Graph transformer for ancient ancestry inference.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Local ancestry inference identifies DNA segments' origins in admixed individuals.
• Older admixture events result in shorter DNA segments, complicating ancestry determination.
• Ancient DNA offers a valuable resource for improving local ancestry inference.

Purpose of the Study:

• To develop an advanced method for local ancestry inference using ancient DNA.
• To improve the accuracy and robustness of local ancestry inference, especially for older admixture events.
• To leverage deep learning on graph structures for genetic ancestry analysis.

Main Methods:

• Introduced ARGMix, a graph transformer integrating deep learning with the ancestral recombination graph (ARG) framework.
• Utilized ancient DNA samples as references within coalescent trees of the ARG.
• Trained the model on European demographic data and applied it to ancient and present-day European samples.

Main Results:

• ARGMix demonstrates improved accuracy and robustness in local ancestry inference, even with demographic misspecification.
• The method successfully infers local ancestry using ancient DNA as references.
• Analysis revealed genetic continuity between Ötzi the Iceman and present-day European populations.

Conclusions:

• ARGMix represents a significant advancement in local ancestry inference by effectively incorporating ancient DNA data.
• The approach enhances our ability to study population history and genetic continuity over extended periods.
• This method opens new avenues for ancestry-specific analyses using ancient and modern genomic data.