Zygosity-Aware DNA Language Modeling Improves Ancestry and Gene Expression Prediction

This repository provides .pkl files containing HyenaDNA-derived embeddings for the MHC region of chromosome 6 in 2,584 individuals from the 1000 Genomes Project. Embeddings are provided for three aggregation strategies (mean, max, and last pooling). Associated files include .csv and .npy data for the XGBoost ancestry classification results using stratified 5-fold cross-validation; these files reproduce all figures in the main paper using analysis code from our GitHub repository. Additionally, text files listing the genes in each 450 kb MHC window are included, supporting the finding that windows with highest classification performance are enriched in HLA and other immunity-related genes.