Construction of a deep learning-based predictive model to evaluate the influence of mechanical stretching stimuli on MMP-2 gene expression levels in fibroblasts

Construction of a deep learning-based predictive model to evaluate the influence of mechanical stretching stimuli on MMP-2 gene expression levels in fibroblasts | 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 Construction of a deep learning-based predictive model to evaluate the influence of mechanical stretching stimuli on MMP-2 gene expression levels in fibroblasts Ruozu Xiao, Haowei Zhou, Zhen Shi, Rong Huang, Yuheng Zhang, Jing Li This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5869090/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 05 Jun, 2025 Read the published version in BioMedical Engineering OnLine → Version 1 posted 8 You are reading this latest preprint version Abstract Background: Matrix metalloproteinase-2 (MMP-2) secretion homeostasis, governed by the multifaceted interplay of skin stretching, is a pivotal determinant influencing wound healing dynamics. This investigation endeavors to devise an artificial intelligence (AI) prediction framework delineating the modulation of MMP-2 expression under stretching conditions, thereby unravelling profound insights into the mechanobiological orchestration of MMP-2secretion and fostering novel mechanotherapeutic strategies targeted at MMP-2 modulation. Methods: Employing a bespoke mechanical tensile loading apparatus, diverse mechanical tensile stimuli were administered to fibroblasts, with parameters such as tensile shape and frequency duration constituting the mechanical loading regimen. Furthermore, reverse transcription polymerase chain reaction (RT‒PCR) assays were conducted to measure MMP-2 gene expression levels in fibroblasts subjected to mechanical stretching. Subsequently, the resulting data were partitioned into training and validation cohorts at a 7:3 ratio, facilitating the development of the deep learning (DL) model via a back propagation neural network predicated on the training set. An external validation set was also curated by culling pertinent literature from the PubMed database to assess the predictive ability of the model. Results: Analysis of 336 data points related to MMP-2 gene expression via RT‒PCR corroborated the variability in MMP-2 gene expression levels in response to distinct mechanical stretching regimens. Consequently, a DL model was successfully crafted via the backpropagation algorithm to delineate the impact of mechanical stretching stimuli on MMP-2 gene expression levels. The model, characterized by an R²value of 0.73, evinced a commendable fit with the training dataset, elucidating the intricate interplay between the input features and the target variable. Notably, the model exhibited minimal prediction errors, as evidenced by a root mean square error (RMSE) of 0.42 and a mean absolute error (MAE) of 0.28. Furthermore, the model showcased robust generalization capabilities during validation, yielding R²values of 0.70 and 0.71 for the validation and external validation sets, respectively, revealing its predictive accuracy. Conclusions: The DL model fashioned through the backpropagation algorithm adeptly forecasts the impact of mechanical stretching stimuli on MMP-2 gene expression levels in fibroblasts with relative precision. These findings provide a foundation for the modulation of MMP homeostasis via mechanical stretching to expedite the healing of recalcitrant chronic refractory wound (CRW). Mechanical stretching Fibroblasts MMP-2 Deep learning Predictive model Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction A chronic refractory wound (CRW) denotes a cutaneous and subcutaneous injury stemming from various factors, thereby failing to heal completely through conventional repair mechanisms and displaying negligible signs of recovery. The increasing prevalence of CRWs aligns with the increasing aging population and the increasing incidence of geriatric and chronic ailments[ 1 ], highlighting CRW as a pressing global public health concern alongside cancer, cardiovascular diseases, diabetes, and obesity. In developed countries, the CRW incidence ranges from approximately 1.67 to 2.21 per 1,000 individuals, with treatment expenses constituting approximately 1–3% of overall healthcare expenditures [ 2 – 4 ]. Presently, the absence of efficacious CRW therapies primarily stems from intricate etiological factors and convoluted pathological mechanisms at play. Excessive inflammatory cascades and extracellular matrix (ECM) impairment are pathological hallmarks of delayed CRW healing[ 5 , 6 ]. Under CRW, ischemic and hypoxic conditions impede the synthesis of collagen types I, II, and III, concurrently inducing aberrant growth factor expression of growth factors and the overexpression of matrix metalloproteinases (MMPs). These factors collectively prevent the compromised ECM from furnishing the requisite mechanical bolstering and structural integrity [ 7 ]. Fibroblasts play a pivotal role in fostering skin wound recovery by secreting copious amounts of ECM, thereby catalyzing the proliferation and differentiation of adjacent tissue cells [ 8 ]. MMPs, vital zinc-dependent enzymes, orchestrate the degradation of the ECM across various stages, directly affecting wound healing, angiogenesis, and tissue remodelling [ 9 , 10 ]. The MMP family is classified into six groups predicated on substrate specificity and function: collagenases, gelatinases, stromelysins, matrilysins, membrane-type matrix metalloproteinases, and other unclassified MMPs [ 11 ]. Matrix metalloproteinase-2 (MMP-2), a gelatinase that primarily targets collagen and basement membrane components such as type IV collagen and gelatin, is a key enzyme involved in ECM degradation and remodelling and directly regulates ECM equilibrium dynamics[ 12 ]. Studies have shown that expression levels of MMP-2 are significantly elevated in chronic wounds[ 13 ]. While excessive MMP-2 expression can lead to ECM degradation, moderate levels of MMP-2 are beneficial, facilitating neovascularization and tissue remodeling[ 14 ]. This evidence indicates that MMP-2 participates plays a critical role in regulating ECM homeostasis, with its activity and expression levels closely associated with wound healing outcomes. Studies have underscored the remarkable responsiveness of MMP-2 in fibroblasts to various mechanical stimuli, including the magnitude, frequency, and duration of mechanical stretching applied to the cells [ 15 , 16 ], suggesting that the modulation of MMP-2 gene expression levels in fibroblasts via mechanical stretching is a viable avenue to bolster CRW repair endeavors. However, notable disparities exist in the ECM composition and mechanical attributes of CRWs stemming from divergent injury factors or distinct wound locations [ 17 , 18 ]. Moreover, ECM substitute materials with diverse compositions exert varying impacts on distinct CRW repair paradigms [ 19 – 21 ]. Mechanical stimulation concurrently influences wound healing and MMP secretion. However, the multiparametric complexity of in vivo mechanical stimuli and their coupled regulatory effects on MMPs remain poorly understood, hindering mechanistic exploration and therapeutic development. As a prototypical multiscale regulatory system, conventional experimental approaches face limitations including high development costs and extended experimental cycles. Recent advances in computational technologies have led to the emergence of artificial intelligence-driven high-throughput screening methods as transformative tools to overcome traditional efficiency bottlenecks[ 22 ]. Cutting-edge studies demonstrate that deep learning-based high-throughput algorithms—as feedforward neural network models—can be trained and optimized via backpropagation algorithms. These systems calculate discrepancies between predicted and experimental outcomes, propagate errors backward from output to hidden/input layers, and iteratively adjust synaptic weights. Additionally, deep learning models utilize gradient descent methods to minimize loss functions, progressively aligning predictions with empirical observations[ 23 ]. This framework holds significant promise for guiding the design of personalized ECM substitute materials through predictive model. To our knowledge, no studies have yet applied deep learning models to establish predictive frameworks for mechanical stimulation's effects on fibroblast-mediated MMP-2 secretion. On the basis of these findings, this study aims to develop precise models for prognosticating the required mechanical stretching parameters to modulate fibroblasts to elicit distinct MMP-2 gene expression levels, thereby providing a cornerstone for devising a meticulous methodology to regulate MMP-2 gene expression in fibroblasts through mechanical stretching stimuli. Results Impact of mechanical stretching parameters on MMP-2 gene expression by fibroblasts Different mechanical stretching conditions induce varying levels of MMP-2 gene expression in fibroblasts, suggesting diverse effects of mechanical forces on gene regulation. At a frequency of 0.05 Hz, MMP-2 gene expression peaked at 3 hours with a 12% stretching intensity but decreased at 24 hours with an 8% intensity (Fig. 1 A, B). Similarly, at 0.1 Hz, the MMP-2 levels increased at 6 hours with a 12% intensity but decreased at 12 hours with a 15% intensity (Fig. 1 C, D). At 0.15 Hz, gene expression increased at 12 hours with 8% intensity but decreased at 24 hours with 22% intensity (Fig. 1 E, F). Notably, at 0.2 Hz, MMP-2 expression increased at 3 hours with 22% intensity but decreased at 6 hours with 8% intensity (Fig. 1 G, H). Overall, MMP-2 expression in fibroblasts initially increased but then decreased with prolonged stretching time, intensity, and frequency, indicating that excessive mechanical stretching can decrease MMP-2 expression. Selecting appropriate mechanical parameters is crucial for regulating MMP-2 gene expression in fibroblasts. Development of the DL model A predictive model for the impact of mechanical stretching on MMP-2 gene expression levels was constructed via experimental data and was supported by a graphical user interface (GUI). This interface enables users to input various mechanical parameters (stretching frequency, intensity, and time) to predict MMP-2 gene expression in fibroblasts. Additionally, users can input MMP-2 gene levels to derive the requisite mechanical parameters. Validation of the model's predictive performance This study assessed the predictive accuracy of the model through scatter plots of actual versus predicted values (Fig. 2 A, B). The well-distributed data points away from the diagonal line indicate minimal prediction errors. The closeness of the data points along the diagonal line at a 45° angle suggests consistency between the predicted and actual values, underscoring the model's accuracy and robustness. Further analysis via line charts (Fig. 2 C, D) revealed consistent performance across different data points, with small deviations between the predicted and actual values. The neural network model achieved an R² value of 0.73 on the training set, indicating a strong fit and capturing the input feature‒target variable relationship. The root mean square error (RMSE) and mean absolute error (MAE) were 0.42 and 0.28, respectively, indicating low prediction errors. For the validation set, the model demonstrated good generalizability, maintaining stable predictive capability beyond the training data, with an R² value of 0.70. The train curve of