Variant effect predictor correlation with functional assays is reflective of clinical classification performance

Area of Science:

• Genomics
• Computational Biology
• Biochemistry

Background:

• Accurate classification of missense variants relies on understanding protein sequence-function relationships.
• Variant effect predictors (VEPs) are crucial but challenging to evaluate due to data circularity.
• Deep mutational scanning (DMS) offers a robust experimental approach for VEP assessment.

Purpose of the Study:

• To benchmark the performance of 97 VEPs using missense DMS data from 36 human proteins.
• To introduce a VEP-centric pairwise approach to mitigate missing prediction impacts.
• To evaluate the correlation between VEP performance on DMS data and clinical variant classification.

Main Methods:

• Benchmarking 97 VEPs against missense DMS measurements across 36 human proteins.
• Employing a novel pairwise, VEP-centric comparison method.
• Analyzing the concordance between VEP performance in functional assays and clinical variant interpretation.

Main Results:

• A strong correlation was observed between VEP performance on DMS benchmarks and clinical variant classification.
• Predictors not trained on human clinical variants showed notable correspondence.
• The pairwise VEP-centric approach effectively handled missing predictions.

Conclusions:

• Comparing VEPs against diverse functional assays (like DMS) is a reliable strategy for assessing clinical variant classification performance.
• Significant challenges remain in the clinical interpretation of VEP scores, necessitating further research.
• Practical guidance is provided for end-users on selecting appropriate VEP methodologies for genetic diagnosis.