Saeed, H.; Huang, W.; Yang, A.

Proteins contain key residues that critically influence their function, stability, and interactions. However, accurately identifying these \"key tuning\" residues remains challenging. In this study, we present an information-theoretic approach, inspired from signal processing methods, to identify key residues in protein sequences. We propose a feature selection framework that integrates both unsupervised and supervised approaches, using Shannon entropy and mutual information as initial priors, respectively. A key aspect of our method is the use of Cramer\'s V and Thiel\'s U to decouple coevolved residues. The supervised approach employed mutual information to assess residue relationships, followed by evolutionary decoupling with Cramer\'s V to better decouple individual contributions. This two-step process demonstrated statistically significant improvement compared to DeepSequence, a leading tool (p-value=0.0149). The unsupervised approach, based solely on Shannon entropy prior to decoupling, showed competitive performance despite a slightly lower statistical significance (pvalue=0.14). Although the supervised method yields higher accuracy given sufficient training data, the unsupervised technique is simpler, requiring only a multiple sequence alignment (MSA) that can be obtained via BLAST. When applied to a system with well-characterised structural information, specifically nanobody-antigen interactions, our methods not only effectively identified the binding residues but also accurately reconstructed a contiguous binding surface, surpassing current state-of-the-art computational tools. Collectively, our methods provide a flexible and interpretable solution for various protein engineering applications. This work contributes to the broader field of protein structure-function mapping by offering a computationally efficient and transparent approach for identifying key tuning residues, paving the way for new strategies in rational protein design and functional modulation with potential applications in drug discovery, enzyme optimization, and novel protein development.