K\u0026ndash;P model, the release exponent (n) was used to determine the dominant drug-release mechanism (Fickian, non-Fickian, or erosion-controlled).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e2.7.3. Correlation Between Kinetic Parameters and Similarity Factor\u003c/h2\u003e \u003cp\u003eTo evaluate whether formulations with similar kinetics also produced dissolution profiles closer to the reference, Pearson correlation coefficients were calculated between f₂ and the K\u0026ndash;P parameters (release rate constant kₖ, exponent n, and model fit R\u0026sup2;). Statistical significance was set at p\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e \u003ch2\u003e2.7.4. Comparison of AI-Predicted and Experimental Dissolution Profiles\u003c/h2\u003e \u003cp\u003eTo assess ChatGPT\u0026rsquo;s (GPT-4, OpenAI) predictive performance, the AI-generated dissolution curves were compared with laboratory data using multiple quantitative metrics: Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Pearson correlation coefficient (r), Concordance correlation coefficient (CCC), Mean percentage bias. Error analyses were performed across all dissolution time points for each formulation (F1\u0026ndash;F5).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e2.7.5. Bland\u0026ndash;Altman and Equivalence Testing\u003c/h2\u003e \u003cp\u003eAgreement between AI-predicted and experimental dissolution values was further evaluated using Bland\u0026ndash;Altman analysis and \u0026plusmn;\u0026thinsp;10% equivalence testing. Formulations were considered statistically equivalent if mean bias was not significantly different from zero and 95% limits of agreement fell within \u0026plusmn;\u0026thinsp;10%.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e \u003ch2\u003e2.7.6. Hierarchical Cluster Analysis (HCA)\u003c/h2\u003e \u003cp\u003eAgglomerative hierarchical cluster analysis (Ward\u0026rsquo;s method) was applied to classify the formulations based on their dissolution kinetics. Standardized values of f₂, kₖ, and n were used to produce a dendrogram identifying similarity clusters among the reference and the test formulations.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"3. Results and Discussion","content":"\u003cp\u003eThe current work evaluates the experimental performance of AI generated metformin extended release matrix formulations. The formulation design strategy and excipient selection are first discussed, followed by an assessment of the physical properties of the prepared tablets. The in vitro dissolution behavior is then examined and compared with the reference product, supported by release kinetics, similarity factor analysis and statistical evaluation.\u003c/p\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Formulation Development\u003c/h2\u003e \u003cp\u003eTables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e summarize the compositions of the different matrix tablet formulations generated by ChatGPT(GPT-4, OpenAI). The AI model proposed varying ratios of hydrophilic polymers, binders, and fillers to modulate the release behavior of Metformin HCl. In all formulations, the total tablet weight was maintained at a range of 695\u0026ndash;700 mg, containing 71% w/w Metformin HCl (500 mg per tablet). The five ChatGPT-generated formulations (F1\u0026ndash;F5) were designed using varying proportions and combinations of hydrophilic matrix-forming polymers, binders, and fillers to explore their influence on drug release modulation and tablet properties.\u003c/p\u003e \u003cp\u003eFormulation F1 employed HPMC K4M DC (17.1% w/w) and HPMC E4M (7.1%) as dual matrix-forming polymers, with microcrystalline cellulose (2.1%) serving as a compressibility enhancer. The combination aimed to form a moderate-viscosity gel upon hydration and maintain mechanical integrity. In F2, the primary matrix polymer was HPMC K100M DC at a relatively high concentration (14.3%), complemented by Carbopol 71G (10%) to enhance gel strength and prolong drug release. Dicalcium phosphate (2.1%) functioned as a filler, while magnesium stearate and talc supported tablet manufacturability. F3 incorporated a blend of HPMC K4M DC (11.5%) and HPMC K100M DC (10.1%) to balance gel formation and release modulation. A notable inclusion of Plasdone K-29/32 (4.3%) acted as both a binder and solubility enhancer. This formulation lacked traditional fillers but maintained consistent weight through lubricant and glidant inclusion.\u003c/p\u003e \u003cp\u003eFormulation F4 focused on maximizing gel matrix formation, using HPMC K100M DC (10%), HPMC E4M (12.9%), and a modest amount of starch (1.4%) as a disintegrant-filler hybrid. Microcrystalline cellulose (2.1%) helped improve compressibility and flow. Lastly, F5 introduced Carbopol 71G at the highest concentration (14.4%) among all formulations, with HPMC K4M DC (7.2%) forming a supporting gel layer. Starch (4.3%) played a dual role in swelling and filling.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComposition of AI generated metformin hydrochloride extended-release matrix tablets (tablet weight indicated in column headers).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eComponent\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eF1 (700 mg)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eF2 (700 mg)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eF3 (695 mg)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF4 (700 mg)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eF5 (695 mg)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMetformin\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e500.0 mg\u003c/p\u003e \u003cp\u003e(71.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e500.0 mg\u003c/p\u003e \u003cp\u003e(71.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e500.0 mg\u003c/p\u003e \u003cp\u003e(71.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e500.0 mg\u003c/p\u003e \u003cp\u003e(71.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e500.0 mg\u003c/p\u003e \u003cp\u003e(71.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHPMC K4M DC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e120.0 mg\u003c/p\u003e \u003cp\u003e(17.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e80.0 mg\u003c/p\u003e \u003cp\u003e(11.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e50.0 mg\u003c/p\u003e \u003cp\u003e(7.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHPMC K100M DC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e100.0 mg\u003c/p\u003e \u003cp\u003e(14.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e70.0 mg\u003c/p\u003e \u003cp\u003e(10.