the model further confirms these findings (Fig. S1 ). External validation of the DL model An external validation set sourced from the relevant literature was utilized to assess the model's generalizability. Scatter plots and line charts of actual versus predicted values (Fig. 3 ) indicated the model's efficacy in predicting the effects of mechanical stretching on MMP-2 gene expression in fibroblasts. The model exhibited an R² value of 0.71, with RMSE and MAE values aligning closely between the validation and external validation sets, emphasizing consistent prediction accuracy across datasets. Discussion The extracellular matrix (ECM) plays pivotal roles in tissue injury and repair processes. Disruption of its structure and function can impede crucial wound healing mechanisms, including proliferation, migration, angiogenesis, and epithelial regeneration, which are essential for chronic wound healing [ 24 , 25 ]. The normal skin ECM predominantly comprises proteins (collagen, glycoproteins, proteoglycans, elastin) and growth factors [ 26 ]. In chronic wounds, such as those in diabetic conditions, ECM components undergo aberrant alterations. Compared with normal skin, diabetic wounds exhibit reduced collagen deposition and elastin expression, with increased MMP-2 expression leading to a microenvironment of elevated protein hydrolysis, resulting in an uneven and rough dermis layer [ 27 ]. MMP-2 influences angiogenesis and fibroblast migration through degradation of ECM components (e.g., type IV collagen), thereby modulating wound healing progression[ 28 ]. Mechanical stimuli (e.g., tension or compression) alter integrin clustering states to activate Focal Adhesion Kinase (FAK) and RhoA/ROCK pathways, promoting YAP/TAZ translocation from the cytoplasm to the nucleus for transcriptional regulation of target genes that drive cell proliferation, migration, and ECM synthesis[ 29 ]. Overexpression of YAP/TAZ suppresses MMP-2 synthesis, whereas YAP/TAZ depletion upregulates MMP-2[ 30 ]. Studies indicate that mechanical stress modulates the extracellular matrix composition and structure by regulating the activity of matrix metalloproteinases, such as MMP-2, in fibroblasts [ 31 , 32 ]. Researchers have investigated the impact of mechanical stretching on MMP-2 gene expression in fibroblasts. Wang et al. utilized an equiaxial stretching device and confirmed that mechanical stretching can upregulate MMP-2 gene expression in fibroblasts under 12% stretching intensity and 48 hours of stretching [ 33 ]. Xie et al. reported that after a 12-hour mechanical stretching duration, 5% and 10% stretching intensities decreased MMP-2 gene expression, whereas 15% and 20% intensities increased its expression [ 15 ]. Despite these findings, few studies have explored the effects of specific mechanical stimulation conditions on MMP-2 gene expression in fibroblasts, indicating a gap in understanding the precise regulatory methods used. Hence, there is a pressing need to develop an AI model for predicting the impact of mechanical parameters on MMP-2 gene expression in fibroblasts. Cellular responses to mechanical stretching are indeed multidimensional. In our preliminary experiments, we observed that stretching intensity, frequency, and duration differentially influence cellular functions[ 34 ]. An analysis of various cell stretching experiments revealed that maintaining stretching intensities between 10% and 15% can sustain the physiological status of the cell and prevent apoptosis, facilitating the modulation of cellular function [ 35 – 37 ]. To enhance model predictive accuracy, we incorporated seven stretching intensities (0%, 5%, 8%, 10%, 12%, 15%, and 22%). Research has shown discrepancies in MMP-2 expression levels in fibroblasts at different durations of mechanical stretching [ 38 , 39 ]. Hence, we varied the duration of stretching to 3 hours, 6 hours, 12 hours, and 24 hours to evaluate the MMP-2 gene expression levels. A stretching frequency of approximately 0.1 Hz was chosen to mimic the normal dynamic physiological environment, which is crucial for studying cellular and tissue responses [ 40 , 41 ]. Orthogonal combination experiments were conducted on fibroblasts with diverse mechanical stretching parameters (intensity, duration, and frequency) to assess MMP-2 gene expression poststretching, revealing various responses on the basis of mechanical parameters. Fibroblasts exhibited a certain tolerance to MMP-2 gene expression stimulation by mechanical stretching, with prolonged stretching durations leading to decreased expression levels, which is consistent with prior research [ 15 , 16 , 33 ]. Notably, MMP-2 gene expression tended to initially increase but then decrease at different stretching intensities, suggesting that moderate stretching conditions may increase MMP-2 expression. Overall, a stretching frequency of approximately 0.1 Hz corresponded to lower MMP-2 gene expression levels, favouring ECM collagen deposition and tissue microenvironment stability. This study employed the error back propagation algorithm to construct a preliminary DL model for the impact of mechanical stretching on MMP-2 gene expression in fibroblasts and validated its predictive performance. The model aids in forecasting MMP-2 gene expression levels under various mechanical stretching conditions, guiding experimental design and reducing costs. Model predictions can be compared with experimental results to verify accuracy and optimize experimental conditions. Through model predictions and experimental validation, insights into how mechanical stretching regulates MMP-2 gene expression in fibroblasts can be gained, shedding light on biological mechanisms such as signal transduction and gene expression regulation and offering new perspectives for cell biology and biomechanics research. The model's application extends to tailoring ECM substitute materials for diverse chronic wound types, adjusting mechanical stretching parameters on the basis of input MMP-2 gene expression levels to regulate fibroblast ECM composition. This optimization can enhance the quality and performance of tissue engineering products. Additionally, the role of MMP-2 in diseases, such as cancer and inflammation, highlights the potential therapeutic implications of modulating mechanical stretching to influence MMP-2 expression in tumor cells. Understanding the interplay between mechanical forces and ECM remodelling through this model provides insights for disease diagnosis and treatment, facilitating personalized treatment plans. Therefore, this model represents a significant step in unravelling complex cell biomechanics and offers a valuable tool for investigating mechanical forces and ECM remodelling relationships. Study limitations include the absence of protein-level verification despite the establishment of a DL model for predicting MMP-2 gene expression levels in fibroblasts under mechanical stretching. This model was developed solely based on mRNA-level data, while MMP-2 protein activity may be influenced by post-translational modifications and feedback mechanisms. Future studies should incorporate Western blot or ELISA to validate protein expression, thereby further improving model applicability. This study forms part of a series investigating mechanical stretching-regulated fibroblast secretion of ECM components, aiming to establish methodological foundations and reliability for subsequent research. To enhance model applicability, future experiments will quantify additional ECM components (e.g., other MMPs, collagen levels). Future research should focus on protein-level validation and expand the model's applicability to refine the model for constructing personalized ECM substitute materials through mechanical stretching. Conclusions In this study, the DL model was leveraged to develop an initial predictive model for assessing the impact of mechanical stretching on MMP-2 gene expression levels in fibroblasts. The model demonstrated the ability to accurately predict MMP-2 gene expression levels in fibroblasts across diverse stretching conditions. This model is poised to serve as an invaluable instrument for scrutinizing the intricate relationship between mechanical forces and ECM remodelling, facilitating advancements in tissue engineering, regenerative medicine, and the treatment of various types of fibrosis and ECM-related disorders. Materials and methods Extraction and culture of fibroblasts In accordance with the approval of the Ethics Committee of the Second Affiliated Hospital of Air Force Medical University (GKJ-Y-202303-082), prepuce samples were obtained from healthy male children aged 8 years (from the Department of Urology at the Second Affiliated Hospital of Air Force Medical University). Fibroblasts were extracted from the samples via density gradient centrifugation and resuspended in cell dishes with Dulbecco's modified Eagle’s medium (DMEM, Procell, Wuhan China) supplemented with 0.584 g/L L-glutamine, 10% fetal bovine serum (FBS), 1% penicillin, and 1% streptomycin. The cells were placed in an incubator at 37℃ with 5% CO₂ and 95% air. When the cell density reached greater than 90%, the cells were digested with 0.25% trypsin and passaged at a ratio of 1:2 or 1:3, with logarithmic phase cells being used for experiments. Mechanical stretching of fibroblasts Mechanical stretching of cells is performed via a spherical automatic cell-stretching device, which consists mainly of a mechanical stretching loading machine and a control system [ 42 ]. The equipment requires a 6-well Flexcell Bioflex® culture plate (TTCF 5001C, Flexcell® International Corporation, USA) as a cell carrier for mechanical stretching loading. The mechanical stretching loading machine comprises six spherical columns, corresponding to each well of the 6-well Flexcell Bioflex® culture plate, with each column capable of vertical displacement within a range of 4–8 mm. Mechanical stretching is achieved by deforming the bottoms of the six wells of the culture plate and establishing six different mechanical stretching parameters. Fibroblasts were seeded onto the culture plate at a density of 1×10⁵/cm². The next day, after the fibroblasts grew to an infiltrative state, the culture medium and nonadherent cells were washed away, and 3 mL of DMEM containing 10% fetal bovine serum was added. The control system was subsequently used to apply mechanical stretching to fibroblasts at different stretching intensities (5%, 8%, 10%, 12%, 15%, and 22%), durations (3 h, 6 h, 12 h, and 24 h), and frequencies (0.05 Hz, 0.1 Hz, 0.15 Hz, and 0.2 Hz). mRNA extraction After mechanical stretching treatment, total RNA was extracted from the cells according to the manufacturer's instructions for the RNeasy Plus Mini Kit (TIANGEN; Beijing China). This kit simplifies the process of RNA extraction from cells via a rapid spin column method. Initially, the cells were lysed under effective denaturing conditions to rapidly inactivate RNases. Subsequently, genomic DNA removal columns were used to homogenize the samples, and the homogenate was added to the RNase-free adsorption column CR4. In the CR4 column, total RNA binds to the membrane. After a series of wash steps, impurities are removed, ultimately producing purified RNA, which can be eluted in 30–100 µL of RNase-free ddH 2 O. The purified RNA was used for subsequent experiments. cDNA synthesis cDNA synthesis was performed via the M5 Super qPCR RT Kit with gDNA Remover (Mei5 Biotechnology, Beijing China). A mixture of 5 µL of mRNA, 5 µL of 5x M5 RT Super Mix, and 10 µL of DEPC-ddH₂O was