Machine-learning tool enhances prediction of splice-altering genetic variants

Researchers introduced Introme, a machine-learning tool that integrates multiple-splice prediction tools, additional splicing rules, and gene architecture features to evaluate the likelihood of a variant impacting splicing. The study was done by Patricia Sullivan, Velimir Gayevskiy, and several others and published by BioMed Central (BMC).

In the study, researchers showcased a learning tool called Introme designed to enhance the prediction of genetic variants' impact on splicing. The study states that splicing, the process by which genetic information is cut and rearranged to form mature RNA molecules, plays a role in gene expression. Existing splice prediction tools, the study explains, have limitations and can miss splice-altering variants, particularly in non-canonical splice sites, leading to missed diagnoses in patients. Introme, the researchers posit, offers a solution by integrating predictions from several splice detection tools, incorporating additional splicing rules and considering gene architecture features. According to the study, through extensive benchmarking across 21,000 splice-altering variants, Introme achieved success, outperforming all other tools with an auPRC (area under the precision-recall curve) of 0.98 for detecting clinically significant splice variants.

According to the study, prior to the development of Introme, researchers faced challenges in predicting the impact of genetic variants on splicing, as existing tools had distinct strengths and weaknesses, requiring experts to choose the appropriate tool for each splicing context. Introme overcomes this barrier by providing a comprehensive approach that detects all types of splice variants in coding and noncoding sequences, utilizing a consensus scoring approach that supplements the shortcomings of constituent tools. The researchers show how Introme can evaluate variants at any location within a gene, from exonic to deep intronic regions, ensuring a more holistic assessment, and point out that the tool has been used by clinicians to make diagnoses, after attempting to utilize existing strategies. The researchers explain how Introme's versatility and "accuracy" make it an important tool for identifying splice-disrupting genetic variants in various clinical settings, from childhood cancer to rare genetic diseases.

While Introme represents a leap forward, the study highlights the need for further research in predicting the functional impact of splice-altering variants and improving methods for identifying splicing regulatory variants. The field, it explains, still lacks sufficient data, linking genetic variants to functional splice alterations and quantitative measures, like the percent spliced in (PSI) value. Initiatives like SpliceVault aim to address this knowledge gap by shedding light on naturally occurring atypical splice site usage. The research argues that Introme offers an avenue for the accurate prediction of splice-altering variants and holds the potential to advance genetic research and diagnostics.

Springer Nature: Patricia Sullivan, Velimir Gayevskiy, et al., Introme accurately predicts the impact of coding and noncoding variants on gene splicing, with clinical applications, Genome Biology (2023). https://doi.org/10.1186/s13059-023-02936-7