Open-Source Generative AI Enables De Novo Design of an Orally Bioavailable Triple GLP-1/GIP/Glucagon Receptor Agonist for Type 2 Diabetes

Open-Source Generative AI Enables De Novo Design of an Orally Bioavailable Triple GLP-1/GIP/Glucagon Receptor Agonist for Type 2 Diabetes | 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 Short Report Open-Source Generative AI Enables De Novo Design of an Orally Bioavailable Triple GLP-1/GIP/Glucagon Receptor Agonist for Type 2 Diabetes Luís Jesuino de Oliveira Andrade, Luisa Correia Matos de Oliveira, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8214757/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 Introduction: Type 2 diabetes mellitus (DM2) demands therapeutic strategies capable of surpassing the limitations of single-target incretin modulation. Triple agonism of GLP-1, GIP, and glucagon receptors (GCGR) offers synergistic metabolic benefits. Advances in open-source generative artificial intelligence (AI) now enable de novo molecular design with unprecedented precision, supporting the development of orally viable multi-receptor agonists. Objective : To design and computationally evaluate a de novo peptide capable of balanced activation of GLP-1, GIP, and GCGR while exhibiting physicochemical and pharmacokinetic properties compatible with oral administration. Methods: A fully in silico pipeline was established integrating ChemGPT-based generative peptide design, ESMFold structural prediction, CB-Dock2 docking, 100-ns molecular dynamics simulations through WebGro, and ADMET profiling using SwissADME, pkCSM, and ProTox-II. Lead candidates were prioritized based on structural stability, multi-receptor affinity, and predicted oral bioavailability. Results: ChemGPT generated 150 peptides, of which 38 met structural quality criteria and seven achieved balanced multi-receptor affinity. Lead candidate TA-071 displayed docking energies of 10.8 (GLP-1R), −10.2 (GIPR), and −10.6 kcal/mol (GCGR) and maintained stable binding throughout molecular dynamics simulations (RMSD <3 Å; persistent hydrogen-bond networks). ADMET predictions indicated high intestinal absorption (87.3%), favorable Caco-2 permeability, absence of major CYP interactions, and low predicted toxicity. Comparative analyses showed binding profiles and permeability surpassing existing injectable incretin mimetics. Conclusion: Open-source generative AI enabled the rational creation of a structurally stable, orally oriented triple incretin/ GCGR agonist. TA-071 represents a promising in silico candidate for future preclinical development in DM2 therapy. Endocrinology & Metabolism Generative AI Triple agonist Oral peptide therapeutics Type 2 diabetes Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 INTRODUCTION Type 2 diabetes mellitus (DM2) affects approximately 537 million adults globally, requiring innovative therapeutic strategies beyond conventional single-target approaches. Multi-receptor agonism targeting glucagon-like peptide-1 (GLP-1), glucose-dependent insulinotropic polypeptide (GIP), and glucagon receptors (GCGR) demonstrates superior glycemic control and weight reduction compared to mono-agonist therapies. 1 Current triple-agonist candidates remain limited to parenteral administration, constraining patient adherence and therapeutic accessibility. 2 The incretin hormones GIP and GLP-1 are essential gastrointestinal peptides secreted from intestinal K-cells and L-cells following nutrient ingestion. 3 These hormones potentiate glucose-dependent insulin secretion from pancreatic beta-cells through G-protein coupled receptor activation and intracellular cyclic adenosine monophosphate elevation. Beyond insulinotropic effects, GLP-1 modulates gastric emptying, inhibits glucagon secretion, and promotes satiety, while GIP enhances postprandial insulin responses and influences lipid metabolism. 4 Together, these incretins mediate approximately 60–70% of postprandial insulin secretion. Artificial intelligence (AI), especially deep learning, has revolutionized drug design by enabling accurate predictions of molecular properties and accelerating lead compound identification. 5 AI-driven generative models now facilitate de novo drug design, improving precision and reducing development timelines significantly. 6 Despite remarkable advances in multi-receptor agonist development, oral bioavailability remains fundamentally limited. Peptide therapeutics exhibit bioavailability below one to two percent due to enzymatic degradation, poor intestinal permeability, and hepatic first-pass metabolism. 7 Current triple-agonist candidates require subcutaneous injection, compromising patient compliance and limiting therapeutic applicability in resource-constrained settings. 8 These gaps hinder the translation of multi-receptor agonists into broadly effective clinical treatments. This study aims to demonstrate how open-source generative AI can enable the de novo design of a triple GLP-1/GIP/ GCGR agonist with favorable oral bioavailability for DM2 management. METHODOLOGY Study Design This computational study employed open-source AI platforms and public-domain webservers to design an orally bioavailable triple GLP-1, GIP, and GCGR agonist. The workflow comprised five sequential phases: receptor structure preparation, de novo peptide generation, molecular docking, molecular dynamics simulations, and pharmacokinetic profiling. Receptor Structure Preparation Crystal structures of human GLP-1 receptor (PDB: 6X18), GIP receptor (PDB: 7DTY), and GCGR (PDB: 5EE7) were retrieved from RCSB Protein Data Bank. Structures were prepared using PyMOL 2.5 for water and heteroatom removal, followed by the addition of hydrogens and energy minimization through the YASARA Energy Minimization Server (http://www.yasara.org/minimizationserver.htm). Active binding sites were identified based on co-crystallized ligand coordinates. De Novo Peptide Design Novel peptide sequences were generated using ChemGPT (https://huggingface.co/ncfrey/ChemGPT-1.2B), an open-source transformer model for molecular design. Design constraints specified peptides of 20-35 amino acids incorporating D-amino acids and N-methylated residues at protease-susceptible positions to enhance oral stability. Generated sequences were optimized through ESMFold webserver (https://esmatlas.com/resources) for three-dimensional structure prediction. Candidates with predicted local distance difference test (pLDDT) scores above 70 and alpha-helical content between 60-80% were selected for subsequent analysis. Molecular Docking Peptide-receptor binding was evaluated using CB-Dock2 webserver (http://clab.labshare.cn/cb-dock2/), which automatically identifies binding sites and performs blind docking. The platform employs the AutoDock Vina algorithm with cavity detection through protein surface analysis. Each peptide was docked against all three target receptors, with binding affinity thresholds set at -8.0 kcal/mol. Candidates demonstrating balanced affinity across the three receptors (differences <3 kcal/mol) were prioritized. Interaction analysis utilized PLIP webserver (https://plip-tool.biotec.tu-dresden.de/) to identify hydrogen bonds and hydrophobic contacts. Molecular Dynamics Simulations Conformational stability of peptide-receptor complexes was assessed through WebGro platform (https://simlab.uams.edu/), performing 100 nanosecond all-atom molecular dynamics simulations using GROMACS with AMBER99SB force field. Systems were solvated in TIP3P water boxes with 150 mM NaCl at 310 K and 1 bar pressure. Trajectory analysis included root mean square deviation (RMSD), root mean square fluctuation, and hydrogen bond monitoring. Complexes maintaining RMSD below 3 Å and hydrogen bond occupancy above 60% were considered stable. ADMET Property Prediction Pharmacokinetic properties were predicted using SwissADME (http://www.swissadme.ch) and pkCSM (http://biosig.lab.uq.edu.au/pkcsm) webservers. Evaluated parameters included human intestinal absorption, Caco-2 permeability, P-glycoprotein substrate liability, blood-brain barrier penetration, drug-likeness according to Lipinski's rules, and BOILED-Egg model predictions. Toxicity endpoints encompassed AMES mutagenicity, hERG channel blockade, hepatotoxicity, and oral acute toxicity through ProTox-II platform (https://tox-new.charite.de/protox_II/). Lead candidates required: predicted intestinal absorption >85%, Caco-2 permeability >0.90 log Papp, negative P-glycoprotein substrate status, bioavailability score ≥0.55, AMES negative, hERG IC50 >10 μM, non-hepatotoxic classification, and oral rat acute toxicity LD50 >2000 mg/kg. Data Visualization and Analysis Molecular structures were visualized using PyMOL 2.5 and protein-ligand interactions mapped through LigPlot + webserver (https://www.ebi.ac.uk/thornton-srv/software/LigPlus/). Binding affinity comparisons employed one-way ANOVA with Tukey's post-hoc test (α=0.05). Correlations between structural descriptors and pharmacokinetic parameters were assessed using Spearman's correlation coefficient. Ethical Considerations This study