You are here:
Development of Classifiers of Exposure Employing Fathead minnow, including Evaluation of an Alternative, Lower Cost, RNA-seq Library Preparation Method
Citation:
Martinson, J., D. Bencic, R. Flick, M. Kostich, W. Huang, D. Lattier, AND A. Biales. Development of Classifiers of Exposure Employing Fathead minnow, including Evaluation of an Alternative, Lower Cost, RNA-seq Library Preparation Method. 2019 SETAC North America Annual Meeting, Toronto, CANADA, November 03 - 07, 2019.
Impact/Purpose:
This presentation will present the results of this experiment, which will answer several important questions, including: Can the new Fathead minnow gene models be successfully employed to develop classifiers of exposure? How does sequencing depth affect classifier performance? How does sample replication affect classifier performance? Does one preparation method offer the potential for significant cost savings relative to the other?
Description:
Developing screening assays for chemical exposure is important to protecting the nation’s waterways. Fathead minnow (FHM) exposure experiments that measure gene expression can be a key tool in developing screening assays for biologically relevant exposures. Due in a large part to the decreasing costs associated with sequencing, RNA-seq has become the preferred tool for expression analysis, and the newly produced Fathead gene models provide a robust framework for organizing and interpreting RNA-seq expression data. Cost remains an issue though, especially if one expects to produce conclusive results, so any tool that can reduce the cost-per-sample has obvious value. This presentation will demonstrate the utility of the new FHM gene models as a platform for mRNA expression analysis and will also compare results derived from two different RNA-seq library preparation methods at different RNA-sequencing depths and different levels of replication. One of the two preparation methods has the potential to provide significant cost savings compared to the other. Thirty-two four-day-post-fertilization Fatheads were exposed for 48 hours to either copper (n=11), biphenthrin (n=11), or were not exposed (negative controls, n=10). Following exposure, RNA-seq libraries were prepared from the 32 individuals using two alternative methods (kits) from the same manufacturer. One kit was the “standard” kit that can provide sequence from any part of the target cDNA, while the second kit specifically targets only the 3'-end of target sequences. The goal of the latter more targeted approach is to achieve deeper sequencing depth of any given transcript, which can allow more multiplexing of samples, therefore reducing costs. Our lab estimated that if the approach worked as the manufacturer claimed we should be able to reduce our price per sample by about 50%. To fairly evaluate the performance of the two preparation methods, we used the same samples, the same amount of total RNA, tagged the samples with the same adaptor sequences, and had the samples sequenced in the same lane orientation (two lanes with 16 samples in each for each kit). Following sequencing, reads were resampled at various depths (ranging from 2-~20M reads per sample). Resampled reads were mapped to the new FHM gene models. Sets of normalized read counts per gene were reduced using a linear model to identify the most informative genes (features), which were used to develop classifiers of exposure using randomForest, employing ten-fold cross-validation to estimate performance (Brier scores).