Open AccessEnsemble QSAR Modeling to Predict Multispecies Fish Toxicity Lethal Concentrations and Points of Departure
Author(s) -
Thomas Sheffield,
Richard S. Judson
Publication year - 2019
Publication title -
environmental science and technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.851
H-Index - 397
eISSN - 1520-5851
pISSN - 0013-936X
DOI - 10.1021/acs.est.9b03957
Subject(s) - quantitative structure–activity relationship , random forest , applicability domain , generalizability theory , test set , support vector machine , data set , regression , chemical toxicity , aquatic toxicology , machine learning , computer science , statistics , chemistry , toxicity , artificial intelligence , mathematics , organic chemistry
QSAR modeling can be used to aid testing prioritization of the thousands of chemical substances for which no ecological toxicity data are available. We drew on the U.S. Environmental Protection Agency's ECOTOX database with additional data from ECHA to build a large data set containing in vivo test data on fish for thousands of chemical substances. This was used to create QSAR models to predict two types of end points: acute LC 50 (median lethal concentration) and points of departure similar to the NOEC (no observed effect concentration) for any duration (named the "LC 50 " and "NOEC" models, respectively). These models used study covariates, such as species and exposure route, as features to facilitate the simultaneous use of varied data types. A novel method of substituting taxonomy groups for species dummy variables was introduced to maximize generalizability to different species. A stacked ensemble of three machine learning methods-random forest, gradient boosted trees, and support vector regression-was implemented to best make use of a large data set with many descriptors. The LC 50 and NOEC models predicted end points within 1 order of magnitude 81% and 76% of the time, respectively, and had RMSEs of roughly 0.83 and 0.98 log 10 (mg/L), respectively. Benchmarks against the existing TEST and ECOSAR tools suggest improved prediction accuracy.