Fuzzy Adaptive Least Squares Applied to Structure-Activity and Structure-Toxicity Correlations

Abstract

A 1991 version of fuzzy adaptive least squares (FALS91) has been developed to analyze structure-activity rating data for generation of QSAR (quantitative structure-activity relationship) models. Improvements were made in this version especially to the error-correcting feedback adaptation of discriminant functions to avoid falling into a local optimum. This paper showed effective applications of FALS91 to structure-activity correlation of 29 antihypertensive arylacryloylpiperazine derivatives and structure-toxicity correlation of human acute toxicity of 504 organic chemicals as typical examples of small- and large-scale data sets, respectively. These examples demonstrated the high reliability of FALS91 in both recognition and leave-one-out prediction.