MutAnt: mutation annotation tool predicts deleteriousness of missense mutations and improves mutation calling from...

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Clinical Genomics

Background:

• Pathogenic variants can cause Mendelian diseases or cancer through loss-of-function, gain-of-function, or other mechanisms.
• Interpreting rare and novel variants is challenging, necessitating computational tools for distinguishing deleterious mutations from neutral variation.
• Accurate variant classification is crucial for clinical genomics and understanding disease mechanisms.

Purpose of the Study:

• To develop and evaluate MutAnt, a machine-learning-based mutation meta-annotator.
• To assess MutAnt's ability to distinguish deleterious from neutral variants using diverse variant properties.
• To explore MutAnt's utility in improving somatic variant calling and its correlation with functional assays.

Main Methods:

• Developed MutAnt, a machine-learning model trained on a large, clinically relevant variant dataset.
• Incorporated multiple variant properties and synchronized predictions from other algorithms into MutAnt's training.
• Evaluated MutAnt's performance using F1 and ROC-AUC scores, correlation with functional assays (deep mutational scanning), and protein stability measurements.

Main Results:

• MutAnt achieved high F1 and ROC-AUC scores (0.88-0.99) on hold-out datasets.
• MutAnt's deleteriousness predictions correlated with functional scores for BRCA1, PTEN, and p53 (ρ = 0.28-0.61) and protein stability.
• MutAnt improved somatic variant calling from RNA sequencing data compared to standard methods.

Conclusions:

• MutAnt demonstrates high performance in distinguishing neutral from protein-disrupting mutations.
• The tool provides well-calibrated probability scores that correlate with experimental functional data.
• MutAnt shows significant potential for clinical utility in variant classification and genomic analysis.