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e70.0 mg\u003c/p\u003e \u003cp\u003e(10.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHPMC E4M\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e50.0 mg\u003c/p\u003e \u003cp\u003e(7.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e90.0 mg\u003c/p\u003e \u003cp\u003e(12.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCarbopol 71G\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e70.0 mg\u003c/p\u003e \u003cp\u003e(10.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e100.0 mg\u003c/p\u003e \u003cp\u003e(14.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMicrocrystalline cellulose\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15.0 mg\u003c/p\u003e \u003cp\u003e(2.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e15.0 mg\u003c/p\u003e \u003cp\u003e(2.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e-\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePlasdone K-29/32\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e30.0 mg\u003c/p\u003e \u003cp\u003e(4.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDicalcium phosphate\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15.0 mg\u003c/p\u003e \u003cp\u003e(2.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eStarch\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10.0 mg\u003c/p\u003e \u003cp\u003e(1.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e30.0 mg\u003c/p\u003e \u003cp\u003e(4.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMagnesium Stearate\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTalc\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7.5 mg\u003c/p\u003e \u003cp\u003e(1.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAll formulations included magnesium stearate and talc at constant low levels (approximately 1.1% each) to ensure proper lubrication and flow during tablet compression. These formulations were developed to systematically investigate how variations in matrix polymer type, binder content, and filler selection could influence the physicochemical behavior of Metformin HCl sustained-release tablets relative to a reference product.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Physical Evaluation of Tablets\u003c/h2\u003e \u003cp\u003eThe physical properties of the AI-generated metformin HCl extended-release matrix tablets were evaluated to assess their suitability for further dissolution and stability testing. The key physical parameters measured included weight variation, tablet thickness, friability, and hardness. These properties are critical indicators of manufacturing reproducibility, mechanical integrity, and overall tablet quality. The results are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePhysical properties of AI generated metformin hydrochloride extended-release tablets.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFormulation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWeight (mg)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness (mm)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHardness (Kp)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFriability (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e31.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.49\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e17.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e13.04\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e77.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eConsistent tablet weight ensures dose uniformity and reflects manufacturing precision. All five formulations (F1\u0026ndash;F5) passed the USP weight variation test, as defined in USP\u0026thinsp;\u0026lt;\u0026thinsp;905\u0026gt;, which requires individual tablet weights to remain within \u0026plusmn;\u0026thinsp;5% of the average. This indicates uniform die filling, adequate powder flow, and reproducible tablet compression across batches [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTablet thickness was constant at 5.5 mm across all formulations, suggesting consistent compaction and tooling settings during manufacture. Although thickness does not directly impact dissolution for non-porous tablets, it can affect tablet handling, blister compatibility, and packaging. The fixed thickness across batches also suggests that the excipient blend and polymer ratios allowed uniform tableting properties despite formulation variability.\u003c/p\u003e \u003cp\u003eTablet hardness is a critical parameter that reflects mechanical strength and resistance to breakage. The formulations showed significant variation in hardness values, ranging from 17.6 Kp (F4) to 90.8 Kp (F2). F2 and F5 exhibited the highest hardness (90.8 Kp and 77.0 Kp, respectively), correlating with better compaction and polymer-filler interaction. In contrast, F4 displayed the lowest hardness (17.6 Kp), which may be due to its high polymeric content (HPMC E4M and K100M) and lack of traditional binders, potentially resulting in reduced cohesion during compression [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. However, as previously demonstrated [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], high hardness does not necessarily correlate with improved friability because excessive compression may induce internal stresses and brittleness within the tablet structure\u003c/p\u003e \u003cp\u003eFriability testing evaluates tablet robustness during handling and transport. USP standards require friability to be \u0026le;\u0026thinsp;1% for conventional tablets. In this study, F2 exhibited the lowest friability value (1.14%), slightly exceeding the strict USP limit, but indicating comparatively better mechanical robustness than the other formulations. The relative low friability of F2 compared to other formulae could be attributed to the synergistic effect of HPMC K100 M, Carbopol and DCP which altogether form strong interparticle bonds that support tablet integrity [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. F1 (2.14%), F3 (3.49%), and F5 (3.19%) showed moderate friability, while F4 exhibited the highest friability at 13.04%, indicating poor mechanical resistance [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. These differences highlight the critical impact of binder and polymer ratios on tablet integrity.\u003c/p\u003e \u003cp\u003eIn summary, the physical evaluation demonstrated that while all formulations (F1\u0026ndash;F5) exhibited acceptable weight uniformity and consistent thickness, only F2 and F5 met the desired mechanical criteria by combining high hardness with relatively low friability. In contrast, F4, due to its low hardness and markedly high friability, may require reformulation or binder optimization to enhance its structural integrity.