RNA-Seq Quantification Methods Differ Dramatically on Equine RNA-Seq Datasets

RNA-seq has recently emerged as the standard method for analyzing the transcriptome across species. There are many existing bioinformatics approaches to RNA-seq quantification, i.e. the conversion of raw sequencing reads into estimates of gene expression.  Reads that map to multiple locations are handled and reported differently by different software and can be a major source of error in RNA-seq quantification.  We utilities four distinct algorithms (1) raw counts with HTSeq, 2) raw counts with subread 3) FPKM with cufflinks 4) raw counts and transcripts per million (TPM) with kallisto to quantify RNA-seq reads on two equine RNA-seq datasets (16 spinal cord r-RNA depleted libraries and 12 muscle poly-A⁺ selected libraries) using custom and publically-available gtf files. While HTseq and subread demonstrated high concordance, the majority of differentially expressed transcripts (DETs), as validated by q-RT-PCR, were not counted accurately with HTSeq or subread. For raw counts, only the count option in kallisto provided accurate counts to identify DETs. Both cufflinks and the TPM estimation in kallisto provided accurate quantification on both equine datasets. In conclusion, similar to results found in other species, the method of RNA-seq quantification may dramatically affect downstream differential gene expression analyses of equine RNA-seq datasets.