Evaluation of the specificity of transcriptional classifiers in larval Pimephales promelas exposed to toxic chemicals with different modes of action.

RNA-seq expression assays were done in larval Pimephales promelas (Fathead minnow, FHM) exposed to either the pyrethroid pesticide Bifenthrin or copper sulfate. Normalized gene expression profiles were used to identify genes differentially expressed between control samples and treated samples, and five machine learning methods were employed to predict the exposure status of the samples in a nested cross-validation framework in a binary manner with each treatment chemical. Classifers trained with one chemical were tested against samples treated with the other chemical to evaluate the chemical specificity of the binary classifiers, and a random forest classifier trained with all three conditions (control, biefnethrin, copper) was also developed and used to predict sample class using a leave-one-out-cross-validation (LOOCV) approach. Functional analysis based on canonical pathways represented by the genes differentially expressed in the two treatment groups was performed to find evidence that toxicant specific reponses were present in the two treatments. Comparative gene expression profiling analysis between controls and treated samples using RNA-seq data from statically exposed larval FHM larvae; untreated controls (n=11), 3.2 ug/L bifenthrin (n=10), or 40 ug/L CuSO4 (n=11).