dnadna

dnadna provides a Python framework for applying and developing deep-learning neural networks to infer population genetic parameters and demographic history from SNP and simulated genetic datasets.

Key Features:

• Workflows and utilities: Provides modular workflows, utility functions, and implementation scaffolds for developing new neural network architectures and tasks.
• Standardized training and testing: Includes standardized training procedures and test environments to run and evaluate models reproducibly.
• Re-optimization and retraining: Supports re-optimizing and retraining implemented networks on user-provided training sets or task-specific data.
• Pre-trained network application: Enables applying pre-trained networks to predict evolutionary history from real and simulated genetic datasets.
• Peer-reviewed demographic inference network: Ships a peer-reviewed neural network specifically for demographic inference from SNP data.
• Toy networks for exploration: Provides toy networks to explore model behavior and validate implementation choices.
• Reproducibility and reusability: Emphasizes reproducible model definitions and reusable components for comparative studies and benchmarking.

Scientific Applications:

• Demographic inference: Infers demographic history and related parameters from SNP datasets using neural-network approaches.
• Population genetic parameter estimation: Estimates population genetic parameters from real and simulated genetic data via deep learning.
• Model development and benchmarking: Facilitates development, evaluation, and benchmarking of custom and pre-trained neural models for evolutionary inference.

Methodology:

Implemented as a Python package (requires Python 3.7+), dnadna uses deep-learning neural networks with standardized training procedures and testing environments, supports re-optimization/retraining of models, and applies pre-trained networks to SNP and simulated genetic datasets.