utilized exclusively in silico methodologies based on publicly available crystallographic data and computational bioinformatics tools. No human subjects, animal models, biological samples, or experimental procedures were involved. Consequently, ethical approval from an institutional review board or animal ethics committee was not required for this research, in accordance with current guidelines for purely computational studies. RESULTS Receptor Structure Preparation and Active Site Characterization Crystal structures of GLP-1 receptor (PDB: 6X18), GIP receptor (PDB: 7DTY), and GCGR (PDB: 5EE7) were successfully retrieved and prepared. Energy minimization converged with final potential energies of -124,847 kJ/mol (GLP-1R), -118,562 kJ/mol (GIPR), and -121,394 kJ/mol (GCGR). Active site analysis identified orthosteric binding pockets at the extracellular-transmembrane interface, with critical residues including Arg121, Glu127, Glu128, Lys197, Glu364, and Arg380 for GLP-1R; Arg183, Glu247, Asp254 for GIPR; and Glu127, Lys187, Asp364 for GCGR (Figure 1). De Novo Peptide Generation and Structure Prediction ChemGPT generated 150 candidate peptide sequences (22-34 amino acids, mean 27.3 ± 3.2) incorporating an average of 4.6 ± 1.8 D-amino acids and strategic N-methylation at protease-susceptible sites. ESMFold structure prediction identified 38 candidates (25.3%) with predicted local distance difference test scores above 70, indicating high structural confidence. These candidates exhibited alpha-helical content between 60-80% (mean 68.4 ± 6.2%) with amphipathic helix formation. Ramachandran analysis confirmed that 96.4 ± 2.1% of residues were in allowed regions, validating structural quality for docking studies (Figure 2). Molecular Docking Analysis CB-Dock2 evaluation revealed seven candidates demonstrating balanced affinity across all three receptors (binding energy differences <3.0 kcal/mol). The lead candidate TA-071 exhibited binding affinities of -10.8 kcal/mol (GLP-1R), -10.2 kcal/mol (GIPR), and -10.6 kcal/mol (GCGR), indicating equipotent multi-receptor engagement. TA-071 sequence: H-Tyr-dAla-Glu-Gly-Thr-Phe(Me)-Ile-Ser-Asp-dVal-Ser-dSer-Tyr-Leu-Glu-Gly-Gln-Ala-Ala-Lys-Glu-Phe-Ile-dAla-Trp-Leu-Val-Arg-NH₂ (28 residues). PLIP analysis revealed extensive hydrogen bonding: eight bonds with GLP-1R (Arg121, Glu127, Glu128, Tyr152, Glu364, Arg380), nine with GIPR, and seven with GCGR. Off-target selectivity was confirmed with significantly weaker binding to peptide YY receptor 1 (-5.3 ± 0.8 kcal/mol) and corticotropin-releasing factor receptor 1 (-4.9 ± 0.9 kcal/mol), both p<0.001 versus target receptors (Figure 3). Molecular Dynamics WebGro simulations spanning 100 nanoseconds confirmed stable peptide-receptor complexes. Root mean square deviation analysis showed equilibration after 16-22 nanoseconds with maintained RMSD values of 2.1 ± 0.3 Å (GLP-1R), 2.4 ± 0.4 Å (GIPR), and 1.9 ± 0.2 Å (GCGR), all below the 3.0 Å threshold. Root mean square fluctuation revealed minimal core fluctuation (0.8-1.4 Å for residues 8-23) with flexibility concentrated at termini (3.2-4.8 Å). Hydrogen bond occupancy exceeded 60% for key interactions, with GLP-1R maintaining average 6.8 ± 1.2 persistent bonds, including Glu3-Arg121 (89% occupancy), Asp9-Tyr152 (76%), and Glu15-Arg380 (82%). Molecular mechanics Poisson-Boltzmann surface area calculations estimated binding free energies of -62.4 ± 8.7 kcal/mol (GLP-1R), -58.9 ± 9.3 kcal/mol (GIPR), and -61.7 ± 8.1 kcal/mol (GCGR), confirming thermodynamically favorable interactions (Figure 4). ADMET Property Predictions SwissADME analysis revealed favorable drug-likeness despite its peptidic nature (molecular weight 3,247 Da, TPSA 1,186 Ų). Lipophilicity consensus yielded log P of -2.8 ± 1.2. P-glycoprotein substrate prediction was negative (probability 0.23), indicating low efflux liability. pkCSM predictions demonstrated promising oral bioavailability characteristics: human intestinal absorption 87.3% (threshold >85%), Caco-2 permeability 1.08 log Papp (threshold >0.90), bioavailability score 0.56 (threshold ≥0.55). Predicted volume of distribution was 0.68 L/kg, plasma protein binding 72.4%, clearance 8.3 mL/min/kg, and half-life 4.2 hours. Cytochrome P450 interaction profiling predicted neither substrate nor inhibitor activity for major isoforms, reducing drug-drug interaction risk (Figure 5). ProTox-II analysis indicated favorable safety predictions: AMES mutagenicity negative (probability 0.12), hERG IC₅₀ 24.6 μM (threshold >10 μM), hepatotoxicity negative (probability 0.31), oral rat acute toxicity LD₅₀ 2,400 mg/kg (toxicity class IV). All parameters met predefined acceptance criteria for lead candidate advancement. Comparative Analysis TA-071 binding affinities compared favorably to approved therapeutics: exenatide (-9.2 kcal/mol), liraglutide (-9.7 kcal/mol), and tirzepatide (-10.1 kcal/mol GLP-1R, -11.3 kcal/mol GIPR). Notably, TA-071's predicted intestinal absorption (87.3%) and Caco-2 permeability (1.08 log Papp) substantially exceeded subcutaneously administered comparators, which exhibit negligible oral bioavailability (<2%). Correlation analysis revealed D-amino acid number strongly correlated with predicted half-life (Spearman's ρ = 0.78, p<0.001) and inversely with hepatic clearance (ρ = -0.71, p<0.001). N-methylation frequency correlated positively with Caco-2 permeability (ρ = 0.64, p<0.01). Alpha-helical content showed positive association with receptor binding affinity (ρ = 0.69-0.73, p<0.001 across all receptors) (Figure 6). TA-071, a de novo designed triple agonist, demonstrated balanced and high-affinity binding to GLP-1R, GIPR, and GCGR. It exhibited exceptional predicted oral bioavailability, favorable ADMET properties, and superior permeability compared to subcutaneous reference therapeutics, establishing its promising preclinical candidacy. DISCUSSION The present study demonstrates that open-source generative AI can effectively facilitate the rational design of an orally bioavailable triple agonist targeting GLP-1, GIP, and glucagon receptors. Our results highlight TA-071’s superior multi-receptor affinity, structural stability, and predicted pharmacokinetic profile, surpassing current injectable therapeutics. This study underscores the transformative potential of AI in creating orally viable peptide therapeutics for complex metabolic diseases like DM2. The preparation of high-resolution receptor structures is an essential prerequisite for rational drug design, enabling precise characterization of orthosteric binding pockets. 9 Advances in crystallography and cryo-electron microscopy have enabled high-resolution receptor models, facilitating identification of orthosteric and allosteric sites essential for selective ligand interaction. 10 Subsequent active site mapping identifies key residues governing ligand recognition and binding affinity, a process that is fundamental for understanding molecular recognition events. 11 Advanced computational tools now facilitate the accurate definition of these pharmacologically relevant cavities, establishing a robust structural foundation for downstream molecular docking and virtual screening campaigns. 12 Aligning with the prerequisite for high-resolution structural characterization in rational drug design, we successfully optimized the receptor models to thermodynamic stability. Unlike generalized models, our targeted structures exhibit detailed active site residues at extracellular-transmembrane interfaces unique to GLP-1R, GIPR, and GCGR. This precision complements broader findings emphasizing conformational adaptations and receptor-ligand specificity, reinforcing a comprehensive structural foundation for rational drug design. De novo peptide generation and structure prediction have advanced through integrating AI and computational modeling, enabling the design of novel peptides with tailored functions and stability. 13 Techniques such as sequence-based language models and physics-based design improve peptide binding specificity and conformational accuracy. 14 Recent developments incorporate generative pre-trained transformers that facilitate high-throughput peptide generation with multimodal screening approaches, integrating sequence and structural information for enhanced bioactivity prediction. 15 These approaches accelerate rational peptide engineering, expanding therapeutic possibilities in drug development. The integration of AI-driven de novo peptide generation and structure prediction, exemplified by models like ChemGPT, allows for efficient design of peptides with tailored sequences and confirmed structural quality. Our results reflect these advancements, demonstrating amphipathic helices and high confidence in structural predictions, and ESMFold-based structure validation confirmed high-confidence predictions with predominant alpha-helical amphipathic architectures, substantiating the literature's premise that integrated sequence-structure approaches yield structurally viable candidates suitable for subsequent functional screening. Molecular docking analysis is pivotal in predicting ligand binding conformations and affinities within receptor sites, employing advanced algorithms to enhance accuracy and efficiency. 