A systematic evaluation of the sources of non-uniform coverage in RNA-Seq and limitations of current practices.
TL;DR
It is demonstrated that no single factor can explain the non-uniformity in coverage across transcripts in RNA-Seq and that a connection of coverage with PCR that depends more on molecular abundance than PCR cycle number is demonstrated.
Abstract
RNA splicing is the focus of a vast array of studies in biology, as it is involved in many normal cellular functions and related abnormalities. RNA-Seq remains the standard quantitative assay to interrogate splicing. The primary technical artifact of RNA-Seq library prep is the non-uniformity in coverage across transcripts. This undermines the accuracy of isoform-level quantification. This non-uniformity is present in both bulk and single-cell RNA-Seq and is purely artifactual as it is observed even when the sample contains only full-length transcripts. Therefore, understanding the source of this non-uniformity is of high importance. Here, we clarify the state of current knowledge and investigate the possible contributing factors. We demonstrate that no single factor can explain the phenomenon. Several approaches have previously been reported to improve uniformity, including reducing PCR ramp rates or cycle numbers, avoiding ribosomal depletion, using shorter fragment lengths, and using thermostable reverse transcriptases. We investigate all of these and demonstrate that none of these produce substantial improvements. We further investigate the possible sources of nonuniformity by assessing existing data sets with varying sample quality, PCR cycle number, reverse transcriptase, and technical or biological replicates. In addition, we generated novel data sets by varying PCR ramp rate, fragment length, and ribosomal depletion method. We demonstrate a connection of coverage with PCR that depends more on molecular abundance than PCR cycle number.