One pipeline to predict them all? On the prediction of alternative splicing from RNA-Seq data

RNA-Seq has become the standard approach to quantify and compare gene expression and alternative splicing in different conditions. In many cases the limiting factor is not the sequencing itself but the bioinformatic analysis. A variety of software tools exist that predict alternative splicing patterns from RNA-Seq data, but surprisingly, a systematic comparison of the predictions obtained from different pipelines has not been performed. Here we compare results from frequently used bioinformatic tools using a high-quality RNA-Seq dataset. We show that there is little overlap in the splicing changes predicted by different tools and that GO-term analysis of the splicing changes predicted by the individual targets yields very different results. Validation of bioinformatic predictions by RT-PCR suggest a high number of false positives in the splicing changes predicated by each pipeline, which probably dominates GO-term analysis. The validation rate is strongly increased for targets predicted by several tools, offering a strategy to reduce false positives. Based on these results we offer some guidelines that may contribute to make alternative splicing predictions more reliable and may thus increase the impact of conclusions drawn from RNA-Seq studies. Furthermore, we created rmappet, a nextflow pipeline that performs alternative splicing analysis using rMATS and Whippet with subsequent overlapping of the results, enabling robust splicing analysis with only one command (https://github.com/didrikolofsson/rmappet/).