Noncongeneric Structure—Toxicity Correlation Using Fuzzy Adaptive Least-Squares
IKUO MORIGUCHI, Q. Liu, Hiroyuki Hirano, S. Hirono · ACS symposium series · 1995
Quantitative relationships between chemical structure and toxicity of miscellaneous organic chemicals were studied to generate predictive models for screening compounds for possible environmental risks and pollutant hazards. The toxicity data included the aquatic toxicity of 394 compounds, rodent carcinogenicity and mutagenicity of 246 compounds, and biodegradability of 1259 compounds. The noncongeneric structure-toxicity relationships were analyzed using fuzzy adaptive least-squares (FALS), a pattern recognition method recently developed in our laboratory for structure-activity rating correlation. A novel feature of FALS is that the degree to which each sample belongs to its activity class is given by a fuzzy membership function. The generated QSAR models were statistically reliable in both recognition and leave-one-out prediction despite the diversity and complexity of the structures of the compounds investigated.