16 Techniques integrate high-throughput virtual screening and scoring functions to identify promising drug candidates. 17 Machine learning-augmented scoring functions, particularly correction-term methodologies, substantially enhance traditional docking algorithms' predictive capabilities across diverse benchmark datasets. 18 This approach, combined with molecular dynamics, refines predictions under physiological conditions, expanding drug discovery capabilities. 19 Our molecular docking results exhibit balanced multi-receptor affinity and extensive hydrogen bonding, aligning with advanced docking algorithms such as CB-Dock2, which incorporates template-based cavity detection to enhance binding pose prediction accuracy. This demonstrates consistent affinity profiles across incretin receptor members, supporting the robustness of our ligand-receptor interaction predictions and confirming the predictive validity of current docking methodologies for rational peptide-receptor optimization. Molecular dynamics simulations have revolutionized biomolecular research by capturing atomic-level protein behavior and dynamics with unprecedented temporal resolution. 20 Advances in force fields and enhanced sampling have improved accuracy, revealing mechanisms underpinning protein function and informing experimental strategies. 21 These computational methodologies facilitate structure-function elucidation, enhanced sampling techniques for conformational exploration, and accelerate structure-based drug design through mechanistic insights into protein-ligand interactions. 22 Our molecular dynamics simulations exemplify how advanced computational methodologies elucidate structure-function relationships through equilibration metrics, fluctuation patterns, and binding thermodynamics. The molecular dynamics simulations in our study provided in-depth, atomic-level data on the behavior of the evaluated proteins, expanding our understanding beyond the peptide-receptor stability results from WebGro. ADMET property predictions have undergone transformative advancement through machine learning integration, addressing critical bottlenecks in pharmaceutical development pipelines. 23 Contemporary methodologies leverage graph neural networks and ensemble frameworks to decipher complex structure-property relationships, substantially outperforming traditional quantitative structure-activity relationship models while providing scalable alternatives. 24 Advanced pretraining strategies incorporating quantum chemistry simulations enhance molecular representation learning, achieving state-of-the-art performance across multiple ADMET endpoints. 25 These computational innovations reduce late-stage attrition by enabling early risk assessment and compound prioritization, thereby expediting therapeutic development. Our SwissADME and pkCSM analyses, focused on drug-likeness, bioavailability, and toxicity profiles, confirm compound suitability, showcasing practical early-stage validation complementing theoretical advances. Employing consensus methodologies for lipophilicity, permeability, and toxicological endpoints, these analyses demonstrated favorable drug-likeness characteristics, minimal efflux liability, and acceptable safety profiles, thereby validating machine learning-driven risk stratification strategies that expedite lead optimization and preclinical decision-making processes. Comparative analysis between marketed drugs and de novo candidates constitutes a fundamental approach in pharmaceutical development. 26 Methodologies leverage existing therapeutic scaffolds through late-stage modifications, prodrug strategies, and repurposing approaches to accelerate candidate optimization. 27 Benchmarking studies reveal successful translation rates and inform structure-activity relationship refinements that distinguish first-in-class innovations from incremental improvements. 28 Such comparative frameworks facilitate evidence-based prioritization while maintaining therapeutic target validation rigor. 29 These strategies complement but differ fundamentally. The comparative analysis between marketed drugs and TA-071 reveals that while traditional therapeutics rely on established molecular scaffolds and modifications, TA-071 exhibits superior receptor binding affinities and enhanced oral bioavailability. Structural features such as D-amino acid incorporation and N-methylation correlate strongly with pharmacokinetic improvements, underscoring TA-071's potential as a multi-receptor agonist surpassing subcutaneous counterparts when analyzed from a computational point of view. This study demonstrates that open-source generative AI effectively enables rational design of orally bioavailable triple GLP-1, GIP, and glucagon receptor agonists. The lead candidate exhibits superior multi-receptor affinity, structural stability, and predicted pharmacokinetics compared to existing injectable therapeutics, highlighting AI’s transformative potential in peptide therapeutic development for DM2. CONCLUSION Open-source generative AI accelerates the design of novel multi-receptor agonists with enhanced oral bioavailability and drug-likeness, offering new avenues for innovative treatments with preclinical potential for metabolic disease management. Declarations Conflict of Interest The authors declare that they have no conflict of interest related to the publication of this manuscript. References Frías JP, Davies MJ, Rosenstock J, Pérez Manghi FC, Fernández Landó L, Bergman BK, et al. Tirzepatide versus Semaglutide Once Weekly in Patients with Type 2 Diabetes. N Engl J Med. 2021;385(6):503-515. Venniyoor A. Tirzepatide Once Weekly for the Treatment of Obesity. N Engl J Med. 2022;387(15):1433-1434. Seino Y, Fukushima M, Yabe D. GIP and GLP-1, the two incretin hormones: Similarities and differences. J Diabetes Investig. 2010;1(1-2):8-23. Baggio LL, Drucker DJ. Biology of incretins: GLP-1 and GIP. Gastroenterology. 2007;132(6):2131-57. Chen H, Engkvist O, Wang Y, Olivecrona M, Blaschke T. The rise of deep learning in drug discovery. Drug Discov Today. 2018;23(6):1241-1250. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596(7873):583-589. Hamman JH, Enslin GM, Kotzé AF. Oral delivery of peptide drugs: barriers and developments. BioDrugs. 2005;19(3):165-77. Finan B, Yang B, Ottaway N, Smiley DL, Ma T, Clemmensen C, et al. A rationally designed monomeric peptide triagonist corrects obesity and diabetes in rodents. Nat Med. 2015;21(1):27-36. Zou Y, Guo T, Fu Z, Guo Z, Bo W, Yan D, et al. A structure-based framework for selective inhibitor design and optimization. Commun Biol. 2025;8(1):422. Vénien-Bryan C, Li Z, Vuillard L, Boutin JA. Cryo-electron microscopy and X-ray crystallography: complementary approaches to structural biology and drug discovery. Acta Crystallogr F Struct Biol Commun. 2017;73(Pt 4):174-183. Du X, Li Y, Xia YL, Ai SM, Liang J, Sang P, et al. Insights into Protein-Ligand Interactions: Mechanisms, Models, and Methods. Int J Mol Sci. 2016;17(2):144. Tavares FM, Gomes AC, Assunção EM, de Medeiros JLS, Scotti MT, Scotti L, et al. Virtual Screening and Molecular Docking: Discovering Novel c-KIT Inhibitors. Curr Med Chem. 2022;29(2):166-188. Lamiable A, Thévenet P, Rey J, Vavrusa M, Derreumaux P, Tufféry P. PEP-FOLD3: faster de novo structure prediction for linear peptides in solution and in complex. Nucleic Acids Res. 2016;44(W1):W449-54. Zhang H, Saravanan KM, Wei Y, Jiao Y, Yang Y, Pan Y, et al. Deep Learning-Based Bioactive Therapeutic Peptide Generation and Screening. J Chem Inf Model. 2023;63(3):835-845. Zhao H, Song G. AVP-GPT2: A Transformer-Powered Platform for De Novo Generation, Screening, and Explanation of Antiviral Peptides. Viruses. 2024;17(1):14. Yang C, Chen EA, Zhang Y. Protein-Ligand Docking in the Machine-Learning Era. Molecules. 2022;27(14):4568. Cardoso MH, Orozco RQ, Rezende SB, Rodrigues G, Oshiro KGN, Cândido ES, et al. Computer-Aided Design of Antimicrobial Peptides: Are We Generating Effective Drug Candidates? Front Microbiol. 2020;10:3097. Zheng L, Meng J, Jiang K, Lan H, Wang Z, Lin M, et al. Improving protein-ligand docking and screening accuracies by incorporating a scoring function correction term. Brief Bioinform. 2022;23(3):bbac051. Wang Z, Sun H, Yao X, Li D, Xu L, Li Y, et al. Comprehensive evaluation of ten docking programs on a diverse set of protein-ligand complexes: the prediction accuracy of sampling power and scoring power. Phys Chem Chem Phys. 2016;18(18):12964-75. Filipe HAL, Loura LMS. Molecular Dynamics Simulations: Advances and Applications. Molecules. 2022;27(7):2105. Lazim R, Suh D, Choi S. Advances in Molecular Dynamics Simulations and Enhanced Sampling Methods for the Study of Protein Systems. Int J Mol Sci. 2020;21(17):6339. Wu X, Xu LY, Li EM, Dong G. Application of molecular dynamics simulation in biomedicine. Chem Biol Drug Des. 2022;99(5):789-800. Venkataraman M, Rao GC, Madavareddi JK, Maddi SR. Leveraging machine learning models in evaluating ADMET properties for drug discovery and development. ADMET DMPK. 2025;13(3):2772. Wang Y, Wang J, Yang Y, Ren Y, Bai H, Li H. From matrix factorization to graph neural networks: Advances in computational drug repositioning. Drug Discov Today. 