Distinguishing between recent balancing selection and incomplete sweep using deep neural networks.

Area of Science:

• Population genetics
• Genomics
• Bioinformatics

Background:

• Balancing selection drives diverse phenotypes but is difficult to detect in genomic data due to signature overlap with positive selection.
• Distinguishing recent balancing selection from incomplete sweeps is a persistent challenge in evolutionary genetics.

Purpose of the Study:

• To develop and evaluate deep neural networks for identifying loci under recent balancing selection or incomplete sweeps.
• To compare the performance of artificial neural networks (ANN) and convolutional neural networks (CNN) in detecting selection signatures.

Main Methods:

• Generated forward-in-time simulations to train and test ANN and CNN models.
• ANN utilized summary statistics, while CNN processed haplotype data directly.
• Deployed trained networks on simulated and real population genomic data (European populations).

Main Results:

• Both ANN and CNN demonstrated high accuracy in predicting loci under recent selection.
• CNN generally outperformed ANN, particularly in differentiating balancing selection from incomplete sweeps and reducing false positives.
• Analysis of the MEFV gene region identified functional variants under incomplete sweep, potentially linked to adaptation against pathogens.

Conclusions:

• Deep neural networks offer a powerful new approach to characterizing selection signals on intermediate frequency variants.
• CNNs provide a robust method for distinguishing complex selection signatures, overcoming limitations of traditional strategies.
• The findings have implications for understanding human adaptation and disease susceptibility.