Inverse QSAR Study Using Evolutionary Algorithm

Abstract

Quantitative Structure-Activity Relationship (QSAR) has been widely used in molecular design and many successful applications have been known. However, when precise model would be made, its chemical interpretation becomes difficult. Therefore, we have developed inverse QSAR system to generate chemical structures with high inhibitory activities. EA-Inventor (Evolutionary Algorithm-Inventor) was used as structure generator in our system. In EA-Inventor, initial structures represented by SMILES strings are modified using cross-over and mutation operations. The inhibitory activities are predicted by QSAR model and their values are used as the scores. The structures with high scores survive and next new structures are generated. After these cycles, the higher scored structures can be obtained. We have applied EA-Inventor to Trypsin inhibitory data. CoMFA (comparative molecular field analysis) is employed as QSAR model. If new structures differ from training set, high penalty values are added to scores in order to guarantee them within chemical space. The generated structures are high complementarily to Trypsin structure and the result seems to be reasonable. This technology can be extended to ADME model and chemical feasibility filter.