2025;30(11):104499. Kim J, Chang W, Ji H, Joung I. Quantum-Informed Molecular Representation Learning Enhancing ADMET Property Prediction. J Chem Inf Model. 2024;64(13):5028-5040. Baró EL, Catti F, Estarellas C, Ghashghaei O, Lavilla R. Drugs from drugs: New chemical insights into a mature concept. Drug Discov Today. 2024;29(12):104212. Zheng L, Wang W, Sun Q. Targeted drug approvals in 2023: breakthroughs by the FDA and NMPA. Signal Transduct Target Ther. 2024;9(1):46. Schuhmacher A, Hinder M, Brief E, Gassmann O, Hartl D. Benchmarking R&D success rates of leading pharmaceutical companies: an empirical analysis of FDA approvals (2006-2022). Drug Discov Today. 2025;30(2):104291. Okuyama R. Chronological Analysis of First-in-Class Drugs Approved from 2011 to 2022: Their Technological Trend and Origin. Pharmaceutics. 2023;15(7):1794. Additional Declarations The authors declare no competing interests. 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-8214757","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Short Report","associatedPublications":[],"authors":[{"id":551303744,"identity":"8eee2c69-91f5-4a00-8de6-0cf00732452d","order_by":0,"name":"Luís Jesuino de Oliveira Andrade","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAElEQVRIiWNgGAWjYBACPgbGBgYGHiCLGYg/MDAkwGQScOhgYEPWwjiDOC1IgJmHKC0Syc0fGGRs7A2O8x58bNtml8fP3sD44WMOQ555Ay4tiW0SDDxpiRsO8yUb57YlF0v2HGCWnLmNoVjmAG4tQL8cTjA4zGMmndvGnLjhRgIbM+82hsQZOB2WCHQYz397sBbLtnqitDQAHXaAcQNIC2PbYSK08DwE+SU5ceZhHmPDnnPHE2f2HGwG+kWiWAKHFn729McfGHvs7PnOnzF88KOsOrGfvfngh4/bbPJwaQEB5r89UBYjOJpAkcuATwMI/IAx/hBQOApGwSgYBSMSAAAthFHtBhERUQAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-7714-0330","institution":"Department of Health of the State University of Santa Cruz - Ilhéus – Bahia – Brazil.","correspondingAuthor":true,"prefix":"","firstName":"Luís","middleName":"Jesuino de Oliveira","lastName":"Andrade","suffix":""},{"id":551303745,"identity":"50f52eb0-8813-40b9-a5ca-1e40ba24dfdb","order_by":1,"name":"Luisa Correia Matos de Oliveira","email":"","orcid":"https://orcid.org/0000-0001-6128-4885","institution":"SENAI CIMATEC University Center – Salvador – Bahia – Brazil.","correspondingAuthor":false,"prefix":"","firstName":"Luisa","middleName":"Correia Matos","lastName":"de Oliveira","suffix":""},{"id":551303746,"identity":"64288ecf-311e-462d-84cd-ec18a3c9928d","order_by":2,"name":"Luanna Lopes da Silva Ramos","email":"","orcid":"https://orcid.org/0009-0002-1499-562X","institution":"Faculty of Medicine of Salvador University – Salvador – Bahia – Brazil.","correspondingAuthor":false,"prefix":"","firstName":"Luanna","middleName":"Lopes da Silva","lastName":"Ramos","suffix":""},{"id":551303747,"identity":"901fd72c-2614-4a43-bb07-2ff513936ff5","order_by":3,"name":"Gabriela Correia Matos de Oliveira","email":"","orcid":"https://orcid.org/0000-0002-3447-3143","institution":"José Silveira Foundation - Salvador - Bahia - Brazil.","correspondingAuthor":false,"prefix":"","firstName":"Gabriela","middleName":"Correia Matos","lastName":"de Oliveira","suffix":""},{"id":551303748,"identity":"4304d116-19e1-4a24-a161-23e5a7e2835a","order_by":4,"name":"Luís Matos de Oliveira","email":"","orcid":"https://orcid.org/0000-0003-4854-6910","institution":"Department of Health of the State University of Santa Cruz - Ilhéus – Bahia – Brazil.","correspondingAuthor":false,"prefix":"","firstName":"Luís","middleName":"Matos","lastName":"de Oliveira","suffix":""}],"badges":[],"createdAt":"2025-11-26 16:02:07","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-8214757/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8214757/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97142070,"identity":"df4ca4a6-9b88-4aa5-9f4e-914ee7386cd0","added_by":"auto","created_at":"2025-12-01 10:07:18","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1540348,"visible":true,"origin":"","legend":"","description":"","filename":"OpenSourceGenerativeAIEnablesDeNovoDesignofanOrallyBioavailableTripleGLP1GIPGlucagonReceptorAgonistforType2Diabetes.docx","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/e90080059b53093027fbc53a.docx"},{"id":97141154,"identity":"8f12b656-5a3f-4c33-8888-b0a84c1d9078","added_by":"auto","created_at":"2025-12-01 10:06:19","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs8214757.json","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/1d31edf32ee607e0967faa88.json"},{"id":97117673,"identity":"a028baf6-3bb1-41fc-94b1-b57952e313c0","added_by":"auto","created_at":"2025-12-01 07:29:30","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":76428,"visible":true,"origin":"","legend":"","description":"","filename":"rs82147570enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/672a6886c1f561acd12816c0.xml"},{"id":97117682,"identity":"395bf791-5676-4c94-b041-c560e886037b","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":289637,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/a59d6148cb29f3bae08dfd8b.png"},{"id":97117683,"identity":"f015110c-4428-4efe-8203-fc04dcd369e5","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":196747,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/d5ea08be3200bc64d7cf2735.png"},{"id":97142067,"identity":"96d73aec-b084-45bd-92fc-3243eca6027b","added_by":"auto","created_at":"2025-12-01 10:07:18","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":209750,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/9b89f18d5bdf8fb705796127.png"},{"id":97117680,"identity":"c88f52c3-5cd6-4c47-9f91-6a8a439b15f0","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":386990,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/f7b32dcf2b4fe65bfa8f8396.png"},{"id":97117679,"identity":"c904a5d6-79a2-4a7f-bfb0-e13eec656ada","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":169262,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/2db6428d9348d5e0a0ba3955.png"},{"id":97117681,"identity":"13cd8083-1037-47ac-9633-439698bf1b1c","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":252166,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/fd58c97c75d01cb732aa7e3b.png"},{"id":97117689,"identity":"b1222509-36a5-4cdf-bd7b-eefb77ca3c67","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29114,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/3a5ed116a8ad9bfdf65d1a2e.png"},{"id":97117690,"identity":"a210975d-0d44-4cfc-9268-0b86ecb8f224","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14125,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/c6820f1aef710565d3a07c91.png"},{"id":97117687,"identity":"8af3556c-677a-4522-a936-0b0c6f338d71","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17945,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/7132ceffc36c236fb6f5f42e.png"},{"id":97117691,"identity":"56b59c85-46e8-4a13-92a5-2d813439f218","added_by":"auto","created_at":"2025-12-01 07:29:32","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":37873,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/a6cb4905a5c89ae6725fe648.png"},{"id":97117685,"identity":"64fe014c-a073-4292-ae12-92db2615e26c","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33518,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/b41527e2d6709ccf442774c4.png"},{"id":97142289,"identity":"d6ac852d-aa5f-48c9-8e31-a7adbe3c305b","added_by":"auto","created_at":"2025-12-01 10:07:29","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":43494,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/e9a8180822f1a9c2b7733e9a.png"},{"id":97117688,"identity":"a37ca446-0d2a-402f-8cf5-bfb9924495ea","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"xml","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":75307,"visible":true,"origin":"","legend":"","description":"","filename":"rs82147570structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/a949c5214207bb3adacdca4a.xml"},{"id":97117693,"identity":"c73f27fc-38f3-4a3d-8024-b3aba14ae58c","added_by":"auto","created_at":"2025-12-01 07:29:32","extension":"html","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":82318,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/bcde026ce752d80cae75b194.html"},{"id":97117670,"identity":"c6be0520-9932-4b70-9889-a673f5ca6a4a","added_by":"auto","created_at":"2025-12-01 07:29:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":289637,"visible":true,"origin":"","legend":"\u003cp\u003eCrystal Structure Preparation and Active Site Characterization of Incretin and GCGR\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/1b62333d3e691ff28501febb.png"},{"id":97117671,"identity":"5ed8894a-f899-4a6b-abd7-dfab2cb3e58a","added_by":"auto","created_at":"2025-12-01 07:29:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":196747,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative De Novo Designed Peptide Structure and Amphipathic Helix Characterization.