Accurate proteome-wide missense variant effect prediction with AlphaMissense

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Most human missense variants have unknown clinical significance, hindering genetic disease research.
• Accurate prediction of variant pathogenicity is crucial for clinical interpretation and understanding genetic variation.

Purpose of the Study:

• To develop and validate AlphaMissense, a novel computational tool for predicting missense variant pathogenicity.
• To create a comprehensive database of variant predictions for the human genome.
• To explore the relationship between variant pathogenicity and gene essentiality.

Main Methods:

• AlphaMissense, an adaptation of AlphaFold, was fine-tuned using human and primate variant population frequency databases.
• The model integrates structural context and evolutionary conservation to predict pathogenicity.
• Performance was evaluated against diverse genetic and experimental benchmarks.

Main Results:

• AlphaMissense achieves state-of-the-art performance in predicting missense variant pathogenicity without explicit training on benchmark data.
• The average pathogenicity score of genes predicts cell essentiality, identifying essential genes missed by other methods.
• A database of predictions for all possible human single amino acid substitutions is provided.

Conclusions:

• AlphaMissense effectively predicts missense variant pathogenicity, classifying 89% of variants as likely benign or pathogenic.
• The tool offers a valuable resource for the scientific community, advancing the interpretation of genetic variants.
• Predictive pathogenicity scores enhance the understanding of gene essentiality and cellular functions.