MuRaL

MuRaL predicts fine-scale germline nucleotide mutation rates from genomic sequence using deep learning to support accurate mutation-rate mapping and downstream functional analyses.

Key Features:

• Deep Learning Approach: MuRaL employs a deep learning model to predict nucleotide-level mutation rates using only genomic sequence as input.
• Efficiency with Limited Data: The framework can train effective models with relatively few observed training mutations and a moderate number of sequenced individuals.
• Transfer Learning Capability: MuRaL leverages transfer learning to reduce data and computational requirements and to adapt models across species.

Scientific Applications:

• Genome-Wide Mutation Rate Mapping: MuRaL has been applied to generate genome-wide mutation rate maps for Homo sapiens, Macaca mulatta, Arabidopsis thaliana, and Drosophila melanogaster.
• Functional Stratification of Genes: Improved mutation-rate estimates enable stratification of human genes into functionally enriched groups, revealing that many developmental genes carry a high mutational burden.

Methodology:

MuRaL trains deep learning models on genomic sequence and human germline variants, performs model training and validation, and applies transfer learning for cross-species adaptation.