Prediction of protein binding residues for metal ions, nucleic acids, and small molecules

Abstract Background Protein–ligand interactions are central to cellular regulation and therapeutic targeting. Experimental identification of binding residues remains costly and time-consuming, motivating the need for efficient computational approaches. We present an interpretable, XGBoost-based framework for residue-level prediction of protein binding sites specific to three ligand classes—metal ions, nucleic acids, and small molecules—emphasizing predictive accuracy and user accessibility. Methods A benchmark dataset [1] comprising 1314 annotated protein sequences was processed into peptide windows (±5, ±11, ±20 residues) centered on binding sites, maintaining a 1:3 ratio of positive to negative samples. Sequence-derived descriptors were generated using the iLearn toolkit, including amino acid composition (AAC), grouped AAC (GAAC), CTD composition, and BLOSUM62 substitution scores, yielding >8000 features subsequently reduced by feature-importance heuristics. More than 60 classifiers were evaluated; XGBoost consistently achieved superior performance after hyperparameter optimization. Results Independent models trained for each ligand type achieved high predictive power and biological interpretability. For small-molecule binding, the model attained an AUC of 0.99 and F1 of 0.91, with key features involving small residues (G, C, S, A) reflecting steric constraints. The nucleic-acid model reached AUC 0.99 and F1 0.95, highlighting glycine-rich and positively charged motifs typical of RNA/DNA interactions. The metal-ion model (AUC 0.92, F1 0.65) emphasized cysteine and histidine, consistent with metalloprotein binding patterns. Compared with reference CNN models, our framework achieved lower log-loss, greater interpretability, and stable cross-validation. Conclusion A standalone graphical user interface (GUI) implemented in Python allows users to input protein sequences and obtain residue-level binding probabilities, exportable as CSV files. This work demonstrates that well-engineered classical machine learning can rival deep learning in protein binding prediction, offering interpretable, high-performance tools to support biomedical and pharmaceutical research. References [1] Littmann M, Heinzinger M, Dallago C, Weissenow K, Rost B. Protein embeddings and deep learning predict binding residues for various ligand classes. Sci Rep. 2021 Dec 13;11(1):23916.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.