MuRaL

KO | EN

gitlite — search

#python #react #machine-learning #rust #docker #macos #nodejs #java #windows #awesome #cli #kubernetes #vue #mcp

(★ 21)

MuRaL is a deep learning framework for building base-resolution mutation rate maps.

MuRaL Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

## MuRaL: inferring genomic mutation rate maps via deep learning
1. [Introduction](#Introduction)
2. [Installation and usage](#Usage)
3. [Pre-trained models and predicted mutation rate maps](#Trained_models)
4. [Citation](#Citation)
5. [Contact](#Contact)

## 1. Introduction <a name="Introduction"></a>
Germline mutation rates are important in genetics, genomics and evolutionary biology. It is long known that mutation rates vary substantially across the genome, but existing methods can only obtain very rough estimates of local mutation rates and are difficult to be applied to non-model species.

**MuRaL**, short for **Mu**tation **Ra**te **L**earner, is a generalizable deep learning framework to estimate base-resolution mutation rates maps.  We have developed two specialized tools for inferring mutation rates: **MuRaL-snv** (Fang et al., 2022) for single nucleotide variants (SNVs) and **MuRaL-indel** (Deng et al., 2025) for short insertions and deletions (INDELs). MuRaL has better predictive performance at different scales than current state-of-the-art methods. Moreover, it can generate genome-wide mutation rate maps with rare variants from a moderate number of sequenced individuals, and can leverage transfer learning to further reduce data and time requirements. It can be applied to many sequenced species with population polymorphism data.

MuRaL supports **recurrent mutation** analysis, where the same genomic site may be mutated independently in multiple individuals. The ``--recurrent`` flag enables count-aware training (per-site weights in loss), prediction, and evaluations.

The MuRaL-snv network architecture shown below has two main modules, one is for learning signals from local genomic regions (e.g. 10bp on each side of the focal nucleotide), the other for learning signals from expanded regions (e.g. 1Kb on each side of the focal nucleotide).

Below is an example showing that MuRaL-predicted rates (colored lines) are highly correlated with observed mutation rates (grey shades) at different scales on Chr3 of *A. thaliana*.

### 2. Installation and usage <a name="Usage"></a> 
For detailed information about installation and usage, please go to [MuRaL documentation site](https://mural.readthedocs.io).

### 3. Trained models and predicted mutation rate maps of multiple species <a name="Trained_models"></a>
Trained models for four species - ***Homo sapiens***, ***Macaca mulatta***, ***Arabidopsis thaliana*** and ***Drosophila melanogaster*** are provided in the 'models/' folder of the package. One can use these model files for prediction or transfer learning.
 
Predicted single-nucleotide mutation rate maps for these genomes are available at [ScienceDB](https://www.doi.org/10.11922/sciencedb.01173).

Predicted indel mutation rate maps for these genomes are available at [ScienceDB](https://doi.org/10.57760/sciencedb.30830).

### 4. Citation <a name="Citation"></a>
Fang Y, Deng S, Li C. A generalizable deep learning framework for inferring fine-scale germline mutation rate maps. *Nature Machine Intelligence* (2022) [doi:10.1038/s42256-022-00574-5](https://doi.org/10.1038/s42256-022-00574-5)

Deng S, Song H, Li C. A deep learning framework for building INDEL mutation rate maps. *Briefings in Bioinformatics* (2025) [doi:10.1093/bib/bbag212](https://doi.org/10.1093/bib/bbag212)

### 5. Contact <a name="Contact"></a>
For reporting issues or requests related to the package, please use GitHub Issues or write to mural-project@outlook.com.

MuRaL

MuRaL Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

MuRaL

MuRaL Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README