prepared, incubated at 42℃ for 15 minutes, and then heated at 96℃ for 5 minutes to inactivate the enzyme, ultimately producing cDNA. Real-time PCR RT‒PCR analysis of MMP-2 and GAPDH was performed on 112 sample groups, with three replicates for each group. PCR amplification was conducted in a 20 µL reaction system, which included 0.8 µL of upstream and downstream primers, 2 µL of cDNA, 10 µL of Taq Pro Universal SYBR qPCR Master Mix (Vazyme, China), and 7.2 µL of RNase-Free ddH₂O. Quantitative analysis was performed via a Bio CFX real-time PCR instrument (Bio-Rad, USA). The sequences and specifications of the PCR primer sets (5’-3’ direction) are shown in Table 1 . The mixture was initially preincubated at 95℃ for 3 minutes to activate the polymerase. The mixture was initially preincubated at 95℃ for 3 minutes to activate the polymerase. The amplification process included 40 cycles, with each cycle consisting of denaturation at 95℃ for 10 seconds, annealing at 60℃ for 30 seconds, and extension at 65℃ for 10 seconds. Relative expression levels were calculated via the 2 −ΔΔCT method (CT = fluorescence threshold; ΔCT = target CT - reference gene (GAPDH) CT; ΔΔCT = sample ΔCT - calibrator sample ΔCT). Table 1 Primers used in qPCR for mRNAs in human Name Sequence Size(bp) MMP-2 F: 5′-CCTACACCAAGAACTTCCGTCTG-3′ 23 R: 5′-GTGCCAAGGTCAATGTCAGGAG-3′ 22 GAPDH F: 5′-ACACCCACTCCTCCACCTTTG-3′ 21 R: 5′-TCCACCACCCTGTTGCTGTAG-3′ 21 Establishing a DL model We performed min-max normalization on the experimental data of mechanical stretching effects on MMP-2 gene expression to standardize variables within the [0, 1] range. The experimental dataset was partitioned into training and validation sets in a 7:3 ratio. The DL model employs the error backpropagation algorithm, learning by comparing the actual values and predicted values of each sample. For each training sample, the weights are adjusted to minimize the mean squared error between the network's predicted values and the actual target values. The DL model’s process was implemented in Python, with the backpropagation neural network architecture constructed using the Python-based TensorFlow 2.8 package. For this model, a grid search was implemented to evaluate 2–4 hidden layer configurations, with the 3-layer architecture (128–64–32 neurons) ultimately selected due to its optimal performance on the validation set (Fig. 4A). ReLU activation functions were employed to capture complex nonlinear relationships in input features through higher dimensionality in the first layer, maintain deep feature extraction capacity in the second layer, and further compress and refine features in the third layer. The output layer employs a linear activation for regression prediction. Additionally, Dropout layers (rate = 0.5) were incorporated after each hidden layer to randomly deactivate neurons, effectively mitigating overfitting while enhancing generalizability. During the training process, the Adam optimizer was selected with an initial learning rate set to 0.001 to leverage its adaptive learning rate advantages, accelerating convergence and improving optimization efficiency. The loss function was set to the mean squared error (MSE), which is suitable for minimizing targets in regression problems. The model training used a fixed random seed to ensure experimental reproducibility, optimizing the model parameters through 1000 epochs and a batch size of 128. To achieve interactivity, we established a GUI via Python 3.9.18 (Fig. 4B). Establishment of the external validation set This study used the PubMed database as the information source and conducted a search using the query “((Fibroblasts[Title]) OR (Fibroblast[Title])) AND ((((Mechanical[Title]) OR (mechanical stretch[Title])) OR (mechanical stretching[Title])) OR (dynamics[Title])) OR (stretch[Title])) OR (stretching[Title]))”, covering the time range from the establishment of the database to December 1, 2024, initially yielding 670 SCI papers. The inclusion criteria were as follows: (1) the research focused on the effects of mechanical stretching on fibroblasts and (2) the experimental content included MMP-2 gene expression levels. The exclusion criteria were as follows: (1) missing information in the articles and (2) data that could not be extracted from the literature. Ultimately, 11 data points were extracted from the statistical graphs in the literature via Origin 2022 software to establish the external validation set. Abbreviations MMP-2 matrix metalloproteinase-2 ECM extracellular matrix CRW chronic refractory wound MMPs Matrix metalloproteinases AI artificial intelligence DL deep learning GUI graphical user interface RMSE root mean square error MAE mean absolute error MSE mean square error. Declarations Acknowledgements The authors thank the Bioinspired Engineering and Biomechanics Centre（BEBC）, Xi’an Jiaotong University, for excellent technical support, collaboration, and provision of facilities for this work. We also thank Feng Xu and Yuanbo Jia for their expert technical assistance and skillful work. Author contributions RX: Project administration, writing - original draft. HZ: Data curation, validation. ZS: Conceptualization, writing - review & editing. YZ: Methodology, data curation. RH: Critical revision of the article, funding acquisition. JL: Project administration, Funding acquisition. All authors reviewed the manuscript. Funding This work was supported by the Key Industrial Innovation Chain Project of Shaanxi Provincial Key Research and Development Plan (No. 2022ZDLSF04-03) and the Key Project for Tackling Key Core Technology of Shaanxi (No. 2024SF-GJHX-20). Data availability No datasets were generated or analysed during the current study. Ethics approval and consent to participate Ethical approval for this study was obtained from the Ethics Committee of the Second Affiliated Hospital of Air Force Medical University (GKJ-Y-202303-082). All procedures conducted in studies involving human participants adhered to the ethical standards of the institutional and/or national research committees, as well as the 1964 Helsinki Declaration and its subsequent amendments or equivalent ethical guidelines. Competing interests The authors declare that they have no competing financial or nonfinancial interests related to this work. References Jiang Y, Huang S, Fu X, Liu H, Chen H. Epidemiology of chronic cutaneous wounds in China . Wound Repair Regen. 2015, 19(2):181-188. Rice JB, Desai U, Cummings AKG, Birnbaum HG, Skornicki M, Parsons NB. Burden of diabetic foot ulcers for medicare and private insurers . Diabetes Care. 2014, 37(3):651-658. Laura, Martinengo, Maja, Olsson, Ram, Bajpai, Michael, Soljak, Zee, Upton. Prevalence of chronic wounds in the general population: systematic review and meta-analysis of observational studies . Ann Epidemiol. 2018. Olsson M, Jarbrink K, Divakar U, Bajpai R, Car J. The humanistic and economic burden of chronic wounds: A systematic review . Wound Repair Regen. 2019, 27(1):114-125. Davis FM, Kimball A, Boniakowski A, Gallagher K. Dysfunctional Wound Healing in Diabetic Foot Ulcers: New Crossroads . Curr Diab Rep. 2018, 18(1):2. Bjarnsholt T, Kirketerp-Møller K, Jensen PØ, Madsen KG, Givskov M. Why chronic wounds will not heal: a novel hypothesis . Wound Repair Regen. 2010, 16(1):2-10. Barrientos S, Brem H, Stojadinovic O, Tomic-Canic M. Clinical application of growth factors and cytokines in wound healing . Wound Repair Regen. 2015, 22(5):569-578. Guo Y, Bian Z, Xu Q, Wen X, Kang J, Lin S, Wang X, Mi Z, Cui J, Zhang Z. Novel tissue-engineered skin equivalent from recombinant human collagen hydrogel and fibroblasts facilitated full-thickness skin defect repair in a mouse model . Mater Sci Eng C Mater Biol Appl. 2021, 130:112469. Pientaweeratch S, Panapisal V, Tansirikongkol A. Antioxidant, anti-collagenase and anti-elastase activities of Phyllanthus emblica , Manilkara zapota and silymarin: an in vitro comparative study for anti-aging applications . Pharm Biol. 2016, 24(7-9):1-8. Manosroi A, Kumguan K, Chankhampan C, Manosroi W, Manosroi J. Nanoscale gelatinase A (MMP-2) inhibition on human skin fibroblasts of Longkong (Lansium domesticum Correa) leaf extracts for anti-aging . J Nanosci Nanotechnol. 2012, 12(9):7187-7197. Sreesada P, Vandana, Krishnan B, Amrutha R, Chavan Y, Alfia H, Jyothis A, Venugopal P, Aradhya R, Suravajhala P. Matrix metalloproteinases: Master regulators of tissue morphogenesis . Gene. 2025, 933. Maybee DV, Ink NL, Ali MA. Novel roles of MT1-MMP and MMP-2: beyond the extracellular milieu . Int J Mol Sci. 2022, 23(17):9513. Przekora A. A Concise Review on Tissue Engineered Artificial Skin Grafts for Chronic Wound Treatment: Can We Reconstruct Functional Skin Tissue In Vitro? Cells. 2020, 9(7). Chen Y, Li C, Xie H, Fan Y, Yang Z, Ma J, He D, Li L. Infiltrating mast cells promote renal cell carcinoma angiogenesis by modulating PI3K→︀AKT→︀GSK3β→︀AM signaling . Oncogene. 2017, 36(20):2879-2888. Xie Y, Ouyang X, Wang G. Mechanical strain affects collagen metabolism-related gene expression in scleral fibroblasts . Biomed Pharmacother. 2020, 126:110095. Liu S, Lin Z. Vascular smooth muscle cells mechanosensitive regulators and vascular remodeling . J Vasc Res. 2022, 59(2):90-113. Maione AG, Brudno Y, Stojadinovic O, Park LK, Smith A, Tellechea A, Leal EC, Kearney CJ, Veves A, Tomic-Canic M. Three-dimensional human tissue models that incorporate diabetic foot ulcer-derived fibroblasts mimic in vivo features of chronic wounds . Tissue Eng Part C Methods. 2015, 21(5):499-508. Hodde JP, Johnson CE. Extracellular matrix as a strategy for treating chronic wounds . Am J Clin Dermatol. 2007, 8:61-66. Zhao H, Li Z, Wang Y, Zhou K, Li H, Bi S, Wang Y, Wu W, Huang Y, Peng B. Bioengineered MSC-derived exosomes in skin wound repair and regeneration . Front Cell Dev Biol. 2023, 11:1029671. Kerstan A, Dieter K, Niebergall-Roth E, Klingele S, Jünger M, Hasslacher C, Daeschlein G, Stemler L, Meyer-Pannwitt U, Schubert K. Translational development of ABCB5+ dermal mesenchymal stem cells for therapeutic induction of angiogenesis in non-healing diabetic foot ulcers . Stem Cell Res Ther. 2022, 13(1):455. Peng Y, Meng H, Li P, Jiang Y, Fu X. Research advances of stem cell-based tissue engineering repair materials in promoting the healing of chronic refractory wounds on the body surface . Zhonghua Shao Shang Yu Chuang Mian Xiu Fu Za Zhi. 2023, 39(3):290-295. Wang H, Fu T, Du Y, Gao W, Huang K, Liu Z, Chandak P, Liu S, Van Katwyk P, Deac A. Scientific discovery in the age of artificial intelligence . Nature. 2023, 620(7972):47-60. Kosarac A, Cep R, Trochta M, Knezev M, Zivkovic A, Mladjenovic C, Antic A. Thermal behavior modeling based on BP neural network in Keras framework for motorized machine tool spindles . Materials (Basel). 2022, 15(21):7782. Chen S, Mienaltowski MJ, Birk DE. Regulation of corneal stroma extracellular matrix assembly . Exp Eye Res. 2015, 133:69-80. Rodrigues M, Kosaric N, Bonham CA, Gurtner GC. Wound healing: a cellular perspective . Physiol Rev. 2018. Leng L, Ma J, Sun X, Guo B, Li F, Zhang W, Chang M, Diao J, Wang Y, Wang W. Comprehensive proteomic atlas of skin biomatrix scaffolds reveals a supportive microenvironment for epidermal development . J Tissue Eng. 2020, 11:2041731420972310. Huang Y, Kyriakides TR. The role of extracellular matrix in the pathophysiology of diabetic wounds . Matrix Biol Plus. 2020, 6:100037. Lim WJ, Chan PF, Abd Hamid R. A 1, 4-benzoquinone derivative isolated from Ardisia crispa (Thunb.) A. DC. root suppresses angiogenesis via its angiogenic signaling cascades . Saudi Pharm J. 2024, 32(1):101891. Do CTP, Prochnau JY, Dominguez A, Wang P, Rao MK. The Road Ahead in Pancreatic Cancer: Emerging Trends and Therapeutic Prospects . Biomedicines. 