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e3.3. In Vitro dissolution testing\u003c/h2\u003e \u003cp\u003eDissolution testing was carried out using USP Apparatus II (paddle method) at 37\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5\u0026deg;C in phosphate buffer (pH 6.8) to evaluate the release behavior of the ChatGPT (GPT-4, OpenAI) generated Metformin HCl extended-release tablets (F1\u0026ndash;F5) versus the reference product. Samples were taken at predefined time points and analyzed to obtain the percentage of drug released. The dissolution profiles for all formulations, together with that of the reference product Formit\u0026reg; XR, are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs can be observed in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, Notable differences in release kinetics were observed among the tested formulations, primarily driven by variations in the polymer systems and excipient ratios employed. Formulation F1, containing HPMC K4M DC and HPMC E4M as matrix-forming agents, exhibited a moderately sustained release with 88.5% drug release at 8 hours, approaching the reference product in the later stages. However, its profile was slower than the reference at early stages (e.g., 34.1% at 2 hours vs. 48.9% for Formit XR). Such delayed release during the first three hours, may be attributed to the combined effect of K4M and E4M (lower-viscosity HPMC grades) that swell quickly with rapid gel formation creating an early cohesive barrier that temporarily slows metformin diffusion [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. As hydration progressed, the matrix relaxed and erosion increased, resulting in a slightly faster release rate between three and six hours.\u003c/p\u003e \u003cp\u003eF2, composed of HPMC K100M DC and Carbopol 71G, demonstrated the slowest release profile throughout the entire testing period, with only 79.4% released by 8 hours. Its deviation from the reference was most evident at later time point, indicating a stronger matrix barrier that may prolong drug retention beyond the intended therapeutic window. The markedly delayed release is consistent with the strong retarding properties of both polymers: K100M hydrates more slowly and forms a dense, high-viscosity gel, while Carbopol rapidly swells into a thick, highly crosslinked matrix, further restricting water penetration and drug diffusion [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFormulation F3, which incorporated a blend of HPMC K4M DC, HPMC K100M DC, and Plasdone K-29/32, exhibited a release profile that closely matched the reference product, particularly between 2 and 6 hours (e.g., 71.5% at 4 hours vs. 71.6% for Formit XR). This suggests a properly balanced matrix capable of maintaining sustained and predictable drug diffusion. As compared to formulations F1 and F2, the complementary hydration behavior of K4M and the strong gel strength of K100M, together with the pore-forming effect of Plasdone, enabled F3 to maintain a sustained and predictable diffusion pattern consistent with the reference [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Similarly, F4, containing high levels of HPMC E4M and HPMC K100M DC, produced the most similar profile to Formit XR across all time points. It achieved 90.2% release at 8 hours and nearly identical intermediate values (e.g., 64.4% at 3 hours vs. 61.4% for the reference), indicating that early hydration from E4M combined effectively with the diffusion-controlling properties of K100M to modulate release throughout the dissolution period.F5, formulated with Carbopol 71G and HPMC K4M DC, demonstrated a slightly faster release, reaching 94.2% at 8 hours, the highest among all formulations. While its early release was modestly higher than the reference (e.g., 26.4% vs. 20.7% at 30 minutes), the overall profile remained within an acceptable range of similarity.\u003c/p\u003e \u003cp\u003eAmong all formulations, F3 and F4 achieved dissolution profiles most comparable to Formit\u0026reg; XR, particularly in the mid-to-late phase of release, suggesting effective matrix integrity and controlled drug diffusion. F5 showed slightly faster release at later stages, while F2 exhibited the most delayed release kinetics. These observations highlight the critical influence of polymer selection and ratio on the extended-release behavior of Metformin HCl matrix tablets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Similarity and Difference Factors\u003c/h2\u003e \u003cp\u003eCalculating the difference factor (f1) and similarity factor (f2) between each formulation and the innovator product (Formit XR) provided insight into the comparative release behavior. As per regulatory guidelines, an f1 value below 15 and an f2 value above 50 indicate that two dissolution profiles are considered similar.\u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, F3 exhibited the highest level of similarity to the reference product, with an f1 value of 2.65 and an f2 value of 81.67, clearly indicating strong similarity. Similarly, F4 also demonstrated strong similarity (f1\u0026thinsp;=\u0026thinsp;3.80, f2\u0026thinsp;=\u0026thinsp;77.29), confirming that its release behavior closely matched that of Formit XR across the full dissolution period. F5 demonstrated moderate similarity (f1\u0026thinsp;=\u0026thinsp;6.83, f2\u0026thinsp;=\u0026thinsp;67.17), whereas F1 showed borderline similarity (f1\u0026thinsp;=\u0026thinsp;11.70, f2\u0026thinsp;=\u0026thinsp;54.81). In contrast, F2 did not meet the similarity criteria, with an f1 value of 17.63 and an f2 value of 47.95, suggesting a dissimilar release profile compared to the reference. These findings further support the selection of F3 and F4 as the most optimized formulations for extended-release Metformin HCl delivery.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of dissolution profiles of the test formulations with the reference product using f₁ and f₂ factors.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCompared Formulae\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ef1 Value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ef2 Value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR versus F1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e54.81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR versus F2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e17.