LEAP: Using machine learning to support variant classification in a clinical setting.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Genome sequencing advances increase novel missense variant discovery.
• Determining clinical significance of these variants is challenging due to limited or conflicting evidence.

Purpose of the Study:

• To present the Learning from Evidence to Assess Pathogenicity (LEAP) machine learning model.
• To evaluate LEAP's performance in classifying missense variants for clinical significance.

Main Methods:

• LEAP utilizes diverse feature categories: functional predictions, splice predictions, population frequencies, conservation scores, protein domain data, and clinical observations.
• L2-regularized logistic regression and random forest models were trained on clinical missense variants from cancer and cardiovascular genes.
• 10-fold cross-validation and validation on withheld genes were employed.

Main Results:

• Logistic regression achieved 97.8% AUROC (cancer) and 98.8% AUROC (cardiovascular).
• Random forest achieved 98.3% AUROC (cancer) and 98.6% AUROC (cardiovascular).
• Validation on withheld genes showed 96.8% AUROC, demonstrating generalizability.

Conclusions:

• LEAP demonstrates high accuracy and broad applicability in classifying missense variants.
• The model is effective as a high-throughput quality control layer in clinical settings.
• LEAP aids in interpreting genetic variants for improved clinical decision-making.