B.D.: Hazard Assessment Modeling: An Evolutionary Ensemble Approach

This paper presents a novel and eective genetic algorithm approach for generating computational models for hazard assessment. With millions of proposed chemicals being registered each year, it is impossible to come even remotely close to completing the battery of tests needed for the proper understanding of the toxic eects of these chemicals. Com-puter models can give quick, cheap, and en-vironmentally friendly hazard assessments of chemicals. Our approach works by rst ex-tracting a hierarchy of theoretical descriptors of the structure of a compound, then ltering these numerous descriptors with a genetic al-gorithm approach to ensemble feature selec-tion. We tested the utility of our approach by modeling the acute aquatic toxicity (LC50) of a congeneric set of 69 benzene derivatives. Our results demonstrate a very important point: that our method is able to accurately predict toxicity directly from structure.