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/a05eaa254688e385b6b880a7.png"},{"id":97117677,"identity":"8a4e0042-6c27-4d6e-bfff-7eebe4c7604d","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":209750,"visible":true,"origin":"","legend":"\u003cp\u003eMolecular docking analysis\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/06227aa312e3a90f41372573.png"},{"id":97142439,"identity":"c2d44f03-c9fb-4f49-be41-0b75ffe77f61","added_by":"auto","created_at":"2025-12-01 10:07:37","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":386990,"visible":true,"origin":"","legend":"\u003cp\u003eMolecular Dynamics Simulation Analysis of Peptide-Receptor Complex Stability\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/78103d22dbcd1f813c16e09a.png"},{"id":97117686,"identity":"7fbc9795-c344-4d38-80e9-b7d0a136bafb","added_by":"auto","created_at":"2025-12-01 07:29:31","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":169262,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eIn Silico\u003c/em\u003eADMET Property Predictions and Drug-Likeness Assessment.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/584f9a1b6f267067996a84d7.png"},{"id":97142594,"identity":"1f782550-50f4-42fa-971c-39b6b0c14e7b","added_by":"auto","created_at":"2025-12-01 10:07:45","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":252166,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of binding affinities, pharmacokinetic parameters, and structural features between TA-071 and approved incretin-based therapeutic.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/a96d927092b46fbff4536766.png"},{"id":97249326,"identity":"fc1b5df4-a88e-4ecd-aae9-99bc66443507","added_by":"auto","created_at":"2025-12-02 13:12:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2132291,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8214757/v1/50c28fc3-ed72-4c69-9eb9-f2d2a23a1af6.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eOpen-Source Generative AI Enables De Novo Design of an Orally Bioavailable Triple GLP-1/GIP/Glucagon Receptor Agonist for Type 2 Diabetes\u003c/p\u003e","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eType 2 diabetes mellitus (DM2) affects approximately 537\u0026nbsp;million adults globally, requiring innovative therapeutic strategies beyond conventional single-target approaches. Multi-receptor agonism targeting glucagon-like peptide-1 (GLP-1), glucose-dependent insulinotropic polypeptide (GIP), and glucagon receptors (GCGR) demonstrates superior glycemic control and weight reduction compared to mono-agonist therapies.\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e Current triple-agonist candidates remain limited to parenteral administration, constraining patient adherence and therapeutic accessibility.\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eThe incretin hormones GIP and GLP-1 are essential gastrointestinal peptides secreted from intestinal K-cells and L-cells following nutrient ingestion.\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e These hormones potentiate glucose-dependent insulin secretion from pancreatic beta-cells through G-protein coupled receptor activation and intracellular cyclic adenosine monophosphate elevation. Beyond insulinotropic effects, GLP-1 modulates gastric emptying, inhibits glucagon secretion, and promotes satiety, while GIP enhances postprandial insulin responses and influences lipid metabolism.\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e Together, these incretins mediate approximately 60\u0026ndash;70% of postprandial insulin secretion.\u003c/p\u003e\u003cp\u003eArtificial intelligence (AI), especially deep learning, has revolutionized drug design by enabling accurate predictions of molecular properties and accelerating lead compound identification.\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e AI-driven generative models now facilitate de novo drug design, improving precision and reducing development timelines significantly.\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eDespite remarkable advances in multi-receptor agonist development, oral bioavailability remains fundamentally limited. Peptide therapeutics exhibit bioavailability below one to two percent due to enzymatic degradation, poor intestinal permeability, and hepatic first-pass metabolism.\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e Current triple-agonist candidates require subcutaneous injection, compromising patient compliance and limiting therapeutic applicability in resource-constrained settings.\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e These gaps hinder the translation of multi-receptor agonists into broadly effective clinical treatments.\u003c/p\u003e\u003cp\u003eThis study aims to demonstrate how open-source generative AI can enable the de novo design of a triple GLP-1/GIP/ GCGR agonist with favorable oral bioavailability for DM2 management.\u003c/p\u003e"},{"header":"METHODOLOGY","content":"\u003cp\u003e\u003cstrong\u003eStudy Design\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis computational study employed open-source AI platforms and public-domain webservers to design an orally bioavailable triple GLP-1, GIP, and GCGR agonist. The workflow comprised five sequential phases: receptor structure preparation, de novo peptide generation, molecular docking, molecular dynamics simulations, and pharmacokinetic profiling.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eReceptor Structure Preparation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCrystal structures of human GLP-1 receptor (PDB: 6X18), GIP receptor (PDB: 7DTY), and GCGR (PDB: 5EE7) were retrieved from RCSB Protein Data Bank. Structures were prepared using PyMOL 2.5 for water and heteroatom removal, followed by the addition of hydrogens and energy minimization through the YASARA Energy Minimization Server (http://www.yasara.org/minimizationserver.htm). Active binding sites were identified based on co-crystallized ligand coordinates.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDe Novo Peptide Design\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNovel peptide sequences were generated using ChemGPT (https://huggingface.co/ncfrey/ChemGPT-1.2B), an open-source transformer model for molecular design. Design constraints specified peptides of 20-35 amino acids incorporating D-amino acids and N-methylated residues at protease-susceptible positions to enhance oral stability. Generated sequences were optimized through ESMFold webserver (https://esmatlas.com/resources) for three-dimensional structure prediction. Candidates with predicted local distance difference test (pLDDT) scores above 70 and alpha-helical content between 60-80% were selected for subsequent analysis.\u003cstrong\u003e\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMolecular Docking\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePeptide-receptor binding was evaluated using CB-Dock2 webserver (http://clab.labshare.cn/cb-dock2/), which automatically identifies binding sites and performs blind docking. The platform employs the AutoDock Vina algorithm with cavity detection through protein surface analysis. Each peptide was docked against all three target receptors, with binding affinity thresholds set at -8.0 kcal/mol. Candidates demonstrating balanced affinity across the three receptors (differences \u0026lt;3 kcal/mol) were prioritized. Interaction analysis utilized PLIP webserver (https://plip-tool.biotec.tu-dresden.de/) to identify hydrogen bonds and hydrophobic contacts.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMolecular Dynamics Simulations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConformational stability of peptide-receptor complexes was assessed through WebGro platform (https://simlab.uams.edu/), performing 100 nanosecond all-atom molecular dynamics simulations using GROMACS with AMBER99SB force field. Systems were solvated in TIP3P water boxes with 150 mM NaCl at 310 K and 1 bar pressure. Trajectory analysis included root mean square deviation (RMSD), root mean square fluctuation, and hydrogen bond monitoring. Complexes maintaining RMSD below 3 \u0026Aring; and hydrogen bond occupancy above 60% were considered stable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eADMET Property Prediction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePharmacokinetic properties were predicted using SwissADME (http://www.swissadme.ch) and pkCSM (http://biosig.lab.uq.edu.au/pkcsm) webservers. Evaluated parameters included human intestinal absorption, Caco-2 permeability, P-glycoprotein substrate liability, blood-brain barrier penetration, drug-likeness according to Lipinski\u0026apos;s rules, and BOILED-Egg model predictions. Toxicity endpoints encompassed AMES mutagenicity, hERG channel blockade, hepatotoxicity, and oral acute toxicity through ProTox-II platform (https://tox-new.charite.de/protox_II/).\u003c/p\u003e\n\u003cp\u003eLead candidates required: predicted intestinal absorption \u0026gt;85%, Caco-2 permeability \u0026gt;0.90 log Papp, negative P-glycoprotein substrate status, bioavailability score \u0026ge;0.55, AMES negative, hERG IC50 \u0026gt;10 \u0026mu;M, non-hepatotoxic classification, and oral rat acute toxicity LD50 \u0026gt;2000 mg/kg.