Haplotype and population structure inference using neural networks in whole-genome sequencing data

Area of Science:

• Population Genetics
• Computational Biology
• Genomics

Background:

• Accurate population structure inference is crucial for population genetics studies.
• Existing methods may struggle with distinguishing closely related populations.

Purpose of the Study:

• Introduce HaploNet, a novel method for dimensionality reduction and clustering of genetic data.
• Enhance the accuracy and scalability of population structure inference.

Main Methods:

• Utilizes local clustering of phased haplotypes with neural networks.
• Employs Gaussian mixtures within a variational autoencoder for a low-dimensional latent space.
• Leverages generative properties for global population structure inference.

Main Results:

• HaploNet effectively clusters haplotypes in a scalable latent space.
• The method successfully infers global population structure and ancestry proportions.
• Demonstrates superior performance in distinguishing closely related populations compared to standard software.
• Validated on simulations, human population sequencing data, and UK Biobank genotype data.

Conclusions:

• HaploNet provides a fast, scalable, and accurate approach for population genetics analysis.
• The method enhances the ability to infer population structure from genetic data.
• Offers a powerful alternative to traditional methods for population genetics research.