Impact of bias correction methods and increasing the biological samples in transcriptomic analysis

RNA-sequencing (RNA-Seq) technologies enable the quantification of gene expression levels, identification of splice junctions, and uncovering of novel transcripts. Comparing the number of sequences (reads) that map to transcripts under different conditions (e.g. genotypes) can detect differentially expressed genes and transcripts. Typically, RNA-Seq experiments have small sample sizes, making accurate detection of differentially abundant transcripts (DATs) challenging under these conditions. Normalization, bias adjustment and modeling of the RNASeq count data can further influence the detection and ranking of DATs. The objectives of this study were to assess the impact of sample size, normalization, and bias correction on the detection, ranking and functional characterization of DATs. RNA-Seq data from the brain macrophages of wild-type and indoleamine 2,3- dioxygenase 1-knockout mice was used. Data subsets (4 and 6 samples per group) and a range of bias adjustments were tested using TopHat and Cufflink routines. Prolactin variant 1 and growth hormone DATs were detected in both 4 and 6 group sample sizes. Average DAT number across sample sizes was 144.5 and 79, respectively. Bias corrections affected the estimate precision, resulting in reranking of the DATs. Despite the additional DATs identified using bias adjustments, functional clustering remained stable. Identification of robust DAT sets requires the evaluation of complementary bias-correction approaches.