Premium
Ensemble Architecture for Prediction of Enzyme‐ligand Binding Residues Using Evolutionary Information
Author(s) -
Pai Priyadarshini P.,
Dattatreya Rohit Kadam,
Mondal Sukanta
Publication year - 2017
Publication title -
molecular informatics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.481
H-Index - 68
eISSN - 1868-1751
pISSN - 1868-1743
DOI - 10.1002/minf.201700021
Subject(s) - computer science , discriminative model , computational biology , docking (animal) , support vector machine , artificial intelligence , machine learning , data mining , chemistry , biology , medicine , nursing
Enzyme interactions with ligands are crucial for various biochemical reactions governing life. Over many years attempts to identify these residues for biotechnological manipulations have been made using experimental and computational techniques. The computational approaches have gathered impetus with the accruing availability of sequence and structure information, broadly classified into template‐based and de novo methods. One of the predominant de novo methods using sequence information involves application of biological properties for supervised machine learning. Here, we propose a support vector machines‐based ensemble for prediction of protein‐ligand interacting residues using one of the most important discriminative contributing properties in the interacting residue neighbourhood, i. e., evolutionary information in the form of position‐specific‐ scoring matrix (PSSM). The study has been performed on a non‐redundant dataset comprising of 9269 interacting and 91773 non‐interacting residues for prediction model generation and further evaluation. Of the various PSSM‐based models explored, the proposed method named ROBBY (p R ediction O f B iologically relevant small molecule B inding residues on enz Y mes) shows an accuracy of 84.0 %, Matthews Correlation Coefficient of 0.343 and F‐measure of 39.0 % on 78 test enzymes. Further, scope of adding domain knowledge such as pocket information has also been investigated; results showed significant enhancement in method precision. Findings are hoped to boost the reliability of small‐molecule ligand interaction prediction for enzyme applications and drug design.