G-Losa: An Efficient Computational Tool for Local Structure-Centric Biological Studies

Abstract

Molecular recognition such as ligand binding and protein-protein interaction (PPI) is a fundamental way for biological molecules to play their functional roles. These specific interactions are involved in the local regions of the molecules, rather than global structures. Therefore, accurate characterization of local structures in protein is needed to better understand biological mechanisms and to rationally design effective drugs. G-LoSA, a recently developed local structure alignment tool, has the advantages of not only predicting the ligand binding sites with high accuracy, but also identifying a single template ligand that is highly similar to the target ligand. Here, we present an improved version of G-LoSA aiming at extending its applicability to broad local structure-centric biological studies. The method generates all possible alignments between two local structures by iterative maximum clique search and fragment superposition and then determines the final optimal alignment by a G-LoSA alignment scoring function, GA-score. GA-score is a length-independent and physicochemical property-based scoring function to measure structural similarity between two local structures. G-LoSA outperforms its previous version in identifying ligand templates and also shows robust performance in detecting similar ligand binding pockets and PPI interfaces from the benchmark sets. Finally, we introduce its application to in silico fragment-based drug design. As demonstrated by this work, G-LoSA is a promising computational tool that can be universally applied to diverse local structure-centric biological studies.