Inference of Population Structure from Time-Series Genotype Data

Area of Science:

• Genetics
• Bioinformatics
• Population History

Background:

• Analyzing ancient DNA is crucial for understanding population history.
• Current tools often assume contemporary samples, limiting their effectiveness with temporally diverse data.

Purpose of the Study:

• To introduce DyStruct, a novel model and inference algorithm for analyzing temporally sampled genotype data.
• To improve the accuracy of inferring shared ancestry and population dynamics from ancient DNA.

Main Methods:

• DyStruct models individuals as mixtures of unobserved populations with allele frequencies drifting over time.
• An efficient inference algorithm using stochastic variational inference was developed.
• The model was tested on simulated and real-world ancient and modern DNA datasets.

Main Results:

• DyStruct demonstrated superior performance compared to state-of-the-art methods on temporally sampled data.
• The model successfully identified a steppe ancestry admixture event in modern European populations.
• Application to Near Eastern farming origins provided new population history insights.

Conclusions:

• DyStruct offers a powerful new approach for analyzing ancient DNA, explicitly accounting for temporal dynamics.
• The model enhances our ability to reconstruct complex population histories and admixture events.
• DyStruct provides accurate insights within feasible computational time.