A transparent and generalizable deep-learning framework for genomic ancestry prediction

Clinical Snapshot

Key Findings

• P Value: Not reported in abstract

• Effect Size: Not reported in abstract; specific accuracy or concordance metrics not available from published abstract

• Primary Outcome: The adapted Diet Network deep-learning model successfully predicts genetic ancestry from SNP data and generalises to diverse populations across CARTaGENE, Montreal Heart Institute, and All of Us biobanks, including emergent representation of North African population structure despite absence from training data

• Nnt Or Sensitivity: Sensitivity and specificity for ancestry classification not reported in abstract; DeepLift attribution identified population-specific SNPs consistent with traditional Fst-based population-genetics signals, suggesting high concordance with established methods

• Confidence Interval: Not reported in abstract

Clinical Application

The framework is designed for large-scale biobank application and demonstrates robustness to missing SNP data, which is a practical advantage in real-world genomic datasets. Implementation requires computational infrastructure for deep-learning model deployment and access to quality-controlled SNP data. The open-source nature of the Diet Network architecture supports reproducibility. Clinical translation would require integration with existing genomic data pipelines and validation in target clinical populations. Australia presents a distinct genomic landscape with significant Indigenous Australian, Pacific Islander, Melanesian, and Southeast Asian ancestries that are substantially underrepresented in both the 1KGP training dataset and the North American validation biobanks used in this study. The Australian Genomics Health Alliance, AGHA, and the National Centre for Indigenous Genomics (NCIG) have highlighted the critical need for ancestry-aware genomic tools that perform equitably for Aboriginal and Torres Strait Islander peoples. Direct application of this model to Australian clinical genomics — including PBS-listed pharmacogenomic testing contexts and TGA-regulated genomic diagnostics — would require prospective validation in Australian cohorts. RACGP guidelines on genomic medicine emphasise the importance of ancestry-appropriate risk stratification. The framework's architecture and explainability approach are potentially transferable to Australian biobank development (e.g., the Australian Genomics Biobank), but performance claims cannot be extrapolated without local validation. Individuals with available SNP genotyping data in large-scale biobank or clinical genomic settings, particularly those from diverse or admixed ancestral backgrounds where ancestry characterisation is needed for genomic study stratification, pharmacogenomic risk adjustment, or disease association analyses

Abstract

Accurately characterizing genetic ancestry is critical for ensuring reproducibility and fairness in genomic studies and downstream health research. This study aims to address the prediction of ancestry from genetic data using deep learning, with a focus on generalizability across datasets with diverse populations and on explainability to improve model transparency. We adapt the Diet Network, a deep-learning architecture proven to be effective in handling high-dimensional data, to learn population ancestry from single-nucleotide polymorphism (SNP) data using the populational Thousand Genomes Project dataset. Our results highlight the model's ability to generalize to diverse populations in the CARTaGENE, Montreal Heart Institute, and All of Us biobanks and that predictions remain robust to high levels of missing SNPs. We show that, despite the lack of North African populations in the training dataset, the model learns latent representations that reflect meaningful population structure for North African individuals in the biobanks. To improve model transparency, we apply Saliency Maps, DeepLift, GradientShap, and Integrated Gradients attribution techniques and evaluate their performance in identifying SNPs leveraged by the model. Using DeepLift, we show that the model's predictions are driven by population-specific signals consistent with those identified by traditional population-genetics metrics. This work presents a generalizable and interpretable deep-learning framework for genetic-ancestry inference in large-scale biobanks with genetic data. By enabling more widespread genomic ancestry characterization in these cohorts, this study contributes practical tools for integrating genetic data into downstream biomedical applications, supporting more inclusive and equitable healthcare solutions.

References

1. 1.Rochefort-Boulanger, C., Scicluna, M., Poujol, R., Grenier, J.-C., Carrier, P. L., Lemieux, S., & Hussin, J. G. (2026). A transparent and generalizable deep-learning framework for genomic ancestry prediction. American Journal of Human Genetics. https://doi.org/10.1016/j.ajhg.2026.06.003