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Visualization and Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMolecular structures were visualized using PyMOL 2.5 and protein-ligand interactions mapped through LigPlot\u003csup\u003e+\u003c/sup\u003e webserver (https://www.ebi.ac.uk/thornton-srv/software/LigPlus/). Binding affinity comparisons employed one-way ANOVA with Tukey\u0026apos;s post-hoc test (\u0026alpha;=0.05). Correlations between structural descriptors and pharmacokinetic parameters were assessed using Spearman\u0026apos;s correlation coefficient.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Considerations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study utilized exclusively \u003cem\u003ein silico\u003c/em\u003e methodologies based on publicly available crystallographic data and computational bioinformatics tools. No human subjects, animal models, biological samples, or experimental procedures were involved. Consequently, ethical approval from an institutional review board or animal ethics committee was not required for this research, in accordance with current guidelines for purely computational studies.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003e\u003cstrong\u003eReceptor Structure Preparation and Active Site Characterization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCrystal structures of GLP-1 receptor (PDB: 6X18), GIP receptor (PDB: 7DTY), and GCGR (PDB: 5EE7) were successfully retrieved and prepared. Energy minimization converged with final potential energies of -124,847 kJ/mol (GLP-1R), -118,562 kJ/mol (GIPR), and -121,394 kJ/mol (GCGR). Active site analysis identified orthosteric binding pockets at the extracellular-transmembrane interface, with critical residues including Arg121, Glu127, Glu128, Lys197, Glu364, and Arg380 for GLP-1R; Arg183, Glu247, Asp254 for GIPR; and Glu127, Lys187, Asp364 for GCGR (Figure 1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDe Novo Peptide Generation and Structure Prediction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eChemGPT generated 150 candidate peptide sequences (22-34 amino acids, mean 27.3 \u0026plusmn; 3.2) incorporating an average of 4.6 \u0026plusmn; 1.8 D-amino acids and strategic N-methylation at protease-susceptible sites. ESMFold structure prediction identified 38 candidates (25.3%) with predicted local distance difference test scores above 70, indicating high structural confidence. These candidates exhibited alpha-helical content between 60-80% (mean 68.4 \u0026plusmn; 6.2%) with amphipathic helix formation. Ramachandran analysis confirmed that 96.4 \u0026plusmn; 2.1% of residues were in allowed regions, validating structural quality for docking studies (Figure 2).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMolecular Docking Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCB-Dock2 evaluation revealed seven candidates demonstrating balanced affinity across all three receptors (binding energy differences \u0026lt;3.0 kcal/mol). The lead candidate TA-071 exhibited binding affinities of -10.8 kcal/mol (GLP-1R), -10.2 kcal/mol (GIPR), and -10.6 kcal/mol (GCGR), indicating equipotent multi-receptor engagement. TA-071 sequence: H-Tyr-dAla-Glu-Gly-Thr-Phe(Me)-Ile-Ser-Asp-dVal-Ser-dSer-Tyr-Leu-Glu-Gly-Gln-Ala-Ala-Lys-Glu-Phe-Ile-dAla-Trp-Leu-Val-Arg-NH₂ (28 residues).\u003c/p\u003e\n\u003cp\u003ePLIP analysis revealed extensive hydrogen bonding: eight bonds with GLP-1R (Arg121, Glu127, Glu128, Tyr152, Glu364, Arg380), nine with GIPR, and seven with GCGR. Off-target selectivity was confirmed with significantly weaker binding to peptide YY receptor 1 (-5.3 \u0026plusmn; 0.8 kcal/mol) and corticotropin-releasing factor receptor 1 (-4.9 \u0026plusmn; 0.9 kcal/mol), both p\u0026lt;0.001 versus target receptors (Figure 3).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMolecular Dynamics\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWebGro simulations spanning 100 nanoseconds confirmed stable peptide-receptor complexes. Root mean square deviation analysis showed equilibration after 16-22 nanoseconds with maintained RMSD values of 2.1 \u0026plusmn; 0.3 \u0026Aring; (GLP-1R), 2.4 \u0026plusmn; 0.4 \u0026Aring; (GIPR), and 1.9 \u0026plusmn; 0.2 \u0026Aring; (GCGR), all below the 3.0 \u0026Aring; threshold. Root mean square fluctuation revealed minimal core fluctuation (0.8-1.4 \u0026Aring; for residues 8-23) with flexibility concentrated at termini (3.2-4.8 \u0026Aring;).\u003c/p\u003e\n\u003cp\u003eHydrogen bond occupancy exceeded 60% for key interactions, with GLP-1R maintaining average 6.8 \u0026plusmn; 1.2 persistent bonds, including Glu3-Arg121 (89% occupancy), Asp9-Tyr152 (76%), and Glu15-Arg380 (82%). Molecular mechanics Poisson-Boltzmann surface area calculations estimated binding free energies of -62.4 \u0026plusmn; 8.7 kcal/mol (GLP-1R), -58.9 \u0026plusmn; 9.3 kcal/mol (GIPR), and -61.7 \u0026plusmn; 8.1 kcal/mol (GCGR), confirming thermodynamically favorable interactions (Figure 4).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eADMET Property Predictions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSwissADME analysis revealed favorable drug-likeness despite its peptidic nature (molecular weight 3,247 Da, TPSA 1,186 Ų). Lipophilicity consensus yielded log P of -2.8 \u0026plusmn; 1.2. P-glycoprotein substrate prediction was negative (probability 0.23), indicating low efflux liability.\u003c/p\u003e\n\u003cp\u003epkCSM predictions demonstrated promising oral bioavailability characteristics: human intestinal absorption 87.3% (threshold \u0026gt;85%), Caco-2 permeability 1.08 log Papp (threshold \u0026gt;0.90), bioavailability score 0.56 (threshold \u0026ge;0.55). Predicted volume of distribution was 0.68 L/kg, plasma protein binding 72.4%, clearance 8.3 mL/min/kg, and half-life 4.2 hours. Cytochrome P450 interaction profiling predicted neither substrate nor inhibitor activity for major isoforms, reducing drug-drug interaction risk (Figure 5).\u003c/p\u003e\n\u003cp\u003eProTox-II analysis indicated favorable safety predictions: AMES mutagenicity negative (probability 0.12), hERG IC₅₀ 24.6 \u0026mu;M (threshold \u0026gt;10 \u0026mu;M), hepatotoxicity negative (probability 0.31), oral rat acute toxicity LD₅₀ 2,400 mg/kg (toxicity class IV). All parameters met predefined acceptance criteria for lead candidate advancement.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eComparative Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTA-071 binding affinities compared favorably to approved therapeutics: exenatide (-9.2 kcal/mol), liraglutide (-9.7 kcal/mol), and tirzepatide (-10.1 kcal/mol GLP-1R, -11.3 kcal/mol GIPR). Notably, TA-071\u0026apos;s predicted intestinal absorption (87.3%) and Caco-2 permeability (1.08 log Papp) substantially exceeded subcutaneously administered comparators, which exhibit negligible oral bioavailability (\u0026lt;2%).\u003c/p\u003e\n\u003cp\u003eCorrelation analysis revealed D-amino acid number strongly correlated with predicted half-life (Spearman\u0026apos;s \u0026rho; = 0.78, p\u0026lt;0.001) and inversely with hepatic clearance (\u0026rho; = -0.71, p\u0026lt;0.001). N-methylation frequency correlated positively with Caco-2 permeability (\u0026rho; = 0.64, p\u0026lt;0.01). Alpha-helical content showed positive association with receptor binding affinity (\u0026rho; = 0.69-0.73, p\u0026lt;0.001 across all receptors) (Figure 6).\u003c/p\u003e\n\u003cp\u003eTA-071, a de novo designed triple agonist, demonstrated balanced and high-affinity binding to GLP-1R, GIPR, and GCGR. It exhibited exceptional predicted oral bioavailability, favorable ADMET properties, and superior permeability compared to subcutaneous reference therapeutics, establishing its promising preclinical candidacy.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eThe present study demonstrates that open-source generative AI can effectively facilitate the rational design of an orally bioavailable triple agonist targeting GLP-1, GIP, and glucagon receptors. Our results highlight TA-071\u0026rsquo;s superior multi-receptor affinity, structural stability, and predicted pharmacokinetic profile, surpassing current injectable therapeutics. This study underscores the transformative potential of AI in creating orally viable peptide therapeutics for complex metabolic diseases like DM2.\u003c/p\u003e\u003cp\u003eThe preparation of high-resolution receptor structures is an essential prerequisite for rational drug design, enabling precise characterization of orthosteric binding pockets.\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e Advances in crystallography and cryo-electron microscopy have enabled high-resolution receptor models, facilitating identification of orthosteric and allosteric sites essential for selective ligand interaction.