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e47.95\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR versus F3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e81.67\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR versus F4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e77.29\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR versus F5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e67.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec25\" class=\"Section2\"\u003e \u003ch2\u003e3.5. Release Kinetics Modeling\u003c/h2\u003e \u003cp\u003eThe in vitro release data of metformin from the prepared matrix tablet formulations (R, F1-F5) were subjected to kinetic analysis to determine the release mechanism. The data were fitted to zero-order, first-order, Higuchi, and Korsmeyer-Peppas mathematical models. The correlation coefficients (R\u0026sup2;) and release constants for each formulation and model are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eAnalysis of the R\u0026sup2; values revealed that the drug release kinetics varied across the formulations. The first-order kinetics model showed high correlation (R\u0026sup2; \u0026gt; 0.99) for the reference formulation (R) and formulations F1, F2, F3, and F4. This indicates that for these formulations, the rate of drug release was dependent on the concentration remaining in the matrix, which is typical for hydrophilic swellable systems containing highly soluble drugs such as metformin [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Conversely, formulation F5 demonstrated the highest correlation (R\u0026sup2; = 0.9973) with the Higuchi model indicating diffusion-dominant release. The R\u0026sup2; values for the Korsmeyer-Peppas model exceeded 0.966 for all formulations, indicating a very strong fit.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCorrelation coefficients for release data of metformin from different formulations after curve fitting to zero-order, first-order, Higuchi, and Korsmeyer Peppas model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eZero Order Model\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eFirst Order Model\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003eHiguchi Model\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c11\" namest=\"c8\"\u003e \u003cp\u003eKorsmeyer Peppas Model\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCode\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eK0\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eR2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eK1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eR2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eKH\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eR2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003ekk\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eR2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eDrug transport mechanism\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.1793\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8869\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.0022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9988\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4.4181\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9926\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.5294\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9904\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eAnomalous release\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.1793\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9283\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.0019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9927\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4.2847\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9769\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.3064\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9668\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eAnomalous release\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.1568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9110\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.0014\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9918\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.8060\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9884\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.3683\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9698\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eAnomalous release\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.1780\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8797\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.0021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9879\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4.3874\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9845\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.5070\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9818\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eAnomalous release\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.1780\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.0021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9911\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4.4321\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9788\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.5189\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9706\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eAnomalous release\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.1736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9184\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.0023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9504\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4.2137\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9973\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.6885\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9883\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eAnomalous release\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eGiven the strong correlation with the Korsmeyer-Peppas model for all formulations, the release exponent (n-value) was used to determine the predominant drug release mechanism. The calculated n values for all formulations ranged from 0.47 to 0.62. According to the Korsmeyer-Peppas model for cylindrical matrix systems, these n values indicate that the drug release mechanism for all tested formulations is dominated by anomalous (non-Fickian) transport [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. This suggests a complex release behavior involving a combination of diffusion and polymer relaxation, swelling, or erosion. Overall, the kinetic analysis demonstrates that while most formulations, including the reference, predominantly follow first-order release behavior, they all share anomalous release mechanism, highlighting the complex release behavior.