Identification and removal of sequencing artifacts produced by mispriming during reverse transcription in multiple RNA-seq technologies

Abstract

The quality of RNA sequencing data relies on specific priming by the primer used for reverse transcription (RT-primer). Non-specific annealing of the RT-primer to the RNA template can generate reads with incorrect cDNA ends and can cause misinterpretation of data (RT mispriming). This kind of artifact in RNA-seq based technologies is underappreciated and currently no adequate tools exist to computationally remove them from published datasets. We show that mispriming can occur with as little as 2 bases of complementarity at the 3 end of the primer followed by intermittent regions of complementarity. We also provide a computational pipeline that identifies cDNA reads produced from RT mispriming, allowing users to filter them out from any aligned dataset. Using this analysis pipeline, we identify thousands of mispriming events in a dozen published datasets from diverse technologies including short RNA-seq, total/mRNA-seq, HITS-CLIP and GRO-seq. We further show how RT-mispriming can lead to misinterpretation of data. In addition to providing a solution to computationally remove RT-misprimed reads, we also propose an experimental solution to avoid RT-mispriming by performing RNA-seq using thermostable group II intron derived reverse transcriptase (TGIRT-seq).