Machine\nLearning Models for Predicting Monoclonal\nAntibody Biophysical Properties from Molecular Dynamics Simulations\nand Deep Learning-Based Surface Descriptors

Monoclonal antibodies (mAbs) have found extensive applications\nand development in treating various diseases. From the pharmaceutical\nindustry’s perspective, the journey from the design and development\nof mAbs to clinical testing and large-scale production is a highly\ntime-consuming and resource-intensive process. During the research\nand development phase, assessing and optimizing the developability\nof mAbs is of paramount importance to ensure their success as candidates\nfor therapeutic drugs. The critical factors influencing mAb development\nare their biophysical properties, such as aggregation propensity,\nsolubility, and viscosity. This study utilized a data set comprising\n12 biophysical properties of 137 antibodies from a previous study\n(Proc Natl Acad Sci USA. 114(5):944–949, 2017). We employed\nfull-length antibody molecular dynamics simulations and machine learning\ntechniques to predict experimental data for these 12 biophysical properties.\nAdditionally, we utilized a newly developed deep learning model called\nDeepSP, which directly predicts the dynamical and structural properties\nof spatial aggregation propensity and spatial charge map in different\nantibody regions from sequences. Our research findings indicate that\nthe machine learning models we developed outperform previous methods\nin predicting most biophysical properties. Furthermore, the DeepSP\nmodel yields similar predictive results compared to molecular dynamic\nsimulations while significantly reducing computational time. The code\nand parameters are freely available at https://github.com/Lailabcode/AbDev. Also, the webapp, AbDev, for 12 biophysical properties prediction\nhas been developed and provided at https://devpred.onrender.com/AbDev.