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section2\"\u003e \u003ch2\u003e3.6. Statistical Evaluation of Dissolution Performance\u003c/h2\u003e \u003cp\u003eTo complement the empirical results, additional statistical analyses are performed on the dissolution data. First, we assessed how each formulation\u0026rsquo;s similarity factor (f₂) (relative to the innovator) correlates with its kinetic parameters from the Korsmeyer\u0026ndash;Peppas (K\u0026ndash;P) model. Second, we evaluated the accuracy of the AI-predicted dissolution profiles against the experimental profiles using several error and correlation metrics. Third, we applied hierarchical cluster analysis (HCA) to group the formulations based on their f₂, K\u0026ndash;P rate constant (kₖ), and diffusion exponent (n).\u003c/p\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003e3.6.1. Correlation of f₂ with Kinetic Parameters\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e lists the f₂ values and K\u0026ndash;P parameters for formulations F1\u0026ndash;F5. We calculated Pearson correlation coefficients between f₂ and each kinetic parameter. As shown in Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, f₂ correlates strongly and positively with kₖ (r\u0026thinsp;=\u0026thinsp;+\u0026thinsp;0.94, p\u0026thinsp;=\u0026thinsp;0.017) and strongly but negatively with n (r = \u0026minus;\u0026thinsp;0.87, p\u0026thinsp;=\u0026thinsp;0.049), both statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). In other words, formulations with higher apparent release rate constants and more Fickian release (lower n) generally tended to show higher similarity to the reference, as seen with F3 (kₖ = 0.5070, n\u0026thinsp;=\u0026thinsp;0.55, f₂ = 81.67) and F4 (kₖ = 0.5189, n\u0026thinsp;=\u0026thinsp;0.56, f₂ = 77.29). F5, however, represents an exception: despite having the highest kₖ value (0.6885) and the lowest n (0.47), its similarity factor was lower (f₂ = 67.17) because its late-stage release became faster than the reference. The correlation between f₂ and R\u0026sup2; of the K\u0026ndash;P fit was moderate (r\u0026thinsp;=\u0026thinsp;+\u0026thinsp;0.62) but not significant (p\u0026thinsp;=\u0026thinsp;0.18), indicating goodness-of-fit alone is a minor factor. Overall, faster diffusion-controlled kinetics (higher kₖ and n\u0026thinsp;\u0026asymp;\u0026thinsp;0.5) produced dissolution curves most similar to the innovator (F3 and F4), whereas slow, relaxation-controlled formulations (F1, F2) had low f₂ values. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e visualizes the correlation of similarity factor f₂ with release rate constant (kₖ), and release exponent (n).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSimilarity factors (f₂) versus innovator and Korsmeyer\u0026ndash;Peppas kinetic parameters\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFormulation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ef₂\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ekₖ (1/h)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eR\u0026sup2; (K\u0026ndash;P)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e54.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.3064\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9668\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e47.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.3683\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9698\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e81.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.5070\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9818\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e77.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.5189\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9706\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e67.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.6885\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9883\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e65.8\u0026thinsp;\u0026plusmn;\u0026thinsp;13.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.48\u0026thinsp;\u0026plusmn;\u0026thinsp;0.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.56\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.975\u0026thinsp;\u0026plusmn;\u0026thinsp;0.008\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePearson correlation of f₂ with K\u0026ndash;P parameters\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePearson \u003cem\u003er\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eInterpretation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ef₂ vs kₖ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e+\u0026thinsp;0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStrong positive correlation (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ef₂ vs \u003cem\u003en\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u0026ndash;0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.049\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStrong negative correlation (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ef₂ vs R\u0026sup2; (K\u0026ndash;P)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e+\u0026thinsp;0.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWeak positive (not significant)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section3\"\u003e \u003ch2\u003e3.6.2. In Vitro vs. AI-Predicted Dissolution Correlation\u003c/h2\u003e \u003cp\u003eWe next quantified how well the AI model\u0026rsquo;s predicted dissolution profiles matched the experimental (in vitro) data. For each formulation, we computed mean absolute percentage error (MAPE), root mean squared error (RMSE), Pearson\u0026rsquo;s correlation (r), concordance correlation coefficient (CCC), and bias (mean percent error) between predicted and observed percent re-leased at each time point. All formulations achieved MAPE\u0026thinsp;\u0026lt;\u0026thinsp;14% (most\u0026thinsp;\u0026lt;\u0026thinsp;10%) and RMSE\u0026thinsp;\u0026lt;\u0026thinsp;11%, indicating acceptable overall accuracy; mean MAPE across F1\u0026ndash;F5 was \u0026asymp;\u0026thinsp;8.6% (below the typical 10% threshold) as illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The AI predictions for F3 and F4 were especially accurate: MAPE\u0026thinsp;\u0026asymp;\u0026thinsp;6% and RMSE\u0026thinsp;\u0026asymp;\u0026thinsp;5%, with r and CCC near 0.96\u0026ndash;0.97 and close to zero.