Ziyi Mo

The persistence and loss of hard selective sweeps amid admixture in ancient Eurasians

This study uses a domain-adaptive neural network to detect selective sweeps in ancient Eurasian genomes, finding that hard sweeps at loci related to neuronal, reproductive, pigmentation, and signaling traits persisted despite admixture.

Domain-adaptive neural networks improve supervised machine learning based on simulated population genetic data

The paper applies domain adaptation to supervised machine learning models trained on simulated population genetic data to improve performance on real-world data.

SIA: Selection Inference Using the Ancestral Recombination Graph

SIA detects and quantifies positive selection using deep learning on the ancestral recombination graph, showing performance comparable to existing methods and identifying novel selection signals in human and bird populations.