2026-10-14T17:15:05Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34027267 2026-09-29T18:18:10Z category_1 category_7 category_15 category_873 category_734 category_931 category_811 category_64 portal_63 item_type_3 month_year_09_2026

Benchmarking Experiments with External Data, Domain-Specific Features, and Learning Strategies for Modeling ADME and Potency of Coronavirus Main Protease Inhibitors Ida Titus (22182380) Ashok Palaniappan (3165975) Biophysics Medicine Neuroscience Chemical Sciences not elsewhere classified Biological Sciences not elsewhere classified Information Systems not elsewhere classified Mathematical Sciences not elsewhere classified Cancer withstand close inspection strongest significant generalization reproducible computational foundation openadmet consortium hosted graph neural network evaluating critical assumptions challenge test set aware attentivefp graph antiviral drug candidates antiviral compounds could transformer model yielded consistently optimal model ensemble integration approaches ensemble fusion strategies distinctive chemical properties classical machine learning >< optimal xgboost model aware modeling strategy curated public data gnn variants across learning strategies model selection pharmacokinetic properties enhanced xgboost based approaches ></ sup work provides statistical benchmarking splitting protocols specific features rigorous setting requiring prediction multitarget settings molecular design mdckii ). identifying antivirals feature space factorial design external data explicitly engineered end point dramatically accelerate clinical evaluation broader tree blind evaluation beyond benchmarking analysis revealed Predictive computational models of potency and pharmacokinetic properties of antiviral drug candidates can dramatically accelerate the path from molecular design to clinical evaluation. The ASAP-Polaris-OpenADMET consortium hosted the open-science Antiviral Drug Discovery Challenge 2025, providing a rigorous setting for this problem and requiring prediction of potency against the SARS-CoV-2 and MERS-CoV Mpro alongside five ADME end points (LogD, KSOL, HLM, MLM, and MDR1-MDCKII). In response, we pursued a domain-aware modeling strategy, reasoning that the distinctive chemical properties of antiviral compounds could be explicitly engineered in the feature space. We tested this premise by isolating the key molecular descriptors capable of identifying antivirals, augmenting challenge data sets with curated public data, and building both classical machine learning and graph neural network (GNN) variants. For ADME modeling, ensemble gradient boosting models trained on our bespoke feature space consistently outperformed both broader tree-based approaches and GNN variants across the five ADME end points in blind evaluation. By a factorial design of experiments and statistical benchmarking, we evaluated the performance of the optimal XGBoost model for each end point. For modeling potency, we explored varied pretraining paradigms and ensemble fusion strategies, and found that a stacking ensemble combining docking-enhanced XGBoost with a stereochemistry-aware AttentiveFP Graph-Transformer model yielded the strongest significant generalization on the Challenge test set, suggesting that orthogonal model families reduced epistemic uncertainty attributable to model selection in data-limited, multitarget settings. Beyond benchmarking, our experiments provided a testbed for evaluating critical assumptions in the field, encompassing data augmentation strategies, splitting protocols, and ensemble integration approaches. Our analysis revealed that a consistently optimal model is elusive, and that several standard industry practices fail to withstand close inspection. Overall, our work provides both practical benchmarked modeling strategies for ADME and potency prediction and a reproducible computational foundation for AI-driven antiviral drug discovery efforts. 2026-09-29T00:00:00Z Dataset Dataset 10.1021/acs.jcim.6c00304.s002 https://figshare.com/articles/dataset/Benchmarking_Experiments_with_External_Data_Domain-Specific_Features_and_Learning_Strategies_for_Modeling_ADME_and_Potency_of_Coronavirus_Main_Protease_Inhibitors/34027267 CC BY-NC 4.0