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003ePearson r\u0026thinsp;\u0026ge;\u0026thinsp;0.90 and CCC\u0026thinsp;\u0026ge;\u0026thinsp;0.90 for all except F2 indicate strong agreement overall. With reference to Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e, the largest deviation occurred for F2 (MAPE 13.1%, bias \u0026minus;\u0026thinsp;6.8%), reflecting its slower-than-predicted release profile. Bland\u0026ndash;Altman and equivalence testing (\u0026plusmn;\u0026thinsp;10% bounds) further confirmed that only F3 and F4 were statistically equivalent to experimental results (bias\u0026thinsp;\u0026asymp;\u0026thinsp;0). These findings quantitatively validate the AI model\u0026rsquo;s predictive power: it reliably captured the dissolution behavior, particularly for the optimal formulations F3 and F4.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab8\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAccuracy metrics comparing AI-predicted vs. experimental dissolution\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFormulation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMAPE (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRMSE (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePearson \u003cem\u003er\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCCC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eBias (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eInterpretation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026ndash;3.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSlightly under-predicted\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e13.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026ndash;6.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eModerate; AI underestimates release\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e+\u0026thinsp;1.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eExcellent prediction\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e+\u0026thinsp;0.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eExcellent prediction\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026ndash;2.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eGood agreement\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026ndash;2.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec29\" class=\"Section3\"\u003e \u003ch2\u003e3.6.3. Hierarchical Cluster Analysis of Formulations\u003c/h2\u003e \u003cp\u003eFinally, we performed agglomerative HCA (Ward\u0026rsquo;s method) on the formulations (F1\u0026ndash;F5) plus the reference (R), using standardized f₂, kₖ, and n values. The resulting dendrogram reveals three clear clusters. Formulations F3 and F4 clustered tightly with the reference R, reflecting their nearly identical kinetics (kₖ\u0026asymp;0.52 h⁻\u0026sup1;, n\u0026thinsp;\u0026asymp;\u0026thinsp;0.55) and high f₂ (\u0026gt;\u0026thinsp;75). A second cluster com-prised F1 and F2, which both have low kₖ (0.31\u0026ndash;0.37) and higher n (\u0026ge;\u0026thinsp;0.58), corresponding to their slow, relaxation-controlled release (f₂\u0026lt;55). F5 formed its own branch (highest kₖ=0.6885, lowest n\u0026thinsp;=\u0026thinsp;0.47), indicating a distinct diffusion-dominant profile as illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThese clustering results statistically reinforce the earlier findings: F3 and F4 are the only formulations that fall into the same cluster as the reference (Cluster I), confirming their kinetic and dissolution similarity. Cluster II (F1\u0026ndash;F2) is distinct with slower kinetics and lower f₂, while F5 (Cluster III) is an isolated fast-release outlier.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Conclusions","content":"\u003cp\u003eThis study demonstrates the feasibility of using a large language model, ChatGPT (GPT-4, OpenAI), as a supportive tool in the early stages of pharmaceutical formulation development. By providing the AI with the available pharmaceutical ingredients in our laboratory and their formulation roles, and a target dissolution profile, ChatGPT (GPT-4, OpenAI) was able to generate multiple sustained release metformin matrix tablet compositions that were subsequently validated through experimental testing. Among the 5 AI generated formulations, F3 and F4 closely reproduced the dissolution behavior of the reference product, as confirmed by similarity factor analysis, release kinetics modelling, and comprehensive statistical evaluation.\u003c/p\u003e \u003cp\u003eThe experimental results demonstrated that ChatGPT (GPT-4, OpenAI) has the capacity to suggest polymer combinations and ratios that are translated into physically manufacturable tablets with predictable release performance. The strong agreement between AI predicted and experimental dissolution profiles for the suggested formulations, F3 and F4, further supports the efficacy of AI assisted formulation design when combined with pharmaceutical expertise and laboratory validation. Importantly, the study also revealed limitations, as certain AI generated formulations deviated from the reference product dissolution profile due to retardation or accelerated late-stage re-lease, emphasizing the need for experimental confirmation.\u003c/p\u003e \u003cp\u003eOverall, this work demonstrates that large language models are not replacements for formulation scientists but can serve as effective decision support tools to reduce trial and error experimentation, support excipients selection, and accelerate early formulation screening when combined with pharmaceutical expertise and experimental validation. The integration of AI driven predictions with experimental validation represents a practical approach that may contribute to more efficient development of extended-release oral dosage forms across a wide range of drug substances.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003e Conflicts of Interest\u003c/b\u003e:\u003c/h2\u003e \u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e \u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThis research was funded by Alfaisal University internal research grant (IRG 23522).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConceptualization: M.A., K.S, G.M.M; Methodology: M.A., K.S., S.A., G.M.M; AI model design and formulation simulation: K.S., S.A.; Experimental supervision and tablet manufacturing: G.M.M., A.F.A.; Investigation and laboratory work: A.B.A.; Kinetic modeling, similarity factor analysis, and statistical evaluation: A.A.I., M.A; Formal analysis: M.A., A.A.I., K.S; Writing original draft preparation: M.A., K.S; Writing review and editing: M.A., A.A.I.; Project administration and leadership: M.A., K.S, G.M.M; Funding acquisition: M.A.