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

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Accurate identification of somatic insertions/deletions (indels) from RNA sequencing (RNA-Seq) data is challenging due to PCR amplification artifacts and the absence of matched normal samples.
• Existing methods struggle to reliably distinguish true somatic variants from technical noise in tumor transcriptomes.

Purpose of the Study:

• To develop and validate a computational tool, RNAIndel, for the accurate prediction of somatic, germline, and artifactual indels from tumor RNA-Seq data.
• To improve the detection of clinically relevant somatic indels, including low-frequency subclonal variants.

Main Methods:

• RNAIndel employs a machine-learning framework utilizing features from indel sequence context and predicted biological impact.
• The tool was evaluated on five diverse cancer datasets (pediatric and adult) to assess its predictive performance.

Main Results:

• RNAIndel achieved robust prediction of somatic indels, with 88-100% accuracy across various cancer types, excluding those with microsatellite instability.
• The tool successfully identified subclonal driver indels (Variant Allele Frequency range 0.01–0.15) that were missed by targeted deep sequencing.
• Compared to current best-practice RNA-Seq variant calling, RNAIndel demonstrated higher sensitivity (88-100% vs. 57%) and significantly fewer false positives (14x reduction).

Conclusions:

• RNAIndel provides a reliable solution for identifying somatic indels in tumor RNA-Seq data, overcoming limitations of existing methods.
• The tool enhances the discovery of critical somatic mutations, including subclonal variants, which are important for understanding cancer evolution and developing targeted therapies.
• RNAIndel is freely available, facilitating its adoption in cancer genomics research.