\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e Subsequent active site mapping identifies key residues governing ligand recognition and binding affinity, a process that is fundamental for understanding molecular recognition events.\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e Advanced computational tools now facilitate the accurate definition of these pharmacologically relevant cavities, establishing a robust structural foundation for downstream molecular docking and virtual screening campaigns.\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e Aligning with the prerequisite for high-resolution structural characterization in rational drug design, we successfully optimized the receptor models to thermodynamic stability. Unlike generalized models, our targeted structures exhibit detailed active site residues at extracellular-transmembrane interfaces unique to GLP-1R, GIPR, and GCGR. This precision complements broader findings emphasizing conformational adaptations and receptor-ligand specificity, reinforcing a comprehensive structural foundation for rational drug design.\u003c/p\u003e\u003cp\u003eDe novo peptide generation and structure prediction have advanced through integrating AI and computational modeling, enabling the design of novel peptides with tailored functions and stability.\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e Techniques such as sequence-based language models and physics-based design improve peptide binding specificity and conformational accuracy.\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e Recent developments incorporate generative pre-trained transformers that facilitate high-throughput peptide generation with multimodal screening approaches, integrating sequence and structural information for enhanced bioactivity prediction.\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e These approaches accelerate rational peptide engineering, expanding therapeutic possibilities in drug development. The integration of AI-driven de novo peptide generation and structure prediction, exemplified by models like ChemGPT, allows for efficient design of peptides with tailored sequences and confirmed structural quality. Our results reflect these advancements, demonstrating amphipathic helices and high confidence in structural predictions, and ESMFold-based structure validation confirmed high-confidence predictions with predominant alpha-helical amphipathic architectures, substantiating the literature's premise that integrated sequence-structure approaches yield structurally viable candidates suitable for subsequent functional screening.\u003c/p\u003e\u003cp\u003eMolecular docking analysis is pivotal in predicting ligand binding conformations and affinities within receptor sites, employing advanced algorithms to enhance accuracy and efficiency.\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e Techniques integrate high-throughput virtual screening and scoring functions to identify promising drug candidates.\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e Machine learning-augmented scoring functions, particularly correction-term methodologies, substantially enhance traditional docking algorithms' predictive capabilities across diverse benchmark datasets.\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e This approach, combined with molecular dynamics, refines predictions under physiological conditions, expanding drug discovery capabilities.\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e Our molecular docking results exhibit balanced multi-receptor affinity and extensive hydrogen bonding, aligning with advanced docking algorithms such as CB-Dock2, which incorporates template-based cavity detection to enhance binding pose prediction accuracy. This demonstrates consistent affinity profiles across incretin receptor members, supporting the robustness of our ligand-receptor interaction predictions and confirming the predictive validity of current docking methodologies for rational peptide-receptor optimization.\u003c/p\u003e\u003cp\u003eMolecular dynamics simulations have revolutionized biomolecular research by capturing atomic-level protein behavior and dynamics with unprecedented temporal resolution.\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e Advances in force fields and enhanced sampling have improved accuracy, revealing mechanisms underpinning protein function and informing experimental strategies.\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e These computational methodologies facilitate structure-function elucidation, enhanced sampling techniques for conformational exploration, and accelerate structure-based drug design through mechanistic insights into protein-ligand interactions.\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e Our molecular dynamics simulations exemplify how advanced computational methodologies elucidate structure-function relationships through equilibration metrics, fluctuation patterns, and binding thermodynamics. The molecular dynamics simulations in our study provided in-depth, atomic-level data on the behavior of the evaluated proteins, expanding our understanding beyond the peptide-receptor stability results from WebGro.\u003c/p\u003e\u003cp\u003eADMET property predictions have undergone transformative advancement through machine learning integration, addressing critical bottlenecks in pharmaceutical development pipelines.\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e Contemporary methodologies leverage graph neural networks and ensemble frameworks to decipher complex structure-property relationships, substantially outperforming traditional quantitative structure-activity relationship models while providing scalable alternatives.\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e Advanced pretraining strategies incorporating quantum chemistry simulations enhance molecular representation learning, achieving state-of-the-art performance across multiple ADMET endpoints.\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e These computational innovations reduce late-stage attrition by enabling early risk assessment and compound prioritization, thereby expediting therapeutic development. Our SwissADME and pkCSM analyses, focused on drug-likeness, bioavailability, and toxicity profiles, confirm compound suitability, showcasing practical early-stage validation complementing theoretical advances. Employing consensus methodologies for lipophilicity, permeability, and toxicological endpoints, these analyses demonstrated favorable drug-likeness characteristics, minimal efflux liability, and acceptable safety profiles, thereby validating machine learning-driven risk stratification strategies that expedite lead optimization and preclinical decision-making processes.\u003c/p\u003e\u003cp\u003eComparative analysis between marketed drugs and de novo candidates constitutes a fundamental approach in pharmaceutical development.\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e Methodologies leverage existing therapeutic scaffolds through late-stage modifications, prodrug strategies, and repurposing approaches to accelerate candidate optimization.\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e Benchmarking studies reveal successful translation rates and inform structure-activity relationship refinements that distinguish first-in-class innovations from incremental improvements.\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e Such comparative frameworks facilitate evidence-based prioritization while maintaining therapeutic target validation rigor.\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e These strategies complement but differ fundamentally. The comparative analysis between marketed drugs and TA-071 reveals that while traditional therapeutics rely on established molecular scaffolds and modifications, TA-071 exhibits superior receptor binding affinities and enhanced oral bioavailability. Structural features such as D-amino acid incorporation and N-methylation correlate strongly with pharmacokinetic improvements, underscoring TA-071's potential as a multi-receptor agonist surpassing subcutaneous counterparts when analyzed from a computational point of view.\u003c/p\u003e\u003cp\u003eThis study demonstrates that open-source generative AI effectively enables rational design of orally bioavailable triple GLP-1, GIP, and glucagon receptor agonists. The lead candidate exhibits superior multi-receptor affinity, structural stability, and predicted pharmacokinetics compared to existing injectable therapeutics, highlighting AI\u0026rsquo;s transformative potential in peptide therapeutic development for DM2.\u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eOpen-source generative AI accelerates the design of novel multi-receptor agonists with enhanced oral bioavailability and drug-likeness, offering new avenues for innovative treatments with preclinical potential for metabolic disease management.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eConflict of Interest\u003c/h2\u003e\u003cp\u003eThe authors declare that they have no conflict of interest related to the publication of this manuscript.