2024, 12(9). Jones DL, Hallström GF, Jiang X, Locke RC, Evans MK, Bonnevie ED, Srikumar A, Leahy TP, Nijsure MP, Boerckel JD, et al. Mechanoepigenetic regulation of extracellular matrix homeostasis via Yap and Taz . Proc Natl Acad Sci U S A 2023, 120(22):e2211947120. Yeganegi A, Whitehead K, de Castro Brás LE, Richardson WJ. Mechanical strain modulates extracellular matrix degradation and byproducts in an isoform-specific manner . Connect Tissue Res. 2023, 1867(3):130286. Pakpahan ND, Kyawsoewin M, Manokawinchoke J, Termkwancharoen C, Egusa H, Limraksasin P, Osathanon T. Effects of mechanical loading on matrix homeostasis and differentiation potential of periodontal ligament cells: A scoping review . J Periodontal Res. 2024. Wang Y, Dang Z, Cui W, Yang L. Mechanical stretch and hypoxia inducible factor-1 alpha affect the vascular endothelial growth factor and the connective tissue growth factor in cultured ACL fibroblasts . Connect Tissue Res. 2017, 58(5):407-413. Dai W, Zhou H, Du J, Xiao R, Su J, Liu Z, Huang R, Li Y, Li J. Mechanical stretching enhances the cellular and paracrine effects of bone marrow mesenchymal stem cells on diabetic wound healing . Burns Trauma. 2025. Lin L-Q, Zeng H-K, Luo Y-L, Chen D-F, Ma X-Q, Chen H-J, Song X-Y, Wu H-K, Li S-Y. Mechanical stretch promotes apoptosis and impedes ciliogenesis of primary human airway basal stem cells . Respir Res. 2023, 24(1):237. Shan S, Fang B, Zhang Y, Wang C, Zhou J, Niu C, Gao Y, Zhao D, He J, Wang J. Mechanical stretch promotes tumoricidal M1 polarization via the FAK/NF‐κB signaling pathway . FASEB J. 2019, 33(12):13254-13266. Ma H, Wang L, Sun H, Yu Q, Yang T, Wang Y, Niu B, Jia Y, Liu Y, Liang Z. MIR-107/HMGB1/FGF-2 axis responds to excessive mechanical stretch to promote rapid repair of vascular endothelial cells . Arch Biochem Biophys. 2023, 744:109686. Liu C, Feng P, Li X, Song J, Chen W. Expression of MMP-2, MT1-MMP, and TIMP-2 by cultured rabbit corneal fibroblasts under mechanical stretch . Exp Biol Med (Maywood). 2014, 239(8):907-912. Shelton L, Rada JS. Effects of cyclic mechanical stretch on extracellular matrix synthesis by human scleral fibroblasts . Exp Eye Res. 2007, 84(2):314-322. Jacho D, Rabino A, Garcia-Mata R, Yildirim-Ayan E. Mechanoresponsive regulation of fibroblast-to-myofibroblast transition in three-dimensional tissue analogues: Mechanical strain amplitude dependency of fibrosis . Sci Rep. 2022, 12(1):16832. Katzengold R, Orlov A, Gefen A. A novel system for dynamic stretching of cell cultures reveals the mechanobiology for delivering better negative pressure wound therapy . Biomech Model Mechanobiol. 2021, 20:193-204. Huang R, Xu L, Wang Y, Zhang Y, Lin B, Lin Z, Li J, Li X. Efficient fabrication of stretching hydrogels with programmable strain gradients as cell sheet delivery vehicles . Mater Sci Eng C Mater Biol Appl. 2021, 129:112415. Additional Declarations No competing interests reported. Supplementary Files FigS1.tif Cite Share Download PDF Status: Published Journal Publication published 05 Jun, 2025 Read the published version in BioMedical Engineering OnLine → Version 1 posted Editorial decision: Accepted 22 May, 2025 Reviews received at journal 18 May, 2025 Reviews received at journal 07 May, 2025 Reviewers agreed at journal 05 May, 2025 Reviewers agreed at journal 30 Apr, 2025 Reviewers invited by journal 30 Apr, 2025 Submission checks completed at journal 30 Apr, 2025 First submitted to journal 28 Apr, 2025 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-5869090","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":450178494,"identity":"44d06186-8b15-42fc-9240-6620a20494f6","order_by":0,"name":"Ruozu Xiao","email":"","orcid":"","institution":"Tangdu Hospital, The Fourth Military Medical University","correspondingAuthor":false,"prefix":"","firstName":"Ruozu","middleName":"","lastName":"Xiao","suffix":""},{"id":450178497,"identity":"028a7ddd-92e6-4cd6-b613-d4610475e177","order_by":1,"name":"Haowei Zhou","email":"","orcid":"","institution":"Tangdu Hospital, The Fourth Military Medical University","correspondingAuthor":false,"prefix":"","firstName":"Haowei","middleName":"","lastName":"Zhou","suffix":""},{"id":450178499,"identity":"962181d7-f0ca-4746-aa41-b66455fd6102","order_by":2,"name":"Zhen Shi","email":"","orcid":"","institution":"Tangdu Hospital, The Fourth Military Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zhen","middleName":"","lastName":"Shi","suffix":""},{"id":450178500,"identity":"e96e3c80-6cb3-454f-a748-f2ca176ea0a1","order_by":3,"name":"Rong Huang","email":"","orcid":"","institution":"Tangdu Hospital, The Fourth Military Medical University","correspondingAuthor":false,"prefix":"","firstName":"Rong","middleName":"","lastName":"Huang","suffix":""},{"id":450178502,"identity":"f2a3205d-0211-40f0-8a1b-3def2326a84a","order_by":4,"name":"Yuheng Zhang","email":"","orcid":"","institution":"Western Theater Air Force Hospital of PLA","correspondingAuthor":false,"prefix":"","firstName":"Yuheng","middleName":"","lastName":"Zhang","suffix":""},{"id":450178503,"identity":"f16784c0-4805-4c30-8192-5c7b2e1bee35","order_by":5,"name":"Jing Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1UlEQVRIiWNgGAWjYBACPmaGhAMMDBJy/MwHwAKMDYS0sEG02BhLtiUQqwVCpSVuOEa0FnaGh4cLfh1m3HyMO3UzD4ON7IYDzM8eEHLY4Zl9h5nNjvFuu83DkGa84QCbuQFBLbw9h9nM7veCtBxO3HCAh02CGC08xm1gW/4TqYXnR5qEARtYywFibWmwMZAA+uXmHINk45lAR+LVws9/Jvkzzx+J+n6gw268qbCT7Tve/AyvFgYGngQGxjYYBxRUzPjVAwH7AQaGPwRVjYJRMApGwUgGAOAqSMgGJh22AAAAAElFTkSuQmCC","orcid":"","institution":"Tangdu Hospital, The Fourth Military Medical University","correspondingAuthor":true,"prefix":"","firstName":"Jing","middleName":"","lastName":"Li","suffix":""}],"badges":[],"createdAt":"2025-01-21 00:38:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5869090/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5869090/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12938-025-01399-0","type":"published","date":"2025-06-05T15:57:48+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":81803098,"identity":"81fff757-4d0c-4f99-9e79-3584d1c1ffc6","added_by":"auto","created_at":"2025-05-02 06:28:55","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":503246,"visible":true,"origin":"","legend":"\u003cp\u003eReal-time PCR quantitative analysis of MMP-2 in fibroblasts after mechanical stretching at different frequencies. \u003cstrong\u003eA, B\u003c/strong\u003e 0.05 Hz. \u003cstrong\u003eC, D\u003c/strong\u003e 0.1 Hz. \u003cstrong\u003eE, F\u003c/strong\u003e0.15 Hz. \u003cstrong\u003eG, H\u003c/strong\u003e 0.2 Hz\u003c/p\u003e","description":"","filename":"1.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5869090/v1/94cf5db2a19ddf42eb830eb1.jpg"},{"id":81803094,"identity":"fb80480e-822b-4e8a-b743-53cafb74a55f","added_by":"auto","created_at":"2025-05-02 06:28:54","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":586813,"visible":true,"origin":"","legend":"\u003cp\u003eScatter plot and line chart of actual values vs. predicted values for the DL model. \u003cstrong\u003eA\u003c/strong\u003e Scatter plot of actual values vs. predicted values for the training set. \u003cstrong\u003eB\u003c/strong\u003e Scatter plot of actual values vs. predicted values for the validation set. \u003cstrong\u003eC\u003c/strong\u003eLine chart of actual values vs. predicted values for the training set. \u003cstrong\u003eD\u003c/strong\u003eLine chart of actual values vs. predicted values for the validation set\u003c/p\u003e","description":"","filename":"2.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5869090/v1/a98df2c49dc4e44fe3454b2f.jpg"},{"id":81803095,"identity":"54ca592e-a7e9-4a8e-a561-53d4c57b3336","added_by":"auto","created_at":"2025-05-02 06:28:55","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":291474,"visible":true,"origin":"","legend":"\u003cp\u003eScatter plot and line chart of actual values vs. predicted values for the external validation set. \u003cstrong\u003eA\u003c/strong\u003eScatter plot of actual values vs. predicted values for the external validation set. \u003cstrong\u003eB\u003c/strong\u003e Line chart of actual values vs. predicted values for the external validation set\u003c/p\u003e","description":"","filename":"3.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5869090/v1/e873e4368a576abde06e3a0e.jpg"},{"id":81803096,"identity":"a1992ab5-85f7-466b-b598-38095543be7a","added_by":"auto","created_at":"2025-05-02 06:28:55","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2518226,"visible":true,"origin":"","legend":"\u003cp\u003eNeural network model architecture. A. Neural network schematic. B. Neural network construction workflow\u003c/p\u003e","description":"","filename":"4.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5869090/v1/e206727f2312845597978528.jpg"},{"id":84243180,"identity":"be1c7f14-733c-4997-8bfc-11b00aa008ba","added_by":"auto","created_at":"2025-06-09 16:12:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4606245,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5869090/v1/3a5f9363-51d7-49f2-bb6e-d169362c99f1.pdf"},{"id":81803101,"identity":"b4411ec8-fb6a-4d85-8bd9-a9fe36fc60ab","added_by":"auto","created_at":"2025-05-02 06:28:55","extension":"tif","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":6417520,"visible":true,"origin":"","legend":"","description":"","filename":"FigS1.tif","url":"https://assets-eu.researchsquare.com/files/rs-5869090/v1/e56b4889e58db70805270f74.tif"}],"financialInterests":"No competing interests reported.","formattedTitle":"Construction of a deep learning-based predictive model to evaluate the influence of mechanical stretching stimuli on MMP-2 gene expression levels in fibroblasts","fulltext":[{"header":"Introduction","content":"\u003cp\u003eA chronic refractory wound (CRW) denotes a cutaneous and subcutaneous injury stemming from various factors, thereby failing to heal completely through conventional repair mechanisms and displaying negligible signs of recovery. The increasing prevalence of CRWs aligns with the increasing aging population and the increasing incidence of geriatric and chronic ailments[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], highlighting CRW as a pressing global public health concern alongside cancer, cardiovascular diseases, diabetes, and obesity. In developed countries, the CRW incidence ranges from approximately 1.67 to 2.21 per 1,000 individuals, with treatment expenses constituting approximately 1\u0026ndash;3% of overall healthcare expenditures [\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Presently, the absence of efficacious CRW therapies primarily stems from intricate etiological factors and convoluted pathological mechanisms at play. Excessive inflammatory cascades and extracellular matrix (ECM) impairment are pathological hallmarks of delayed CRW healing[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eUnder CRW, ischemic and hypoxic conditions impede the synthesis of collagen types I, II, and III, concurrently inducing aberrant growth factor expression of growth factors and the overexpression of matrix metalloproteinases (MMPs). These factors collectively prevent the compromised ECM from furnishing the requisite mechanical bolstering and structural integrity [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Fibroblasts play a pivotal role in fostering skin wound recovery by secreting copious amounts of ECM, thereby catalyzing the proliferation and differentiation of adjacent tissue cells [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. MMPs, vital zinc-dependent enzymes, orchestrate the degradation of the ECM across various stages, directly affecting wound healing, angiogenesis, and tissue remodelling [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. The MMP family is classified into six groups predicated on substrate specificity and function: collagenases, gelatinases, stromelysins, matrilysins, membrane-type matrix metalloproteinases, and other unclassified MMPs [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Matrix metalloproteinase-2 (MMP-2), a gelatinase that primarily targets collagen and basement membrane components such as type IV collagen and gelatin, is a key enzyme involved in ECM degradation and remodelling and directly regulates ECM equilibrium dynamics[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Studies have shown that expression levels of MMP-2 are significantly elevated in chronic wounds[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. While excessive MMP-2 expression can lead to ECM degradation, moderate levels of MMP-2 are beneficial, facilitating neovascularization and tissue remodeling[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. This evidence indicates that MMP-2 participates plays a critical role in regulating ECM homeostasis, with its activity and expression levels closely associated with wound healing outcomes.