Using sequence-to-function models to interpret archaic hominin introgression
Abstract
Understanding the functional impact of archaic hominin introgression remains challenging due to the poor representation of global introgression in publicly available genomics resources. Sequence-to-function models can predict the effects of any possible variant in the human genome and may fill this gap. Here, we used AlphaGenome to predict the effects of 144,139 introgressed SNPs segregating in present-day individuals of Papuan genetic ancestry. AlphaGenome’s chromatin accessibility predictions recapitulate experimentally observed effects, but gene expression performs no better than chance. Predictions correlate more strongly with an independent reporter assay of single-variant activity than with the same variants’ effects in live cells, indicating that AlphaGenome captures the regulatory potential of individual variants more reliably. Predictions carry tissue specificity, allowing us to predict specific tissues potentially impacted by introgressed haplotypes. We identify genes, including JAK1 and TAB2, that are associated with haplotypes that contain an excess of variants predicted by AlphaGenome to have large impacts on chromatin accessibility. Finally, we highlight the challenges and limitations associated with using sequence-to-function models for introgressed variant effect prediction, and show that while AlphaGenome’s chromatin accessibility predictions can aid in prioritising candidate functional regions, expression predictions and the assignment of variants to target genes remain as open challenges.
Related
- Joint ancestry inference reveals the landscape of archaic introgression in admixed populations
- Genomic structural variations distinguishing archaic and modern humans are associated with neurodevelopment
- S2F-Agent: Harnessing sequence-to-function models for verifiable genome interpretation
- Archaic ancestry inference in imputed ancient human genomes
- QLoRA Fine-Tuning of Ministral LLM for Sequence-to-Function Protein Annotation
BibTeX
@misc{comerford2026using,
title = {{Using sequence-to-function models to interpret archaic hominin introgression}},
author = {Maddy Comerford and Isabela Alvim and Timothy M Johanson and Danat Yermakovich and Christopher Kinipi and Matthew Leavesley and François-Xavier Ricaut and Michael Dannemann and Murray P Cox and Rhys Allan and Nicolas Brucato and Irene Gallego Romero},
howpublished = {HAL},
year = {2026},
month = aug,
doi = {10.64898/2026.08.31.748430},
url = {https://hal.science/hal-05769538v1},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Using sequence-to-function models to interpret archaic hominin introgression
AU - Maddy Comerford
AU - Isabela Alvim
AU - Timothy M Johanson
AU - Danat Yermakovich
AU - Christopher Kinipi
AU - Matthew Leavesley
AU - François-Xavier Ricaut
AU - Michael Dannemann
AU - Murray P Cox
AU - Rhys Allan
AU - Nicolas Brucato
AU - Irene Gallego Romero
PY - 2026
DA - 2026/08/31
PB - HAL
DO - 10.64898/2026.08.31.748430
UR - https://hal.science/hal-05769538v1
AB - Understanding the functional impact of archaic hominin introgression remains challenging due to the poor representation of global introgression in publicly available genomics resources. Sequence-to-function models can predict the effects of any possible variant in the human genome and may fill this gap. Here, we used AlphaGenome to predict the effects of 144,139 introgressed SNPs segregating in present-day individuals of Papuan genetic ancestry. AlphaGenome’s chromatin accessibility predictions recapitulate experimentally observed effects, but gene expression performs no better than chance. Predictions correlate more strongly with an independent reporter assay of single-variant activity than with the same variants’ effects in live cells, indicating that AlphaGenome captures the regulatory potential of individual variants more reliably. Predictions carry tissue specificity, allowing us to predict specific tissues potentially impacted by introgressed haplotypes. We identify genes, including JAK1 and TAB2, that are associated with haplotypes that contain an excess of variants predicted by AlphaGenome to have large impacts on chromatin accessibility. Finally, we highlight the challenges and limitations associated with using sequence-to-function models for introgressed variant effect prediction, and show that while AlphaGenome’s chromatin accessibility predictions can aid in prioritising candidate functional regions, expression predictions and the assignment of variants to target genes remain as open challenges.
ER -