{"doi":"10.1145/3768322.3769033","title":"Systematic evaluation of 566 sequence-based features for predicting protein stability changes induced by mutations using machine learning","abstract":"Accurately predicting protein stability changes (ΔΔG) upon amino acid substitutions is essential for understanding disease mechanisms and guiding protein engineering. While recent deep learning frameworks can extract representations directly from protein sequences or structures, it remains unclear which sequence-derived features are most predictive and interpretable, and how they may complement large embeddings. Here, we systematically evaluated 566 descriptors from the AAIndex database to identify compact, biochemically meaningful feature subsets for ΔΔG prediction. Using the side-chain stability contribution (S3) dataset as a benchmark, we tested sliding windows from 5 to 21 residues and found that a 13-residue window consistently optimized performance across models. Random Forest (ROC-AUC = 0.787) and XGBoost (ROC-AUC = 0.776) achieved the best baseline results. Grouping features into categories revealed hydrophobicity, physicochemical, and stability-related indices as the most informative. Restricting models to the top 20–30 ranked features improved AUC to 0.820, while a minimal subset limited to hydrophobicity and stability features achieved AUC = 0.813, maintaining performance with enhanced interpretability. These findings establish a concise, robust feature panel that not only provides mechanistic insights but can also be seamlessly integrated with deep learning embeddings, paving the way for hybrid approaches that combine predictive power with interpretability in ΔΔG prediction.","journal":null,"year":2025,"id":584348,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":0,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9465,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2025-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":1497188,"name":"Junyan Li","orcid":"0000-0002-1392-5796","position":1,"is_corresponding":false},{"id":255755,"name":"Dongxiao Liu","orcid":"0000-0001-9656-8976","position":2,"is_corresponding":false},{"id":1497189,"name":"Krish Wahi","orcid":"0009-0003-6486-1387","position":3,"is_corresponding":false},{"id":255757,"name":"Shaolei Teng","orcid":"0000-0001-8326-9889","position":4,"is_corresponding":false},{"id":1297786,"name":"Qiaobin Yao","orcid":"0009-0009-3832-1524","position":0,"is_corresponding":true}],"reference_count":16,"raw_metadata":null,"created_at":"2026-07-19T02:59:11.978098Z","pmid":null,"pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}