\u003c/p\u003e \u003cp\u003eStudies have underscored the remarkable responsiveness of MMP-2 in fibroblasts to various mechanical stimuli, including the magnitude, frequency, and duration of mechanical stretching applied to the cells [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e], suggesting that the modulation of MMP-2 gene expression levels in fibroblasts via mechanical stretching is a viable avenue to bolster CRW repair endeavors. However, notable disparities exist in the ECM composition and mechanical attributes of CRWs stemming from divergent injury factors or distinct wound locations [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Moreover, ECM substitute materials with diverse compositions exert varying impacts on distinct CRW repair paradigms [\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Mechanical stimulation concurrently influences wound healing and MMP secretion. However, the multiparametric complexity of in vivo mechanical stimuli and their coupled regulatory effects on MMPs remain poorly understood, hindering mechanistic exploration and therapeutic development. As a prototypical multiscale regulatory system, conventional experimental approaches face limitations including high development costs and extended experimental cycles.\u003c/p\u003e \u003cp\u003eRecent advances in computational technologies have led to the emergence of artificial intelligence-driven high-throughput screening methods as transformative tools to overcome traditional efficiency bottlenecks[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Cutting-edge studies demonstrate that deep learning-based high-throughput algorithms\u0026mdash;as feedforward neural network models\u0026mdash;can be trained and optimized via backpropagation algorithms. These systems calculate discrepancies between predicted and experimental outcomes, propagate errors backward from output to hidden/input layers, and iteratively adjust synaptic weights. Additionally, deep learning models utilize gradient descent methods to minimize loss functions, progressively aligning predictions with empirical observations[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. This framework holds significant promise for guiding the design of personalized ECM substitute materials through predictive model. To our knowledge, no studies have yet applied deep learning models to establish predictive frameworks for mechanical stimulation's effects on fibroblast-mediated MMP-2 secretion. On the basis of these findings, this study aims to develop precise models for prognosticating the required mechanical stretching parameters to modulate fibroblasts to elicit distinct MMP-2 gene expression levels, thereby providing a cornerstone for devising a meticulous methodology to regulate MMP-2 gene expression in fibroblasts through mechanical stretching stimuli.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eImpact of mechanical stretching parameters on MMP-2 gene expression by fibroblasts\u003c/h2\u003e \u003cp\u003eDifferent mechanical stretching conditions induce varying levels of MMP-2 gene expression in fibroblasts, suggesting diverse effects of mechanical forces on gene regulation. At a frequency of 0.05 Hz, MMP-2 gene expression peaked at 3 hours with a 12% stretching intensity but decreased at 24 hours with an 8% intensity (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA, B). Similarly, at 0.1 Hz, the MMP-2 levels increased at 6 hours with a 12% intensity but decreased at 12 hours with a 15% intensity (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC, D). At 0.15 Hz, gene expression increased at 12 hours with 8% intensity but decreased at 24 hours with 22% intensity (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eE, F). Notably, at 0.2 Hz, MMP-2 expression increased at 3 hours with 22% intensity but decreased at 6 hours with 8% intensity (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eG, H). Overall, MMP-2 expression in fibroblasts initially increased but then decreased with prolonged stretching time, intensity, and frequency, indicating that excessive mechanical stretching can decrease MMP-2 expression. Selecting appropriate mechanical parameters is crucial for regulating MMP-2 gene expression in fibroblasts.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eDevelopment of the DL model\u003c/h3\u003e\n\u003cp\u003eA predictive model for the impact of mechanical stretching on MMP-2 gene expression levels was constructed via experimental data and was supported by a graphical user interface (GUI). This interface enables users to input various mechanical parameters (stretching frequency, intensity, and time) to predict MMP-2 gene expression in fibroblasts. Additionally, users can input MMP-2 gene levels to derive the requisite mechanical parameters.\u003c/p\u003e\n\u003ch3\u003eValidation of the model's predictive performance\u003c/h3\u003e\n\u003cp\u003eThis study assessed the predictive accuracy of the model through scatter plots of actual versus predicted values (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA, B). The well-distributed data points away from the diagonal line indicate minimal prediction errors. The closeness of the data points along the diagonal line at a 45\u0026deg; angle suggests consistency between the predicted and actual values, underscoring the model's accuracy and robustness. Further analysis via line charts (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC, D) revealed consistent performance across different data points, with small deviations between the predicted and actual values.\u003c/p\u003e \u003cp\u003eThe neural network model achieved an R\u0026sup2; value of 0.73 on the training set, indicating a strong fit and capturing the input feature‒target variable relationship. The root mean square error (RMSE) and mean absolute error (MAE) were 0.42 and 0.28, respectively, indicating low prediction errors. For the validation set, the model demonstrated good generalizability, maintaining stable predictive capability beyond the training data, with an R\u0026sup2; value of 0.70. The train curve of the model further confirms these findings (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eExternal validation of the DL model\u003c/h3\u003e\n\u003cp\u003eAn external validation set sourced from the relevant literature was utilized to assess the model's generalizability. Scatter plots and line charts of actual versus predicted values (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e) indicated the model's efficacy in predicting the effects of mechanical stretching on MMP-2 gene expression in fibroblasts. The model exhibited an R\u0026sup2; value of 0.71, with RMSE and MAE values aligning closely between the validation and external validation sets, emphasizing consistent prediction accuracy across datasets.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe extracellular matrix (ECM) plays pivotal roles in tissue injury and repair processes. Disruption of its structure and function can impede crucial wound healing mechanisms, including proliferation, migration, angiogenesis, and epithelial regeneration, which are essential for chronic wound healing [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. The normal skin ECM predominantly comprises proteins (collagen, glycoproteins, proteoglycans, elastin) and growth factors [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. In chronic wounds, such as those in diabetic conditions, ECM components undergo aberrant alterations. Compared with normal skin, diabetic wounds exhibit reduced collagen deposition and elastin expression, with increased MMP-2 expression leading to a microenvironment of elevated protein hydrolysis, resulting in an uneven and rough dermis layer [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. MMP-2 influences angiogenesis and fibroblast migration through degradation of ECM components (e.g., type IV collagen), thereby modulating wound healing progression[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Mechanical stimuli (e.g., tension or compression) alter integrin clustering states to activate Focal Adhesion Kinase (FAK) and RhoA/ROCK pathways, promoting YAP/TAZ translocation from the cytoplasm to the nucleus for transcriptional regulation of target genes that drive cell proliferation, migration, and ECM synthesis[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Overexpression of YAP/TAZ suppresses MMP-2 synthesis, whereas YAP/TAZ depletion upregulates MMP-2[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Studies indicate that mechanical stress modulates the extracellular matrix composition and structure by regulating the activity of matrix metalloproteinases, such as MMP-2, in fibroblasts [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eResearchers have investigated the impact of mechanical stretching on MMP-2 gene expression in fibroblasts. Wang et al. utilized an equiaxial stretching device and confirmed that mechanical stretching can upregulate MMP-2 gene expression in fibroblasts under 12% stretching intensity and 48 hours of stretching [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Xie et al. reported that after a 12-hour mechanical stretching duration, 5% and 10% stretching intensities decreased MMP-2 gene expression, whereas 15% and 20% intensities increased its expression [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Despite these findings, few studies have explored the effects of specific mechanical stimulation conditions on MMP-2 gene expression in fibroblasts, indicating a gap in understanding the precise regulatory methods used. Hence, there is a pressing need to develop an AI model for predicting the impact of mechanical parameters on MMP-2 gene expression in fibroblasts.