AlphaMissenseR

AlphaMissenseR facilitates performant and reproducible access to AlphaMissense pathogenicity predictions for missense mutations in human protein-coding genes and provides functionality for analysis, visualization, validation, and benchmarking.

Key Features:

• Data Access: Downloads individual AlphaMissense files to DuckDB databases for exploration and integration into R and Bioconductor workflows.
• Genomic Region Analysis: Integrates with Bioconductor facilities for genomic region analysis.
• Multi-level Visualization: Provides visualization and interactive exploration of variant pathogenicity in a genome browser and on 3D protein structures.
• Database Integration: Integrates with major clinical and experimental variant databases for contrasting predicted and clinically derived pathogenicity scores.
• Benchmarking: Enables systematic benchmarking of existing and new variant effect prediction methods across a large collection of deep mutational scanning assays.

Scientific Applications:

• Missense Variant Pathogenicity Analysis: Investigates missense mutations in human protein-coding genes using AlphaMissense predictions.
• Clinical Variant Interpretation: Contrasts predicted pathogenicity scores with clinically derived scores from major clinical variant databases.
• Variant Effect Prediction Benchmarking: Systematically benchmarks existing and new variant effect prediction methods against deep mutational scanning assays.
• Protein Structure-Function Exploration: Interactively explores variant pathogenicity on 3D protein structures.

Methodology:

AlphaMissenseR leverages AlphaMissense, an AI model from Google DeepMind based on a variant of AlphaFold, which predicts the pathogenicity of every possible missense mutation in the human proteome. The package downloads individual data files to DuckDB databases for local access and analysis.