In Vitro Biodescriptors Derived from Time Series Toxicogenomics Data in In Silico QSAR Improve the Phenotypic Toxicity Prediction
Sheikh Mokhlesur Rahman, Jiaqi Lan, Na Gou, Akram N. Alshawabkeh, April Z. Gu · ACS ES&T Water · 2023
In this study, we explored the bioassay-based Quantitative Biological Activity Relationship (QBAR) model with in vitro high-throughput screening-based temporal transcriptomic and proteomic data of known stress response pathways, as the biological descriptors, to predict in vivo toxicity. We, for the first time, constructed and compared QBAR models with quantitative biodescriptors derived at three biologically relevant levels─individual biomarker, specific pathway, and whole cellular level. For all the models, two separate case studies were conducted including cytotoxicity ( Escherichia coli half-maximal effective concentration) prediction from an E. coli- based transcriptomic assay and genotoxicity ( in vivo carcinogenicity) prediction from a yeast-based proteomic assay. For both case studies, QBAR models using biodescriptors perform better (with higher R 2 and accuracy values) than the conventional QSAR models or the Quantitative Structure and Biological Activity Relationship (QSBAR) models with combined physical-chemical and biological descriptors. Among the three biological levels for biodescriptors derivation, the QBAR model with descriptors indicative of the pathway-level molecular disturbance led to the best prediction ( R 2 = 0.78 for cytotoxicity prediction; accuracy = 75% for carcinogenicity prediction). In addition, the QBAR model identified top ranked biodescriptors that contributed significantly to the toxicity prediction, thus providing insights into mode of action. Applicability domain analysis reveals that the derived QBAR models can produce reliable predictions using the biological descriptors.