Machine Learning Tool for Detecting Clinically Relevant Splice Variants

A machine learning-based method for predicting the effects of coding and noncoding variants on gene splicing is presented in Genome Biology this week. The accurate prediction of genetic variants that cause or drive disease is a key challenge in genomic medicine. Genome sequencing for rare genetic diseases can provide a diagnosis in up to 60 percent of cases, but current diagnostic rates generally only consider coding, copy number, and canonical splice site variants. Identifying splice-altering variants and interpreting their functional role, therefore, could boost diagnostic yields. To that end, a group led by scientists from the Lowy Cancer Research Centre in Sydney created Introme, an in silico splicing analysis tool that evaluates a variant's likelihood of altering splicing by combining predictions from multiple splice-scoring tools, combined with additional splicing rules, and gene architecture features. Through extensive benchmarking across 21,000 splice-altering variants, the researchers show that Introme outperforms other tools for the detection of clinically significant splice variants.