\u003c/p\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eFr\u0026iacute;as JP, Davies MJ, Rosenstock J, P\u0026eacute;rez Manghi FC, Fern\u0026aacute;ndez Land\u0026oacute; L, Bergman BK, et al. Tirzepatide versus Semaglutide Once Weekly in Patients with Type 2 Diabetes. N Engl J Med. 2021;385(6):503-515.\u003c/li\u003e\n \u003cli\u003eVenniyoor A. Tirzepatide Once Weekly for the Treatment of Obesity. N Engl J Med. 2022;387(15):1433-1434.\u003c/li\u003e\n \u003cli\u003eSeino Y, Fukushima M, Yabe D. GIP and GLP-1, the two incretin hormones: Similarities and differences. J Diabetes Investig. 2010;1(1-2):8-23.\u003c/li\u003e\n \u003cli\u003eBaggio LL, Drucker DJ. Biology of incretins: GLP-1 and GIP. Gastroenterology. 2007;132(6):2131-57.\u003c/li\u003e\n \u003cli\u003eChen H, Engkvist O, Wang Y, Olivecrona M, Blaschke T. The rise of deep learning in drug discovery. Drug Discov Today. 2018;23(6):1241-1250.\u003c/li\u003e\n \u003cli\u003eJumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596(7873):583-589.\u003c/li\u003e\n \u003cli\u003eHamman JH, Enslin GM, Kotz\u0026eacute; AF. Oral delivery of peptide drugs: barriers and developments. BioDrugs. 2005;19(3):165-77.\u003c/li\u003e\n \u003cli\u003eFinan B, Yang B, Ottaway N, Smiley DL, Ma T, Clemmensen C, et al. A rationally designed monomeric peptide triagonist corrects obesity and diabetes in rodents. Nat Med. 2015;21(1):27-36.\u003c/li\u003e\n \u003cli\u003eZou Y, Guo T, Fu Z, Guo Z, Bo W, Yan D, et al. A structure-based framework for selective inhibitor design and optimization. Commun Biol. 2025;8(1):422.\u003c/li\u003e\n \u003cli\u003eV\u0026eacute;nien-Bryan C, Li Z, Vuillard L, Boutin JA. Cryo-electron microscopy and X-ray crystallography: complementary approaches to structural biology and drug discovery. Acta Crystallogr F Struct Biol Commun. 2017;73(Pt 4):174-183.\u003c/li\u003e\n \u003cli\u003eDu X, Li Y, Xia YL, Ai SM, Liang J, Sang P, et al. Insights into Protein-Ligand Interactions: Mechanisms, Models, and Methods. Int J Mol Sci. 2016;17(2):144.\u003c/li\u003e\n \u003cli\u003eTavares FM, Gomes AC, Assun\u0026ccedil;\u0026atilde;o EM, de Medeiros JLS, Scotti MT, Scotti L, et al. Virtual Screening and Molecular Docking: Discovering Novel c-KIT Inhibitors. Curr Med Chem. 2022;29(2):166-188.\u003c/li\u003e\n \u003cli\u003eLamiable A, Th\u0026eacute;venet P, Rey J, Vavrusa M, Derreumaux P, Tuff\u0026eacute;ry P. PEP-FOLD3: faster de novo structure prediction for linear peptides in solution and in complex. Nucleic Acids Res. 2016;44(W1):W449-54.\u003c/li\u003e\n \u003cli\u003eZhang H, Saravanan KM, Wei Y, Jiao Y, Yang Y, Pan Y, et al. Deep Learning-Based Bioactive Therapeutic Peptide Generation and Screening. J Chem Inf Model. 2023;63(3):835-845.\u003c/li\u003e\n \u003cli\u003eZhao H, Song G. AVP-GPT2: A Transformer-Powered Platform for De Novo Generation, Screening, and Explanation of Antiviral Peptides. Viruses. 2024;17(1):14.\u003c/li\u003e\n \u003cli\u003eYang C, Chen EA, Zhang Y. Protein-Ligand Docking in the Machine-Learning Era. Molecules. 2022;27(14):4568.\u003c/li\u003e\n \u003cli\u003eCardoso MH, Orozco RQ, Rezende SB, Rodrigues G, Oshiro KGN, C\u0026acirc;ndido ES, et al. Computer-Aided Design of Antimicrobial Peptides: Are We Generating Effective Drug Candidates? Front Microbiol. 2020;10:3097.\u003c/li\u003e\n \u003cli\u003eZheng L, Meng J, Jiang K, Lan H, Wang Z, Lin M, et al. Improving protein-ligand docking and screening accuracies by incorporating a scoring function correction term. Brief Bioinform. 2022;23(3):bbac051.\u003c/li\u003e\n \u003cli\u003eWang Z, Sun H, Yao X, Li D, Xu L, Li Y, et al. Comprehensive evaluation of ten docking programs on a diverse set of protein-ligand complexes: the prediction accuracy of sampling power and scoring power. Phys Chem Chem Phys. 2016;18(18):12964-75.\u003c/li\u003e\n \u003cli\u003eFilipe HAL, Loura LMS. Molecular Dynamics Simulations: Advances and Applications. Molecules. 2022;27(7):2105.\u003c/li\u003e\n \u003cli\u003eLazim R, Suh D, Choi S. Advances in Molecular Dynamics Simulations and Enhanced Sampling Methods for the Study of Protein Systems. Int J Mol Sci. 2020;21(17):6339.\u003c/li\u003e\n \u003cli\u003eWu X, Xu LY, Li EM, Dong G. Application of molecular dynamics simulation in biomedicine. Chem Biol Drug Des. 2022;99(5):789-800.\u003c/li\u003e\n \u003cli\u003eVenkataraman M, Rao GC, Madavareddi JK, Maddi SR. Leveraging machine learning models in evaluating ADMET properties for drug discovery and development. ADMET DMPK. 2025;13(3):2772.\u003c/li\u003e\n \u003cli\u003eWang Y, Wang J, Yang Y, Ren Y, Bai H, Li H. From matrix factorization to graph neural networks: Advances in computational drug repositioning. Drug Discov Today. 2025;30(11):104499.\u003c/li\u003e\n \u003cli\u003eKim J, Chang W, Ji H, Joung I. Quantum-Informed Molecular Representation Learning Enhancing ADMET Property Prediction. J Chem Inf Model. 2024;64(13):5028-5040.\u003c/li\u003e\n \u003cli\u003eBar\u0026oacute; EL, Catti F, Estarellas C, Ghashghaei O, Lavilla R. Drugs from drugs: New chemical insights into a mature concept. Drug Discov Today. 2024;29(12):104212.\u003c/li\u003e\n \u003cli\u003eZheng L, Wang W, Sun Q. Targeted drug approvals in 2023: breakthroughs by the FDA and NMPA. Signal Transduct Target Ther. 2024;9(1):46.\u003c/li\u003e\n \u003cli\u003eSchuhmacher A, Hinder M, Brief E, Gassmann O, Hartl D. Benchmarking R\u0026amp;D success rates of leading pharmaceutical companies: an empirical analysis of FDA approvals (2006-2022). Drug Discov Today. 2025;30(2):104291.\u003c/li\u003e\n \u003cli\u003eOkuyama R. Chronological Analysis of First-in-Class Drugs Approved from 2011 to 2022: Their Technological Trend and Origin. Pharmaceutics. 2023;15(7):1794.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"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":"Generative AI, Triple agonist, Oral peptide therapeutics, Type 2 diabetes","lastPublishedDoi":"10.21203/rs.3.rs-8214757/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8214757/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eIntroduction: \u003c/strong\u003eType 2 diabetes mellitus (DM2) demands therapeutic strategies capable of surpassing the limitations of single-target incretin modulation. Triple agonism of GLP-1, GIP, and glucagon receptors (GCGR) offers synergistic metabolic benefits. Advances in open-source generative artificial intelligence (AI) now enable de novo molecular design with unprecedented precision, supporting the development of orally viable multi-receptor agonists.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eObjective\u003c/strong\u003e: To design and computationally evaluate a de novo peptide capable of balanced activation of GLP-1, GIP, and GCGR while exhibiting physicochemical and pharmacokinetic properties compatible with oral administration.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eA fully \u003cem\u003ein silico\u003c/em\u003e pipeline was established integrating ChemGPT-based generative peptide design, ESMFold structural prediction, CB-Dock2 docking, 100-ns molecular dynamics simulations through WebGro, and ADMET profiling using SwissADME, pkCSM, and ProTox-II. Lead candidates were prioritized based on structural stability, multi-receptor affinity, and predicted oral bioavailability.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eChemGPT generated 150 peptides, of which 38 met structural quality criteria and seven achieved balanced multi-receptor affinity. Lead candidate TA-071 displayed docking energies of 10.8 (GLP-1R), −10.2 (GIPR), and −10.6 kcal/mol (GCGR) and maintained stable binding throughout molecular dynamics simulations (RMSD \u0026lt;3 Å; persistent hydrogen-bond networks). ADMET predictions indicated high intestinal absorption (87.3%), favorable Caco-2 permeability, absence of major CYP interactions, and low predicted toxicity. Comparative analyses showed binding profiles and permeability surpassing existing injectable incretin mimetics.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eOpen-source generative AI enabled the rational creation of a structurally stable, orally oriented triple incretin/ GCGR agonist. TA-071 represents a promising \u003cem\u003ein silico \u003c/em\u003ecandidate for future preclinical development in DM2 therapy.\u003c/p\u003e","manuscriptTitle":"Open-Source Generative AI Enables De Novo Design of an Orally Bioavailable Triple GLP-1/GIP/Glucagon Receptor Agonist for Type 2 Diabetes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-01 07:29:26","doi":"10.21203/rs.3.rs-8214757/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":"ab6e42e9-7d0a-4b1a-b6b5-00f76d505028","owner":[],"postedDate":"December 1st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":58860257,"name":"Endocrinology \u0026 Metabolism"}],"tags":[],"updatedAt":"2025-12-01T07:29:26+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-01 07:29:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8214757","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8214757","identity":"rs-8214757","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}