\u003c/p\u003e \u003cp\u003eCellular responses to mechanical stretching are indeed multidimensional. In our preliminary experiments, we observed that stretching intensity, frequency, and duration differentially influence cellular functions[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. An analysis of various cell stretching experiments revealed that maintaining stretching intensities between 10% and 15% can sustain the physiological status of the cell and prevent apoptosis, facilitating the modulation of cellular function [\u003cspan additionalcitationids=\"CR36\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. To enhance model predictive accuracy, we incorporated seven stretching intensities (0%, 5%, 8%, 10%, 12%, 15%, and 22%). Research has shown discrepancies in MMP-2 expression levels in fibroblasts at different durations of mechanical stretching [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Hence, we varied the duration of stretching to 3 hours, 6 hours, 12 hours, and 24 hours to evaluate the MMP-2 gene expression levels. A stretching frequency of approximately 0.1 Hz was chosen to mimic the normal dynamic physiological environment, which is crucial for studying cellular and tissue responses [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOrthogonal combination experiments were conducted on fibroblasts with diverse mechanical stretching parameters (intensity, duration, and frequency) to assess MMP-2 gene expression poststretching, revealing various responses on the basis of mechanical parameters. Fibroblasts exhibited a certain tolerance to MMP-2 gene expression stimulation by mechanical stretching, with prolonged stretching durations leading to decreased expression levels, which is consistent with prior research [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Notably, MMP-2 gene expression tended to initially increase but then decrease at different stretching intensities, suggesting that moderate stretching conditions may increase MMP-2 expression. Overall, a stretching frequency of approximately 0.1 Hz corresponded to lower MMP-2 gene expression levels, favouring ECM collagen deposition and tissue microenvironment stability.\u003c/p\u003e \u003cp\u003eThis study employed the error back propagation algorithm to construct a preliminary DL model for the impact of mechanical stretching on MMP-2 gene expression in fibroblasts and validated its predictive performance. The model aids in forecasting MMP-2 gene expression levels under various mechanical stretching conditions, guiding experimental design and reducing costs. Model predictions can be compared with experimental results to verify accuracy and optimize experimental conditions. Through model predictions and experimental validation, insights into how mechanical stretching regulates MMP-2 gene expression in fibroblasts can be gained, shedding light on biological mechanisms such as signal transduction and gene expression regulation and offering new perspectives for cell biology and biomechanics research.\u003c/p\u003e \u003cp\u003eThe model's application extends to tailoring ECM substitute materials for diverse chronic wound types, adjusting mechanical stretching parameters on the basis of input MMP-2 gene expression levels to regulate fibroblast ECM composition. This optimization can enhance the quality and performance of tissue engineering products. Additionally, the role of MMP-2 in diseases, such as cancer and inflammation, highlights the potential therapeutic implications of modulating mechanical stretching to influence MMP-2 expression in tumor cells. Understanding the interplay between mechanical forces and ECM remodelling through this model provides insights for disease diagnosis and treatment, facilitating personalized treatment plans. Therefore, this model represents a significant step in unravelling complex cell biomechanics and offers a valuable tool for investigating mechanical forces and ECM remodelling relationships.\u003c/p\u003e \u003cp\u003eStudy limitations include the absence of protein-level verification despite the establishment of a DL model for predicting MMP-2 gene expression levels in fibroblasts under mechanical stretching. This model was developed solely based on mRNA-level data, while MMP-2 protein activity may be influenced by post-translational modifications and feedback mechanisms. Future studies should incorporate Western blot or ELISA to validate protein expression, thereby further improving model applicability. This study forms part of a series investigating mechanical stretching-regulated fibroblast secretion of ECM components, aiming to establish methodological foundations and reliability for subsequent research. To enhance model applicability, future experiments will quantify additional ECM components (e.g., other MMPs, collagen levels). Future research should focus on protein-level validation and expand the model's applicability to refine the model for constructing personalized ECM substitute materials through mechanical stretching.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eIn this study, the DL model was leveraged to develop an initial predictive model for assessing the impact of mechanical stretching on MMP-2 gene expression levels in fibroblasts. The model demonstrated the ability to accurately predict MMP-2 gene expression levels in fibroblasts across diverse stretching conditions. This model is poised to serve as an invaluable instrument for scrutinizing the intricate relationship between mechanical forces and ECM remodelling, facilitating advancements in tissue engineering, regenerative medicine, and the treatment of various types of fibrosis and ECM-related disorders.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eExtraction and culture of fibroblasts\u003c/h2\u003e \u003cp\u003e In accordance with the approval of the Ethics Committee of the Second Affiliated Hospital of Air Force Medical University (GKJ-Y-202303-082), prepuce samples were obtained from healthy male children aged 8 years (from the Department of Urology at the Second Affiliated Hospital of Air Force Medical University). Fibroblasts were extracted from the samples via density gradient centrifugation and resuspended in cell dishes with Dulbecco's modified Eagle\u0026rsquo;s medium (DMEM, Procell, Wuhan China) supplemented with 0.584 g/L L-glutamine, 10% fetal bovine serum (FBS), 1% penicillin, and 1% streptomycin. The cells were placed in an incubator at 37℃ with 5% CO₂ and 95% air. When the cell density reached greater than 90%, the cells were digested with 0.25% trypsin and passaged at a ratio of 1:2 or 1:3, with logarithmic phase cells being used for experiments.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eMechanical stretching of fibroblasts\u003c/h2\u003e \u003cp\u003eMechanical stretching of cells is performed via a spherical automatic cell-stretching device, which consists mainly of a mechanical stretching loading machine and a control system [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. The equipment requires a 6-well Flexcell Bioflex\u0026reg; culture plate (TTCF 5001C, Flexcell\u0026reg; International Corporation, USA) as a cell carrier for mechanical stretching loading. The mechanical stretching loading machine comprises six spherical columns, corresponding to each well of the 6-well Flexcell Bioflex\u0026reg; culture plate, with each column capable of vertical displacement within a range of 4\u0026ndash;8 mm. Mechanical stretching is achieved by deforming the bottoms of the six wells of the culture plate and establishing six different mechanical stretching parameters. Fibroblasts were seeded onto the culture plate at a density of 1\u0026times;10⁵/cm\u0026sup2;. The next day, after the fibroblasts grew to an infiltrative state, the culture medium and nonadherent cells were washed away, and 3 mL of DMEM containing 10% fetal bovine serum was added. The control system was subsequently used to apply mechanical stretching to fibroblasts at different stretching intensities (5%, 8%, 10%, 12%, 15%, and 22%), durations (3 h, 6 h, 12 h, and 24 h), and frequencies (0.05 Hz, 0.1 Hz, 0.15 Hz, and 0.2 Hz).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003emRNA extraction\u003c/h2\u003e \u003cp\u003eAfter mechanical stretching treatment, total RNA was extracted from the cells according to the manufacturer's instructions for the RNeasy Plus Mini Kit (TIANGEN; Beijing China). This kit simplifies the process of RNA extraction from cells via a rapid spin column method. Initially, the cells were lysed under effective denaturing conditions to rapidly inactivate RNases. Subsequently, genomic DNA removal columns were used to homogenize the samples, and the homogenate was added to the RNase-free adsorption column CR4. In the CR4 column, total RNA binds to the membrane. After a series of wash steps, impurities are removed, ultimately producing purified RNA, which can be eluted in 30\u0026ndash;100 \u0026micro;L of RNase-free ddH\u003csub\u003e2\u003c/sub\u003eO. The purified RNA was used for subsequent experiments.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ecDNA synthesis\u003c/h2\u003e \u003cp\u003ecDNA synthesis was performed via the M5 Super qPCR RT Kit with gDNA Remover (Mei5 Biotechnology, Beijing China). A mixture of 5 \u0026micro;L of mRNA, 5 \u0026micro;L of 5x M5 RT Super Mix, and 10 \u0026micro;L of DEPC-ddH₂O was prepared, incubated at 42℃ for 15 minutes, and then heated at 96℃ for 5 minutes to inactivate the enzyme, ultimately producing cDNA.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eReal-time PCR\u003c/h2\u003e \u003cp\u003eRT‒PCR analysis of MMP-2 and GAPDH was performed on 112 sample groups, with three replicates for each group. PCR amplification was conducted in a 20 \u0026micro;L reaction system, which included 0.8 \u0026micro;L of upstream and downstream primers, 2 \u0026micro;L of cDNA, 10 \u0026micro;L of Taq Pro Universal SYBR qPCR Master Mix (Vazyme, China), and 7.2 \u0026micro;L of RNase-Free ddH₂O. Quantitative analysis was performed via a Bio CFX real-time PCR instrument (Bio-Rad, USA). The sequences and specifications of the PCR primer sets (5\u0026rsquo;-3\u0026rsquo; direction) are shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The mixture was initially preincubated at 95℃ for 3 minutes to activate the polymerase. The mixture was initially preincubated at 95℃ for 3 minutes to activate the polymerase. The amplification process included 40 cycles, with each cycle consisting of denaturation at 95℃ for 10 seconds, annealing at 60℃ for 30 seconds, and extension at 65℃ for 10 seconds. Relative expression levels were calculated via the 2\u003csup\u003e\u0026minus;ΔΔCT\u003c/sup\u003e method (CT\u0026thinsp;=\u0026thinsp;fluorescence threshold; ΔCT\u0026thinsp;=\u0026thinsp;target CT - reference gene (GAPDH) CT; ΔΔCT\u0026thinsp;=\u0026thinsp;sample ΔCT - calibrator sample ΔCT).