RNAIndel: discovering somatic coding indels from tumor RNA-Seq data

Abstract

Abstract Motivation Reliable identification of expressed somatic insertions/deletions (indels) is an unmet need due to artifacts generated in PCR-based RNA-Seq library preparation and the lack of normal RNA-Seq data, presenting analytical challenges for discovery of somatic indels in tumor transcriptome. Results We present RNAIndel, a tool for predicting somatic, germline and artifact indels from tumor RNA-Seq data. RNAIndel leverages features derived from indel sequence context and biological effect in a machine-learning framework. Except for tumor samples with microsatellite instability, RNAIndel robustly predicts 88–100% of somatic indels in five diverse test datasets of pediatric and adult cancers, even recovering subclonal (VAF range 0.01–0.15) driver indels missed by targeted deep-sequencing, outperforming the current best-practice for RNA-Seq variant calling which had 57% sensitivity but with 14 times more false positives. Availability and implementation RNAIndel is freely available at https://github.com/stjude/RNAIndel. Supplementary information Supplementary data are available at Bioinformatics online.