TorchGWAS2: Cost-Effective Phenome- and Genome-Wide Association Testing in Related Samples
- 1New York University
Abstract
Modern large-scale genetic association analyses of imaging and omics data reveal unprecedented details of the genetic architecture of complex traits. Such analyses involve scanning thousands of phenotypes using linear mixed model-based genome-wide association study tools to control for sample relatedness. However, current LMM tools are not designed for such scale, creating a computational burden that hinders discovery. We propose TorchGWAS2, a cost-effective solution that overcomes the bottleneck using a deterministic variance-correction algorithm for LMMs, making it well-suited for GPU acceleration. TorchGWAS2 is applicable to unrelated and related individuals, cross-sectional and longitudinal studies, with and without missing data, and its computational complexity scales linearly with the number of phenotypes, genetic variants, and individuals. TorchGWAS2 showed more powerful association testing across 128 retinal image-derived endophenotypes of pairs of eyes from 64,703 UK Biobank participants and achieved two orders of magnitude speed-up analyzing 1,023 circulating metabolites in 16,352 Trans-Omics for Precision Medicine participants.
Related
- Decoding Atrial Fibrillation Heterogeneity : A Nationwide Phenome-Wide Association Study in England
- A Clinical Atlas for Phenome-Wide Association Study of Obstructive Sleep Apnea Across Electronic Health Record Domains
- Accelerometer-derived real-world sleep stages and risk of incident diseases: A UK Biobank cohort study and phenome-wide association analysis
- Thrombophilia Testing After VTE and in Related Clinical Contexts: Is It Necessary, and Does It Change Management? Appraisal of Evidence, Guidelines, and Clinical Yield
- A novel predictive model for DNA mismatch repair deficient colorectal carcinoma for cost-effective MMR testing in low-resource settings: HERALD-MMR (Histological Evaluation and Risk Assessment using Local Demographics for Mismatch Repair).
BibTeX
@misc{zhang2026torchgwas2,
title = {{TorchGWAS2: Cost-Effective Phenome- and Genome-Wide Association Testing in Related Samples}},
author = {M. Zhang and Z. Xie and S. Salehi nasab and N. Wang and X. Zhao and T. Alkis and J. Barnard and T. W. Blackwell and R. P. Bowler and S. Chung and M. Cho and C. B. Clish and E. Drzymalla and A. M. Evans and N. Franceschini and R. E. Gerszten and M. G. Gillman and M. L. Grove and N. Heard-Costa and S. R. Hutton and R. S. Kelly and C. Kooperberg and M. G. Larson and J. Lasky-Su and D. A. Meyers and F. P. Ockerman and L. M. Raffield and A. P. Reiner and S. S. Rich and J. I. Rotter and A. V. Smith and K. D. Taylor and R. S. Vasan and S. T. Weiss and K. E. Wong and A. C. Wood and P. G. Woodruff and L. Wu and R. I. Yarden and J. Yu and L. Y. Zhou and B. Yu and D. Zhi and H. Chen},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.10.26362744},
url = {https://www.medrxiv.org/content/10.64898/2026.09.10.26362744v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - TorchGWAS2: Cost-Effective Phenome- and Genome-Wide Association Testing in Related Samples
AU - M. Zhang
AU - Z. Xie
AU - S. Salehi nasab
AU - N. Wang
AU - X. Zhao
AU - T. Alkis
AU - J. Barnard
AU - T. W. Blackwell
AU - R. P. Bowler
AU - S. Chung
AU - M. Cho
AU - C. B. Clish
AU - E. Drzymalla
AU - A. M. Evans
AU - N. Franceschini
AU - R. E. Gerszten
AU - M. G. Gillman
AU - M. L. Grove
AU - N. Heard-Costa
AU - S. R. Hutton
AU - R. S. Kelly
AU - C. Kooperberg
AU - M. G. Larson
AU - J. Lasky-Su
AU - D. A. Meyers
AU - F. P. Ockerman
AU - L. M. Raffield
AU - A. P. Reiner
AU - S. S. Rich
AU - J. I. Rotter
AU - A. V. Smith
AU - K. D. Taylor
AU - R. S. Vasan
AU - S. T. Weiss
AU - K. E. Wong
AU - A. C. Wood
AU - P. G. Woodruff
AU - L. Wu
AU - R. I. Yarden
AU - J. Yu
AU - L. Y. Zhou
AU - B. Yu
AU - D. Zhi
AU - H. Chen
PY - 2026
DA - 2026/09/10
PB - medRxiv
DO - 10.64898/2026.09.10.26362744
UR - https://www.medrxiv.org/content/10.64898/2026.09.10.26362744v1
AB - Modern large-scale genetic association analyses of imaging and omics data reveal unprecedented details of the genetic architecture of complex traits. Such analyses involve scanning thousands of phenotypes using linear mixed model-based genome-wide association study tools to control for sample relatedness. However, current LMM tools are not designed for such scale, creating a computational burden that hinders discovery. We propose TorchGWAS2, a cost-effective solution that overcomes the bottleneck using a deterministic variance-correction algorithm for LMMs, making it well-suited for GPU acceleration. TorchGWAS2 is applicable to unrelated and related individuals, cross-sectional and longitudinal studies, with and without missing data, and its computational complexity scales linearly with the number of phenotypes, genetic variants, and individuals. TorchGWAS2 showed more powerful association testing across 128 retinal image-derived endophenotypes of pairs of eyes from 64,703 UK Biobank participants and achieved two orders of magnitude speed-up analyzing 1,023 circulating metabolites in 16,352 Trans-Omics for Precision Medicine participants.
ER -