\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\u003ePrimers used in qPCR for mRNAs in human\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=\"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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eName\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSequence\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSize(bp)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMMP-2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eF: 5\u0026prime;-CCTACACCAAGAACTTCCGTCTG-3\u0026prime;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eR: 5\u0026prime;-GTGCCAAGGTCAATGTCAGGAG-3\u0026prime;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGAPDH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eF: 5\u0026prime;-ACACCCACTCCTCCACCTTTG-3\u0026prime;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eR: 5\u0026prime;-TCCACCACCCTGTTGCTGTAG-3\u0026prime;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21\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=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eEstablishing a DL model\u003c/h2\u003e \u003cp\u003eWe performed min-max normalization on the experimental data of mechanical stretching effects on MMP-2 gene expression to standardize variables within the [0, 1] range. The experimental dataset was partitioned into training and validation sets in a 7:3 ratio. The DL model employs the error backpropagation algorithm, learning by comparing the actual values and predicted values of each sample. For each training sample, the weights are adjusted to minimize the mean squared error between the network's predicted values and the actual target values. The DL model\u0026rsquo;s process was implemented in Python, with the backpropagation neural network architecture constructed using the Python-based TensorFlow 2.8 package. For this model, a grid search was implemented to evaluate 2\u0026ndash;4 hidden layer configurations, with the 3-layer architecture (128\u0026ndash;64\u0026ndash;32 neurons) ultimately selected due to its optimal performance on the validation set (Fig.\u0026nbsp;4A). ReLU activation functions were employed to capture complex nonlinear relationships in input features through higher dimensionality in the first layer, maintain deep feature extraction capacity in the second layer, and further compress and refine features in the third layer. The output layer employs a linear activation for regression prediction. Additionally, Dropout layers (rate\u0026thinsp;=\u0026thinsp;0.5) were incorporated after each hidden layer to randomly deactivate neurons, effectively mitigating overfitting while enhancing generalizability.\u003c/p\u003e \u003cp\u003eDuring the training process, the Adam optimizer was selected with an initial learning rate set to 0.001 to leverage its adaptive learning rate advantages, accelerating convergence and improving optimization efficiency. The loss function was set to the mean squared error (MSE), which is suitable for minimizing targets in regression problems. The model training used a fixed random seed to ensure experimental reproducibility, optimizing the model parameters through 1000 epochs and a batch size of 128. To achieve interactivity, we established a GUI via Python 3.9.18 (Fig.\u0026nbsp;4B).\u003c/p\u003e\u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eEstablishment of the external validation set\u003c/h2\u003e \u003cp\u003eThis study used the PubMed database as the information source and conducted a search using the query \u0026ldquo;((Fibroblasts[Title]) OR (Fibroblast[Title])) AND ((((Mechanical[Title]) OR (mechanical stretch[Title])) OR (mechanical stretching[Title])) OR (dynamics[Title])) OR (stretch[Title])) OR (stretching[Title]))\u0026rdquo;, covering the time range from the establishment of the database to December 1, 2024, initially yielding 670 SCI papers. The inclusion criteria were as follows: (1) the research focused on the effects of mechanical stretching on fibroblasts and (2) the experimental content included MMP-2 gene expression levels. The exclusion criteria were as follows: (1) missing information in the articles and (2) data that could not be extracted from the literature. Ultimately, 11 data points were extracted from the statistical graphs in the literature via Origin 2022 software to establish the external validation set.\u003c/p\u003e \u003c/div\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMMP-2\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ematrix metalloproteinase-2\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eECM extracellular matrix\u003c/div\u003e \u003cdiv class=\"Description\"\u003e\u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCRW\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003echronic refractory wound\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMMPs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMatrix metalloproteinases\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eartificial intelligence\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDL\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003edeep learning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGUI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003egraphical user interface\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRMSE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eroot mean square error\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMAE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emean absolute error\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMSE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emean square error.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank the Bioinspired Engineering and Biomechanics Centre（BEBC）, Xi\u0026rsquo;an Jiaotong University, for excellent technical support, collaboration, and provision of facilities for this work. We also thank Feng Xu and Yuanbo Jia for their expert technical assistance and skillful work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRX: Project administration, writing - original draft. HZ: Data curation, validation. ZS: Conceptualization, writing - review \u0026amp; editing. YZ: Methodology, data curation. RH: Critical revision of the article, funding acquisition. JL: Project administration, Funding acquisition.\u0026nbsp;All authors reviewed the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Key Industrial Innovation Chain Project of Shaanxi Provincial Key Research and Development Plan (No. 2022ZDLSF04-03) and the Key Project for Tackling Key Core Technology of Shaanxi (No. 2024SF-GJHX-20).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo datasets were generated or analysed during the current study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthical approval for this study was obtained from the Ethics Committee of the Second Affiliated Hospital of Air Force Medical University (GKJ-Y-202303-082). All procedures conducted in studies involving human participants adhered to the ethical standards of the institutional and/or national research committees, as well as the 1964 Helsinki Declaration and its subsequent amendments or equivalent ethical guidelines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing financial or nonfinancial interests related to this work.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eJiang Y, Huang S, Fu X, Liu H, Chen H. Epidemiology of chronic cutaneous wounds in China\u003cstrong\u003e.\u003c/strong\u003e Wound Repair Regen. 2015, 19(2):181-188.\u003c/li\u003e\n\u003cli\u003eRice JB, Desai U, Cummings AKG, Birnbaum HG, Skornicki M, Parsons NB. Burden of diabetic foot ulcers for medicare and private insurers\u003cstrong\u003e.\u003c/strong\u003e Diabetes Care. 2014, 37(3):651-658.\u003c/li\u003e\n\u003cli\u003eLaura, Martinengo, Maja, Olsson, Ram, Bajpai, Michael, Soljak, Zee, Upton. Prevalence of chronic wounds in the general population: systematic review and meta-analysis of observational studies\u003cstrong\u003e.\u003c/strong\u003e Ann Epidemiol. 2018.\u003c/li\u003e\n\u003cli\u003eOlsson M, Jarbrink K, Divakar U, Bajpai R, Car J. The humanistic and economic burden of chronic wounds: A systematic review\u003cstrong\u003e.\u003c/strong\u003e Wound Repair Regen. 2019, 27(1):114-125.\u003c/li\u003e\n\u003cli\u003eDavis FM, Kimball A, Boniakowski A, Gallagher K. Dysfunctional Wound Healing in Diabetic Foot Ulcers: New Crossroads\u003cstrong\u003e.\u003c/strong\u003e Curr Diab Rep. 2018, 18(1):2.\u003c/li\u003e\n\u003cli\u003eBjarnsholt T, Kirketerp-M\u0026oslash;ller K, Jensen P\u0026Oslash;, Madsen KG, Givskov M. Why chronic wounds will not heal: a novel hypothesis\u003cstrong\u003e.\u003c/strong\u003e Wound Repair Regen. 2010, 16(1):2-10.\u003c/li\u003e\n\u003cli\u003eBarrientos S, Brem H, Stojadinovic O, Tomic-Canic M. Clinical application of growth factors and cytokines in wound healing\u003cstrong\u003e.\u003c/strong\u003e Wound Repair Regen. 2015, 22(5):569-578.\u003c/li\u003e\n\u003cli\u003eGuo Y, Bian Z, Xu Q, Wen X, Kang J, Lin S, Wang X, Mi Z, Cui J, Zhang Z. Novel tissue-engineered skin equivalent from recombinant human collagen hydrogel and fibroblasts facilitated full-thickness skin defect repair in a mouse model\u003cstrong\u003e.\u003c/strong\u003e Mater Sci Eng C Mater Biol Appl. 2021, 130:112469.\u003c/li\u003e\n\u003cli\u003ePientaweeratch S, Panapisal V, Tansirikongkol A. Antioxidant, anti-collagenase and anti-elastase activities of Phyllanthus emblica , Manilkara zapota and silymarin: an in vitro comparative study for anti-aging applications\u003cstrong\u003e.\u003c/strong\u003e Pharm Biol. 2016, 24(7-9):1-8.\u003c/li\u003e\n\u003cli\u003eManosroi A, Kumguan K, Chankhampan C, Manosroi W, Manosroi J. Nanoscale gelatinase A (MMP-2) inhibition on human skin fibroblasts of Longkong (Lansium domesticum Correa) leaf extracts for anti-aging\u003cstrong\u003e.\u003c/strong\u003e J Nanosci Nanotechnol. 2012, 12(9):7187-7197.\u003c/li\u003e\n\u003cli\u003eSreesada P, Vandana, Krishnan B, Amrutha R, Chavan Y, Alfia H, Jyothis A, Venugopal P, Aradhya R, Suravajhala P. Matrix metalloproteinases: Master regulators of tissue morphogenesis\u003cstrong\u003e.\u003c/strong\u003e Gene. 2025, 933.\u003c/li\u003e\n\u003cli\u003eMaybee DV, Ink NL, Ali MA. Novel roles of MT1-MMP and MMP-2: beyond the extracellular milieu\u003cstrong\u003e.\u003c/strong\u003e Int J Mol Sci. 2022, 23(17):9513.\u003c/li\u003e\n\u003cli\u003ePrzekora A. A Concise Review on Tissue Engineered Artificial Skin Grafts for Chronic Wound Treatment: Can We Reconstruct Functional Skin Tissue In Vitro? Cells. 2020, 9(7).\u003c/li\u003e\n\u003cli\u003eChen Y, Li C, Xie H, Fan Y, Yang Z, Ma J, He D, Li L. Infiltrating mast cells promote renal cell carcinoma angiogenesis by modulating PI3K\u0026rarr;︀AKT\u0026rarr;︀GSK3\u0026beta;\u0026rarr;︀AM signaling\u003cstrong\u003e.\u003c/strong\u003e Oncogene. 2017, 36(20):2879-2888.\u003c/li\u003e\n\u003cli\u003eXie Y, Ouyang X, Wang G. Mechanical strain affects collagen metabolism-related gene expression in scleral fibroblasts\u003cstrong\u003e.\u003c/strong\u003e Biomed Pharmacother. 2020, 126:110095.\u003c/li\u003e\n\u003cli\u003eLiu S, Lin Z. Vascular smooth muscle cells mechanosensitive regulators and vascular remodeling\u003cstrong\u003e.\u003c/strong\u003e J Vasc Res. 2022, 59(2):90-113.\u003c/li\u003e\n\u003cli\u003eMaione AG, Brudno Y, Stojadinovic O, Park LK, Smith A, Tellechea A, Leal EC, Kearney CJ, Veves A, Tomic-Canic M. Three-dimensional human tissue models that incorporate diabetic foot ulcer-derived fibroblasts mimic in vivo features of chronic wounds\u003cstrong\u003e.\u003c/strong\u003e Tissue Eng Part C Methods. 2015, 21(5):499-508.\u003c/li\u003e\n\u003cli\u003eHodde JP, Johnson CE. Extracellular matrix as a strategy for treating chronic wounds\u003cstrong\u003e.\u003c/strong\u003e Am J Clin Dermatol. 2007, 8:61-66.\u003c/li\u003e\n\u003cli\u003eZhao H, Li Z, Wang Y, Zhou K, Li H, Bi S, Wang Y, Wu W, Huang Y, Peng B. Bioengineered MSC-derived exosomes in skin wound repair and regeneration\u003cstrong\u003e.\u003c/strong\u003e Front Cell Dev Biol. 2023, 11:1029671.\u003c/li\u003e\n\u003cli\u003eKerstan A, Dieter K, Niebergall-Roth E, Klingele S, J\u0026uuml;nger M, Hasslacher C, Daeschlein G, Stemler L, Meyer-Pannwitt U, Schubert K. Translational development of ABCB5+ dermal mesenchymal stem cells for therapeutic induction of angiogenesis in non-healing diabetic foot ulcers\u003cstrong\u003e.\u003c/strong\u003e Stem Cell Res Ther. 2022, 13(1):455.