Analysis of AlphaMissense data in different protein groups and structural context.

Area of Science:

• Genomics
• Proteomics
• Computational Biology

Background:

• Missense variants, single amino acid substitutions, significantly impact protein function.
• Predicting variant pathogenicity is crucial for therapeutic development and research.
• Experimental validation of variants is often limited, necessitating computational tools.

Purpose of the Study:

• To comprehensively evaluate the performance of AlphaMissense, a missense variant pathogenicity predictor.
• To assess AlphaMissense's efficacy across various protein types and regions.
• To validate AlphaMissense predictions using ClinVar and CFTR2 databases.

Main Methods:

• Assessed AlphaMissense performance on diverse protein groups (soluble, transmembrane, mitochondrial) and regions (intramembrane, membrane-interacting).
• Utilized ClinVar data for initial validation.
• Benchmarked performance against the CFTR2 database for specific protein analysis.

Main Results:

• AlphaMissense demonstrated outstanding performance with Matthews Correlation Coefficient (MCC) scores generally between 0.6 and 0.74.
• Lower performance was observed on disordered protein regions and ClinVar data for the CFTR ABC protein.
• Superior performance was achieved when AlphaMissense was benchmarked against the high-quality CFTR2 database.

Conclusions:

• AlphaMissense is a highly effective tool for predicting missense variant pathogenicity across various protein contexts.
• The predictor shows significant potential for identifying functional hotspots, particularly when validated with curated datasets like CFTR2.
• Further refinement may be needed for disordered regions and specific proteins like CFTR when relying solely on broad databases like ClinVar.