Accurate prediction of enzyme mutant activity based on a multibody statistical potential

• DOI: 10.1093/bioinformatics/btm509
• Publication Date: 2007-11-01T00:33:58Z
• Authors: Majid Masso; Iosif I. Vaisman
• Keywords: Models, Molecular; 0303 health sciences; Molecular Sequence Data; Enzymes; Pattern Recognition, Automated; Enzyme Activation; Structure-Activity Relationship; 03 medical and health sciences; Amino Acid Substitution; Models, Chemical; Artificial Intelligence; Sequence Analysis, Protein; Data Interpretation, Statistical; Mutation; Mutagenesis, Site-Directed; Computer Simulation; Amino Acid Sequence; Sequence Alignment; Algorithms

An important area of research in biochemistry and molecular biology focuses on characterization enzyme mutants. However, synthesis analysis experimental mutants is time consuming expensive. We describe a machine-learning approach for inferring the activity levels all unexplored single point an enzyme, based training set such with experimentally measured activity.Based Delaunay tessellation-derived four-body statistical potential function, perturbation vector measuring environmental changes relative to wild type (wt) at every residue position uniquely characterizes each mutant model development prediction. First, measure performance utilizing (AUC) under receiver operating characteristic (ROC) curve surpasses 0.83 0.77 data sets HIV-1 protease T4 lysozyme mutants, respectively. Additionally, novel method introduced evaluating significance associated number correct test predictions obtained from trained model. Third, 100 stratified random splits into achieve 77.0% 80.8% mean accuracy, Next, models are used predict remaining mutants; subsequent search publications reporting dozens these reveals that results matched by 79% 86% predictions, Finally, learning curves system indicate influence size performance.Prediction databases http://proteins.gmu.edu/automute/