dnadna: a deep learning framework for population genetics inference.

Area of Science:

• Population genetics
• Computational biology
• Bioinformatics

Background:

• Deep learning is increasingly applied to complex biological data.
• Population genetics involves analyzing genetic variation within and between populations.
• Developing and sharing reusable deep learning tools for population genetics is challenging.

Purpose of the Study:

• Introduce dnadna, a flexible Python software for deep learning inference in population genetics.
• Facilitate the development, reproducibility, dissemination, and re-usability of neural networks for population genetic data.
• Enable researchers to perform demographic inference and other analyses without extensive deep learning expertise.

Main Methods:

• dnadna provides user-friendly workflows for implementing and optimizing neural network architectures.
• The software includes utility functions, a standardized training procedure, and a testing environment.
• Pre-trained networks and new architectures can be shared within the community.

Main Results:

• dnadna supports task-agnostic development, allowing adaptation to various population genetics problems.
• Users can apply pre-trained networks for tasks like demographic inference from SNP data.
• The software includes a peer-reviewed, exchangeable neural network and toy networks for ease of use.

Conclusions:

• dnadna lowers the barrier to entry for applying deep learning in population genetics.
• It promotes reproducible research and facilitates the sharing of novel deep learning approaches.
• The software is extensible and encourages community contributions for expanding its capabilities.