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors gratefully acknowledge the Research Office at Alfaisal University for financial support through the Internal Research Grant that enabled the completion of this work. The authors also acknowledge the technical and logistical support provided during the experimental phase of the study.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe data generated and analyzed during this study are included in this published article. Additional datasets generated during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eKant S, Deepika. Roy \u0026middot; Saheli. Artificial intelligence in drug discovery and development: transforming challenges into opportunities. Discover Pharm Sci. 2025;1(1):7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/S44395-025-00007-3\u003c/span\u003e\u003cspan address=\"10.1007/S44395-025-00007-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. 2025 1:1.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlanco-Gonz\u0026aacute;lez A, Cabez\u0026oacute;n A, Seco-Gonz\u0026aacute;lez A, et al. The Role of AI in Drug Discovery: Challenges, Opportunities, and Strategies. Pharmaceuticals. 2023;16(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/PH16060891/S1\u003c/span\u003e\u003cspan address=\"10.3390/PH16060891/S1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlmanasra S, Suwais K. Analysis of ChatGPT-Generated Codes Across Multiple Programming Languages. IEEE Access. 2025;13:23580\u0026ndash;96. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ACCESS.2025.3538050\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2025.3538050\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Angelis L, Baglivo F, Arzilli G, et al. ChatGPT and the rise of large language models: the new AI-driven infodemic threat in public health. Front Public Health. 2023;11:1166120. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/FPUBH.2023.1166120/BIBTEX\u003c/span\u003e\u003cspan address=\"10.3389/FPUBH.2023.1166120/BIBTEX\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSallam M, Sallam M. ChatGPT Utility in Healthcare Education, Research, and Practice: Systematic Review on the Promising Perspectives and Valid Concerns. Healthcare 2023, Vol 11,. 2023;11(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/HEALTHCARE11060887\u003c/span\u003e\u003cspan address=\"10.3390/HEALTHCARE11060887\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu J, Ma Y, Wang J, Xiao M. The Application of ChatGPT in Medicine: A Scoping Review and Bibliometric Analysis. J Multidiscip Healthc. 2024;17:1681\u0026ndash;92. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2147/JMDH.S463128\u003c/span\u003e\u003cspan address=\"10.2147/JMDH.S463128\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMu Y, He D. The Potential Applications and Challenges of ChatGPT in the Medical Field. Int J Gen Med. 2024;17:817\u0026ndash;26. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2147/IJGM.S456659\u003c/span\u003e\u003cspan address=\"10.2147/IJGM.S456659\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaity S, Saikia MJ, Maity S, Saikia MJ. Large Language Models in Healthcare and Medical Applications: A Review. Bioengineering 2025, Vol 12,. 2025;12(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/BIOENGINEERING12060631\u003c/span\u003e\u003cspan address=\"10.3390/BIOENGINEERING12060631\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao A, Wu Y. Future implications of ChatGPT in pharmaceutical industry: drug discovery and development. Front Pharmacol. 2023;14:1194216. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/FPHAR.2023.1194216\u003c/span\u003e\u003cspan address=\"10.3389/FPHAR.2023.1194216\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRavi Kiran AVVV, Kumari GK, Krishnamurthy PT. ChatGPT in Drug Discovery Process. Adv Pharm Bull. 2023;14(1):5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.34172/APB.2024.007\u003c/span\u003e\u003cspan address=\"10.34172/APB.2024.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEuropean Federation of Pharmaceutical Industries and Associations (EFPIA). The Pharmaceutical Industry in Figures: Key Data 2023. Brussels, Belgium: EFPIA. 2023. Accessed December 13, 2025.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTimmons J, Boyle J. Metformin. Diabetes Drug Notes. Published online August. 2023;17:30\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/9781119785033.ch2\u003c/span\u003e\u003cspan address=\"10.1002/9781119785033.ch2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng M, Ren L, Jia X, Wang J, Cong B. Understanding the action mechanisms of metformin in the gastrointestinal tract. Front Pharmacol. 2024;15:1347047. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/FPHAR.2024.1347047\u003c/span\u003e\u003cspan address=\"10.3389/FPHAR.2024.1347047\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCirilli M, Moutaharrik S, Palugan L, et al. Gastroprotective systems for prolonged release of metformin based on osmotically driven expansion. Int J Pharm. 2025;681. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijpharm.2025.125856\u003c/span\u003e\u003cspan address=\"10.1016/j.ijpharm.2025.125856\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThe United States Pharmacopeial Convention. USP 1216 tablet friability. United States Pharmacopeia. 32:22\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eU.S, Pharmacopoeial, Convention. (905) Uniformity of Dosage Units. Stage 6 Harmonization. Vol 3. United States Pharmacopeia; 2011.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eU.S. of Pharmacopoeia. The United States Pharmacopeia\u0026thinsp;\u0026lt;\u0026thinsp;701\u0026thinsp;\u0026gt;\u0026thinsp;Disintegration. United States Pharmacopeial Convention. 2020;4:2\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePharmacopeial Convention US. Pharmacopeial Guidelines Dissolution, \u0026lt;\u0026thinsp;711\u0026thinsp;\u0026gt;\u0026thinsp;Dissolution. United States Pharmacopeia. 2011;1:1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDange YD, Honmane SM, Bhinge SD, Salunkhe VR, Jadge DR. Development and validation of uv-spectrophotometric method for estimation of metformin in bulk and tablet dosage form. Indian J Pharm Educ Res. 2017;51(4):S754\u0026ndash;60. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5530/IJPER.51.4S.109\u003c/span\u003e\u003cspan address=\"10.5530/IJPER.51.4S.109\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHamzah A, Maswadeh, Othman A, Al-Hanbali RA, Kanaan, Ashok K, Shakya A, Maraqa. Testing lyoequivalency for three commercially sustained-release tablets containing diltiazem hydrochloride. Acta Pol Pharm. 