Cross-task benchmarking reveals variable transferability of deep learning models across molecular quantitative trait loci.
6 affiliations
- 1Monash Biomedicine Discovery Institute and Department of Biochemistry and Molecular Biology, Monash University, Wellington Road, Clayton, Melbourne, Victoria 3800, Australia.
- 2Department of Infectious Diseases, The Alfred Hospital and School of Translational Medicine, Monash University, 85 Commercial Road, Melbourne, Victoria 3004, Australia.
- 3Monash Biomedicine Discovery Institute, Department of Microbiology, Monash University, Wellington Road, Clayton, Melbourne, Victoria 3800, Australia.
- 4Centre to Impact AMR, Monash University, Wellington Road, Clayton, Melbourne, Victoria 3800, Australia.
- 5South Australian immunoGENomics Cancer Institute (SAiGENCI), Adelaide University, 4 North Terrace, Adelaide, South Australia 5005, Australia.
- 6Department of Data Science and Artificial Intelligence, Faculty of Information Technology, Monash University, Wellington Road, Clayton, Melbourne, Victoria 3800, Australia.
Abstract
Most genetic variants associated with complex traits reside in non-coding regions of the genome, complicating their functional interpretation. Recent deep learning models have advanced the prediction of expression quantitative trait loci (eQTLs), but whether these models generalize across diverse molecular phenotypes is unknown. Here, we assess the cross-task transferability of pretrained regulatory features from six representative sequence- and annotation-based models across nine classes of molecular quantitative trait loci (xQTLs). We observe that while all models perform well in eQTL prediction, their performance varies substantially across other xQTL predictions. Chromatin accessibility, transcription factor binding, and histone QTLs are generally well predicted, whereas chromatin looping and microRNA expression QTLs remain challenging. Foundation models, particularly AlphaGenome and Borzoi, consistently outperform models based on local sequence context and exhibit improved robustness across class-imbalance settings. Tree-based ensemble classifiers further enhance predictive performance, suggesting that nonlinear interactions among regulatory features contribute substantially to variant prioritization. Applying AlphaGenome and Borzoi to coronary artery disease-associated genome-wide association studies loci identifies 337 high-confidence candidate chromatin accessibility QTLs, including 11 previously reported and 326 candidates that display significantly larger allele-specific differences in predicted chromatin accessibility signals. Collectively, our results reveal substantial variability in the cross-task performance of current eQTL-centric models, highlight the advantages of long-range genomic representations and inform the prioritization of candidate regulatory variants contributing to complex diseases.
Related
- Quantitative Delineation of the Morphological Heterogeneity of Anisotropic Nanostructures via Machine-Learning-Augmented Electron Microscopy.
- AutoWrinkleID: a machine learning pipeline for biofilm wrinkle identification and quantitative analysis
- Bimodal spectroscopy integrating multi-wavelength time-resolved photoacoustic spectroscopy and near-infrared spectroscopy with deep learning for quantitative detection of serum biochemical indicators
- Machine Learning Integrating Quantitative MRI and Clinical Features Predicts Loss of Ambulation in Duchenne Muscular Dystrophy.
- Immune-related molecular links between asthma and Alzheimer’s disease identified by multi-omics, machine learning and single-cell analysis
BibTeX
@article{chen2026crosstask,
title = {{Cross-task benchmarking reveals variable transferability of deep learning models across molecular quantitative trait loci.}},
author = {Sihan Chen and Xiaoyu Wang and Anton Y Peleg and Shuting Xu and Fuyi Li and Geoffrey I Webb and Jiangning Song},
journal = {Briefings in bioinformatics},
year = {2026},
month = sep,
doi = {10.1093/bib/bbag599},
url = {https://europepmc.org/article/MED/42960994},
note = {CC BY-NC},
}
RIS
TY - JOUR
TI - Cross-task benchmarking reveals variable transferability of deep learning models across molecular quantitative trait loci.
AU - Sihan Chen
AU - Xiaoyu Wang
AU - Anton Y Peleg
AU - Shuting Xu
AU - Fuyi Li
AU - Geoffrey I Webb
AU - Jiangning Song
PY - 2026
DA - 2026/09/01
JO - Briefings in bioinformatics
DO - 10.1093/bib/bbag599
UR - https://europepmc.org/article/MED/42960994
AB - Most genetic variants associated with complex traits reside in non-coding regions of the genome, complicating their functional interpretation. Recent deep learning models have advanced the prediction of expression quantitative trait loci (eQTLs), but whether these models generalize across diverse molecular phenotypes is unknown. Here, we assess the cross-task transferability of pretrained regulatory features from six representative sequence- and annotation-based models across nine classes of molecular quantitative trait loci (xQTLs). We observe that while all models perform well in eQTL prediction, their performance varies substantially across other xQTL predictions. Chromatin accessibility, transcription factor binding, and histone QTLs are generally well predicted, whereas chromatin looping and microRNA expression QTLs remain challenging. Foundation models, particularly AlphaGenome and Borzoi, consistently outperform models based on local sequence context and exhibit improved robustness across class-imbalance settings. Tree-based ensemble classifiers further enhance predictive performance, suggesting that nonlinear interactions among regulatory features contribute substantially to variant prioritization. Applying AlphaGenome and Borzoi to coronary artery disease-associated genome-wide association studies loci identifies 337 high-confidence candidate chromatin accessibility QTLs, including 11 previously reported and 326 candidates that display significantly larger allele-specific differences in predicted chromatin accessibility signals. Collectively, our results reveal substantial variability in the cross-task performance of current eQTL-centric models, highlight the advantages of long-range genomic representations and inform the prioritization of candidate regulatory variants contributing to complex diseases.
SN - 1467-5463
SN - 1477-4054
ER -