Inferring Historical Introgression with Deep Learning.

Area of Science:

• Evolutionary biology
• Genomics
• Bioinformatics

Background:

• Phylogenetic relationship resolution is complex due to genetic admixture across many species.
• Existing genome-wide methods for disentangling evolutionary history are often computationally intensive and require specific data formats or large sample sizes.

Purpose of the Study:

• To develop a novel deep learning-based approach for inferring genome-wide evolutionary relationships and identifying local introgressed regions.
• To provide an efficient and effective tool for analyzing complex genomic data, overcoming limitations of current methods.

Main Methods:

• Developed ERICA (Evolutionary Relationship Inference with Convolutional Autoencoders), a deep learning model accepting sequence alignments.
• ERICA identifies discordant genealogies and introgression patterns across genomes.
• Validated ERICA using population genomic data from Heliconius butterflies and whole genome assemblies.

Main Results:

• ERICA efficiently infers genome-wide evolutionary relationships and pinpoints introgressed genomic segments.
• The method successfully identified hybridization and introgression patterns in Heliconius butterflies.
• Application to rice revealed significant introgression's role in domestication and adaptation.

Conclusions:

• ERICA offers an effective deep learning-based solution for analyzing whole genome data to resolve evolutionary relationships.
• The tool facilitates the study of hybridization and introgression, crucial processes in evolution and adaptation.
• ERICA's application in rice highlights its utility in understanding crop evolution.