Refinement of the clinical variant interpretation framework by statistical evidence and machine learning

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• The American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP) guidelines are widely used for clinical variant interpretation.
• Knowledge-based guidelines like ACMG/AMP can benefit from data-driven improvements for enhanced accuracy.

Purpose of the Study:

• To statistically assess the deleteriousness of start-lost, stop-lost, and in-frame indel variants.
• To develop and validate a machine learning model for predicting the pathogenicity of start-lost variants.
• To propose refinements to the ACMG/AMP guidelines based on data-driven analysis.

Main Methods:

• Statistical analysis of variant deleteriousness using Genome Aggregation Database (gnomAD) population data.
• Development of the PoStaL model, a machine learning approach predicting pathogenicity of start-lost variants.
• Deep learning for predicting translation initiation sites and random forest training on known pathogenic variants.

Main Results:

• Stop-lost variants showed the highest proportion of rare variants, followed by in-frame indels and start-lost variants.
• Current ACMG/AMP guideline classifications for start-lost (PVS) and stop-lost/in-frame indel (PM) variants may require revision.
• The PoStaL model demonstrated superior performance in assessing start-lost variant pathogenicity compared to existing tools.

Conclusions:

• The study offers data-driven insights to refine the ACMG/AMP guidelines for variant interpretation.
• Developed resources, including the PoStaL model and variant lists, aid future research in clinical genetics.
• This work exemplifies improving knowledge-based frameworks through data-driven methodologies.