\u003c/li\u003e\n\u003cli\u003ePeng Y, Meng H, Li P, Jiang Y, Fu X. Research advances of stem cell-based tissue engineering repair materials in promoting the healing of chronic refractory wounds on the body surface\u003cstrong\u003e.\u003c/strong\u003e Zhonghua Shao Shang Yu Chuang Mian Xiu Fu Za Zhi. 2023, 39(3):290-295.\u003c/li\u003e\n\u003cli\u003eWang H, Fu T, Du Y, Gao W, Huang K, Liu Z, Chandak P, Liu S, Van Katwyk P, Deac A. Scientific discovery in the age of artificial intelligence\u003cstrong\u003e.\u003c/strong\u003e Nature. 2023, 620(7972):47-60.\u003c/li\u003e\n\u003cli\u003eKosarac A, Cep R, Trochta M, Knezev M, Zivkovic A, Mladjenovic C, Antic A. Thermal behavior modeling based on BP neural network in Keras framework for motorized machine tool spindles\u003cstrong\u003e.\u003c/strong\u003e Materials (Basel). 2022, 15(21):7782.\u003c/li\u003e\n\u003cli\u003eChen S, Mienaltowski MJ, Birk DE. Regulation of corneal stroma extracellular matrix assembly\u003cstrong\u003e.\u003c/strong\u003e Exp Eye Res. 2015, 133:69-80.\u003c/li\u003e\n\u003cli\u003eRodrigues M, Kosaric N, Bonham CA, Gurtner GC. Wound healing: a cellular perspective\u003cstrong\u003e.\u003c/strong\u003e Physiol Rev. 2018.\u003c/li\u003e\n\u003cli\u003eLeng L, Ma J, Sun X, Guo B, Li F, Zhang W, Chang M, Diao J, Wang Y, Wang W. Comprehensive proteomic atlas of skin biomatrix scaffolds reveals a supportive microenvironment for epidermal development\u003cstrong\u003e.\u003c/strong\u003e J Tissue Eng. 2020, 11:2041731420972310.\u003c/li\u003e\n\u003cli\u003eHuang Y, Kyriakides TR. The role of extracellular matrix in the pathophysiology of diabetic wounds\u003cstrong\u003e.\u003c/strong\u003e Matrix Biol Plus. 2020, 6:100037.\u003c/li\u003e\n\u003cli\u003eLim WJ, Chan PF, Abd Hamid R. A 1, 4-benzoquinone derivative isolated from Ardisia crispa (Thunb.) A. DC. root suppresses angiogenesis via its angiogenic signaling cascades\u003cstrong\u003e.\u003c/strong\u003e Saudi Pharm J. 2024, 32(1):101891.\u003c/li\u003e\n\u003cli\u003eDo CTP, Prochnau JY, Dominguez A, Wang P, Rao MK. The Road Ahead in Pancreatic Cancer: Emerging Trends and Therapeutic Prospects\u003cstrong\u003e.\u003c/strong\u003e Biomedicines. 2024, 12(9).\u003c/li\u003e\n\u003cli\u003eJones DL, Hallstr\u0026ouml;m GF, Jiang X, Locke RC, Evans MK, Bonnevie ED, Srikumar A, Leahy TP, Nijsure MP, Boerckel JD, et al. Mechanoepigenetic regulation of extracellular matrix homeostasis via Yap and Taz\u003cstrong\u003e.\u003c/strong\u003e Proc Natl Acad Sci U S A 2023, 120(22):e2211947120.\u003c/li\u003e\n\u003cli\u003eYeganegi A, Whitehead K, de Castro Br\u0026aacute;s LE, Richardson WJ. Mechanical strain modulates extracellular matrix degradation and byproducts in an isoform-specific manner\u003cstrong\u003e.\u003c/strong\u003e Connect Tissue Res. 2023, 1867(3):130286.\u003c/li\u003e\n\u003cli\u003ePakpahan ND, Kyawsoewin M, Manokawinchoke J, Termkwancharoen C, Egusa H, Limraksasin P, Osathanon T. Effects of mechanical loading on matrix homeostasis and differentiation potential of periodontal ligament cells: A scoping review\u003cstrong\u003e.\u003c/strong\u003e J Periodontal Res. 2024.\u003c/li\u003e\n\u003cli\u003eWang Y, Dang Z, Cui W, Yang L. Mechanical stretch and hypoxia inducible factor-1 alpha affect the vascular endothelial growth factor and the connective tissue growth factor in cultured ACL fibroblasts\u003cstrong\u003e.\u003c/strong\u003e Connect Tissue Res. 2017, 58(5):407-413.\u003c/li\u003e\n\u003cli\u003eDai W, Zhou H, Du J, Xiao R, Su J, Liu Z, Huang R, Li Y, Li J. Mechanical stretching enhances the cellular and paracrine effects of bone marrow mesenchymal stem cells on diabetic wound healing\u003cstrong\u003e.\u003c/strong\u003e Burns Trauma. 2025.\u003c/li\u003e\n\u003cli\u003eLin L-Q, Zeng H-K, Luo Y-L, Chen D-F, Ma X-Q, Chen H-J, Song X-Y, Wu H-K, Li S-Y. Mechanical stretch promotes apoptosis and impedes ciliogenesis of primary human airway basal stem cells\u003cstrong\u003e.\u003c/strong\u003e Respir Res. 2023, 24(1):237.\u003c/li\u003e\n\u003cli\u003eShan S, Fang B, Zhang Y, Wang C, Zhou J, Niu C, Gao Y, Zhao D, He J, Wang J. Mechanical stretch promotes tumoricidal M1 polarization via the FAK/NF‐\u0026kappa;B signaling pathway\u003cstrong\u003e.\u003c/strong\u003e FASEB J. 2019, 33(12):13254-13266.\u003c/li\u003e\n\u003cli\u003eMa H, Wang L, Sun H, Yu Q, Yang T, Wang Y, Niu B, Jia Y, Liu Y, Liang Z. MIR-107/HMGB1/FGF-2 axis responds to excessive mechanical stretch to promote rapid repair of vascular endothelial cells\u003cstrong\u003e.\u003c/strong\u003e Arch Biochem Biophys. 2023, 744:109686.\u003c/li\u003e\n\u003cli\u003eLiu C, Feng P, Li X, Song J, Chen W. Expression of MMP-2, MT1-MMP, and TIMP-2 by cultured rabbit corneal fibroblasts under mechanical stretch\u003cstrong\u003e.\u003c/strong\u003e Exp Biol Med (Maywood). 2014, 239(8):907-912.\u003c/li\u003e\n\u003cli\u003eShelton L, Rada JS. Effects of cyclic mechanical stretch on extracellular matrix synthesis by human scleral fibroblasts\u003cstrong\u003e.\u003c/strong\u003e Exp Eye Res. 2007, 84(2):314-322.\u003c/li\u003e\n\u003cli\u003eJacho D, Rabino A, Garcia-Mata R, Yildirim-Ayan E. Mechanoresponsive regulation of fibroblast-to-myofibroblast transition in three-dimensional tissue analogues: Mechanical strain amplitude dependency of fibrosis\u003cstrong\u003e.\u003c/strong\u003e Sci Rep. 2022, 12(1):16832.\u003c/li\u003e\n\u003cli\u003eKatzengold R, Orlov A, Gefen A. A novel system for dynamic stretching of cell cultures reveals the mechanobiology for delivering better negative pressure wound therapy\u003cstrong\u003e.\u003c/strong\u003e Biomech Model Mechanobiol. 2021, 20:193-204.\u003c/li\u003e\n\u003cli\u003eHuang R, Xu L, Wang Y, Zhang Y, Lin B, Lin Z, Li J, Li X. Efficient fabrication of stretching hydrogels with programmable strain gradients as cell sheet delivery vehicles\u003cstrong\u003e.\u003c/strong\u003e Mater Sci Eng C Mater Biol Appl. 2021, 129:112415.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"biomedical-engineering-online","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmeo","sideBox":"Learn more about [BioMedical Engineering OnLine](http://biomedical-engineering-online.biomedcentral.com/)","snPcode":"12938","submissionUrl":"https://submission.nature.com/new-submission/12938/3","title":"BioMedical Engineering OnLine","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Mechanical stretching, Fibroblasts, MMP-2, Deep learning, Predictive model","lastPublishedDoi":"10.21203/rs.3.rs-5869090/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5869090/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003eMatrix metalloproteinase-2 (MMP-2) secretion homeostasis, governed by the multifaceted interplay of skin stretching, is a pivotal determinant influencing wound healing dynamics. This investigation endeavors to devise an artificial intelligence (AI) prediction framework delineating the modulation of MMP-2 expression under stretching conditions, thereby unravelling profound insights into the mechanobiological orchestration of MMP-2secretion and fostering novel mechanotherapeutic strategies targeted at MMP-2 modulation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003eEmploying a bespoke mechanical tensile loading apparatus, diverse mechanical tensile stimuli were administered to fibroblasts, with parameters such as tensile shape and frequency duration constituting the mechanical loading regimen. Furthermore, reverse transcription polymerase chain reaction (RT‒PCR) assays were conducted to measure MMP-2 gene expression levels in fibroblasts subjected to mechanical stretching. Subsequently, the resulting data were partitioned into training and validation cohorts at a 7:3 ratio, facilitating the development of the deep learning (DL) model via a back propagation neural network predicated on the training set. An external validation set was also curated by culling pertinent literature from the PubMed database to assess the predictive ability of the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eAnalysis of 336 data points related to MMP-2 gene expression via RT‒PCR corroborated the variability in MMP-2 gene expression levels in response to distinct mechanical stretching regimens. Consequently, a DL model was successfully crafted via the backpropagation algorithm to delineate the impact of mechanical stretching stimuli on MMP-2 gene expression levels. The model, characterized by an R²value of 0.73, evinced a commendable fit with the training dataset, elucidating the intricate interplay between the input features and the target variable. Notably, the model exhibited minimal prediction errors, as evidenced by a root mean square error (RMSE) of 0.42 and a mean absolute error (MAE) of 0.28. Furthermore, the model showcased robust generalization capabilities during validation, yielding R²values of 0.70 and 0.71 for the validation and external validation sets, respectively, revealing its predictive accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions: \u003c/strong\u003eThe DL model fashioned through the backpropagation algorithm adeptly forecasts the impact of mechanical stretching stimuli on MMP-2 gene expression levels in fibroblasts with relative precision. These findings provide a foundation for the modulation of MMP homeostasis via mechanical stretching to expedite the healing of recalcitrant chronic refractory wound (CRW).\u003c/p\u003e","manuscriptTitle":"Construction of a deep learning-based predictive model to evaluate the influence of mechanical stretching stimuli on MMP-2 gene expression levels in fibroblasts","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-02 06:28:50","doi":"10.21203/rs.3.rs-5869090/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Accepted","date":"2025-05-22T19:55:06+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-18T21:29:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-07T13:49:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"186806234233814038034343944598584325153","date":"2025-05-05T12:01:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"113400231136858966131567981212183321361","date":"2025-04-30T11:29:07+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-30T11:23:28+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-30T07:07:51+00:00","index":"","fulltext":""},{"type":"submitted","content":"BioMedical Engineering OnLine","date":"2025-04-28T17:27:07+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"biomedical-engineering-online","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmeo","sideBox":"Learn more about [BioMedical Engineering OnLine](http://biomedical-engineering-online.biomedcentral.com/)","snPcode":"12938","submissionUrl":"https://submission.nature.com/new-submission/12938/3","title":"BioMedical Engineering OnLine","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0ad18f72-2b76-4ec3-86bf-b13769de9c28","owner":[],"postedDate":"May 2nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-06-09T16:10:27+00:00","versionOfRecord":{"articleIdentity":"rs-5869090","link":"https://doi.org/10.1186/s12938-025-01399-0","journal":{"identity":"biomedical-engineering-online","isVorOnly":false,"title":"BioMedical Engineering OnLine"},"publishedOn":"2025-06-05 15:57:48","publishedOnDateReadable":"June 5th, 2025"},"versionCreatedAt":"2025-05-02 06:28:50","video":"","vorDoi":"10.1186/s12938-025-01399-0","vorDoiUrl":"https://doi.org/10.1186/s12938-025-01399-0","workflowStages":[]},"version":"v1","identity":"rs-5869090","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5869090","identity":"rs-5869090","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}