Audio as a Molecular Modality: Oscillatory Spectral Encoding of Protein Binding Pockets
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Current computational representations of protein binding pockets (graphs, images, sequences) lack an intrinsic temporal dimension. Here we show that mapping pocket residues to localized oscillatory functions and exploiting wave superposition produces spectral fingerprints that encode collective spatialchemical organization inaccessible to conventional feature aggregation. Across 1,489 co-crystal structures spanning 15 protein families, this deterministic, training-free method achieves 85.7% classification accuracy with a 3.0-fold effect-size improvement over aggregation baselines using identical input features. Within-family kinase subfamily discrimination (78.8% across 42 subfamilies) and DFG conformational state detection (92.6% minority-class recall) demonstrate sensitivity to the fine-grained structural variation that governs drug selectivity. By introducing time as a representational dimension for molecular structure, this work establishes audio as a new modality for structural bioinformatics, positioning pocket analysis to leverage the rapid advances in self-supervised audio AI.
Fields of Study & Subjects
Computational Drug Discovery Methods
Protein Structure and Dynamics
Machine Learning in Bioinformatics
Publication details
- Journal / source
- ChemRxiv
- Publication date
- 31 March 2026
- Language
- EN
- Document type
- Preprint