Performance evaluation of pathogenicity-computation methods for missense variants.

Area of Science:

• Medical Genetics
• Computational Biology
• Bioinformatics

Background:

• Next-generation sequencing (NGS) applications in medical genetics necessitate accurate computational prediction of missense variant pathogenicity.
• Numerous computational methods exist, but their performance across varied conditions remains unevaluated, hindering optimal tool selection for gene identification.

Purpose of the Study:

• To comprehensively evaluate and compare the performance of 23 computational methods for predicting missense variant pathogenicity.
• To identify the most effective methods for different types of genetic variants and datasets.

Main Methods:

• Comparison of 12 performance measures for 23 prediction methods.
• Utilized three independent benchmark datasets: clinical variants (ClinVar), cancer somatic variants (IARC TP53, ICGC), and experimentally validated PPARG variants.
• Evaluated method performance on de novo mutations.

Main Results:

• Method performance varied across different datasets and conditions, indicating limited generalizability for some tools.
• Most methods exhibited lower specificity than sensitivity, especially on experimental datasets, suggesting a need for stricter thresholds.
• REVEL, VEST3, and their combined approach (ReVe) consistently showed the best overall performance across all benchmark datasets.
• ReVe demonstrated the highest performance for de novo mutations.

Conclusions:

• The study provides a comparative analysis of computational pathogenicity prediction tools under various conditions.
• REVEL, VEST3, and ReVe are recommended as top-performing methods for missense variant pathogenicity prediction.
• Guidance is offered for selecting optimal computational tools in medical genetics research.