Abstract
Structure-based computational methods have been widely used in exploring protein-ligand interactions, including predicting the binding ligands of a given protein based on their structural complementarity. Compared to other protein and ligand representations, the advantages of a surface representation include reduced sensitivity to subtle changes in the pocket and ligand conformation and fast search speed. Here we developed a novel method named PL-PatchSurfer (Protein-Ligand PatchSurfer). PL-PatchSurfer represents the protein binding pocket and the ligand molecular surface as a combination of segmented surface patches. Each patch is characterized by its geometrical shape and the electrostatic potential, which are represented using the 3D Zernike descriptor (3DZD). We first tested PL-PatchSurfer on binding ligand prediction and found it outperformed the pocket-similarity based ligand prediction program. We then optimized the search algorithm of PL-PatchSurfer using the PDBbind dataset. Finally, we explored the utility of applying PL-PatchSurfer to a larger and more diverse dataset and showed that PL-PatchSurfer was able to provide a high early enrichment for most of the targets. To the best of our knowledge, PL-PatchSurfer is the first surface patch-based method that treats ligand complementarity at protein binding sites. We believe that using a surface patch approach to better understand protein-ligand interactions has the potential to significantly enhance the design of new ligands for a wide array of drug-targets.
Highlights
Interactions with small ligand molecules are essential aspects of proteins
PL-PatchSurfer takes a complementary approach to Patch-Surfer in that it predicts the binding ligand of a given protein based on the molecular surface complementarity between the ligand and the protein pocket
First we have demonstrated that PL-PatchSurfer works well in predicting binding ligands for pockets of target proteins
Summary
Prediction of binding ligands for a protein provides important clues of biological functions of proteins. Computational prediction can help building hypothesis of protein functions that can be later tested by experiments. Binding ligands for proteins can be in principle predicted by identifying similar known binding pockets from known protein structures. There are several strategies proposed to predict binding ligands by pocket comparison in the past [2,3,4,5,6,7,8,9]. Catalytic Site Atlas [10] and AFT [11] compare a few functional residues in binding pockets and quantify the pocket similarity with the root mean square deviation (RMSD) of the residues
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have