2010;67(1):93\u0026ndash;97. Accessed December 13, 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pubmed.ncbi.nlm.nih.gov/20210085/\u003c/span\u003e\u003cspan address=\"https://pubmed.ncbi.nlm.nih.gov/20210085/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbdulkarim M, Adam DR, Zahrani Y, Al, et al. Development of a metformin matrix tablet: a comparative study with marketed sustained release formulation. Farmacia. 2025;73:4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.31925/farmacia.2025.4.11\u003c/span\u003e\u003cspan address=\"10.31925/farmacia.2025.4.11\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatel S, Kaushal AM, Bansal AK. Compression Physics in the Formulation Development of Tablets. Crit Rev Ther Drug Carrier Syst. 2006;23(1):1\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1615/CRITREVTHERDRUGCARRIERSYST.V23\u003c/span\u003e\u003cspan address=\"10.1615/CRITREVTHERDRUGCARRIERSYST.V23\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNokhodchi A, Ford JL, Rowe PH, Rubinstein MH. The effects of compression rate and force on the compaction properties of different viscosity grades of hydroxy propyl methyl cellulose 2208. Int J Pharm. 1996;129(1\u0026ndash;2):21\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/0378-5173(95)04236-9\u003c/span\u003e\u003cspan address=\"10.1016/0378-5173(95)04236-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eObeidat WM, Nokhodchi A, Alkhatib H. Evaluation of Matrix Tablets Based on Eudragit\u0026reg;E100/Carbopol\u0026reg;971P Combinations for Controlled Release and Improved Compaction Properties of Water-Soluble Model Drug Paracetamol. AAPS PharmSciTech 2015 16:5. 2015;16(5):1169\u0026ndash;1179. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1208/S12249-015-0301-5\u003c/span\u003e\u003cspan address=\"10.1208/S12249-015-0301-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePramanik A, Sahoo RN, Nandi S, Nanda A, Mallick S. Characterization of Hydration Behaviour and Modeling of Film Formulation. Acta Chim Slov. 2021;68:159\u0026ndash;69. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.17344/acsi.2020.6298\u003c/span\u003e\u003cspan address=\"10.17344/acsi.2020.6298\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNokhodchi A, Raja S, Patel P, Asare-Addo K. The role of oral controlled release matrix tablets in drug delivery systems. Bioimpacts. 2012;2(4):175\u0026ndash;87. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5681/bi.2012.027\u003c/span\u003e\u003cspan address=\"10.5681/bi.2012.027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHardy IJ, Windberg-Baarup A, Neri C, Byway PV, Booth SW, Fitzpatrick S. Modulation of drug release kinetics from hydroxypropyl methyl cellulose matrix tablets using polyvinyl pyrrolidone. Int J Pharm. 2007;337(1\u0026ndash;2):246\u0026ndash;53. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/J.IJPHARM.2007.01.026\u003c/span\u003e\u003cspan address=\"10.1016/J.IJPHARM.2007.01.026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAskarizadeh M, Esfandiari N, Honarvar B, Sajadian SA, Azdarpour A. Kinetic Modeling to Explain the Release of Medicine from Drug Delivery Systems. Chem Bio Eng Reviews. 2023;10(6):1006\u0026ndash;49. .202300027; W group: String: Publication.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial intelligence, GPT-4, Extended-release matrix tablets, Hydrophilic matrix systems, Metformin hydrochloride, Formulation development, Dissolution kinetics","lastPublishedDoi":"10.21203/rs.3.rs-8755354/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8755354/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground/Objectives\u003c/strong\u003e: The development of extended-release oral dosage forms remains a complicated and time consuming process, particularly during early formulation screening. Recently, artificial intelligence (AI) tools have emerged as supportive tools for formulation development in pharmaceutical research. However, experimental validation of AI-generated formulations remains limited. The present study explores a novel approach in which a large language model (ChatGPT, GPT-4, OpenAI) was used to generate sustained-release metformin hydrochloride matrix tablet compositions under predefined pharmaceutical ingredients, followed by comprehensive experimental evaluation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e: Several AI-generated formulations were prepared using the direct compression technique and evaluated for their physical attributes, in vitro dissolution behavior, release kinetics, and statistical similarity to a marketed reference product. Dissolution profiles were analyzed using similarity and difference factors (f₂ and f₁), kinetic modelling, and advanced statistical tools, including correlation analysis and clustering.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: Among the AI generated formulations, F3 and F4 showed dissolution profiles closest to the reference product, as indicated by f₂ values above 75 together with f₁ values below 15 and comparable release kinetics. The remaining formulations exhibited either slower or faster release behavior, highlighting the importance of experimental validation of AI generated outputs.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions\u003c/strong\u003e: This study demonstrates that, when provided with pharmaceutical ingredients and formulation considerations, GPT-4 proposed tablet matrix compositions that resulted in manufacturable dosage forms with predictable release behavior upon experimental evaluation. The integration of AI-driven predictions with experimental validation represents a novel and practical strategy to support rational excipient selection, reduce trial and error experimentation, and accelerate early-stage formulation development.\u003c/p\u003e","manuscriptTitle":"AI-Driven Simulation and Design of Sustained Release Metformin Tablets: Experimental Validation and Predictive Accuracy Assessment","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-16 18:55:23","doi":"10.21203/rs.3.rs-8755354/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a0334f0c-397b-4a4e-84f7-e62273642314","owner":[],"postedDate":"February 16th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-15T22:08:25+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-16 18:55:23","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8755354","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8755354","identity":"rs-8755354","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}