Diet Network: Trained models and resources for genetic ancestry prediction (1KGP, 24 populations)

This Zenodo record accompanies the paper A Transparent and Generalizable Deep Learning Framework for Genomic Ancestry Prediction (AJHG, 2026). The record contains the resources for the Diet Network models trained for genetic ancestry prediction using data of the 1000 Genomes Project (1KGP). The source code and usage instructions are available at https://github.com/HussinLab/DIETNETWORK.

This record contains

• Dietnet_inference.tar.gz: 15 trained model weights and files for inference.

• Dietnet_training.tar.gz: Files used for model training, including the 1KGP dataset, configuration files, normalization statistics, SNPs embedding and fold splits, to enable reproducibility.

The 1KGP dataset is provided as part of this repository exclusively to ensure reproducibility of the analyses presented in this paper. The authoritative source for these data is The International Genome Sample Resource (IGSR), which maintains and publishes the dataset. Any reuse of the 1KGP data should reference and cite the official IGSR resource.