sMultiXcan

Summary

sMultiXcan is a multi-tissue gene-level association tool developed as part of the MetaXcan software suite. It integrates predicted gene expression levels from multiple tissues using GWAS summary statistics and reference expression models (such as GTEx) to identify genes associated with a specific trait. By combining information across tissues, sMultiXcan increases statistical power to discover gene-trait associations, particularly when the causal tissue is unknown or lacks well-powered eQTL models.

Methodology and Use Cases

sMultiXcan addresses the tissue-specific limitations of single-tissue TWAS by modeling multivariate associations: - Tissue Integration: It builds a joint regression model that tests the association between a trait and predicted expression across multiple tissues simultaneously, accounting for the correlation of predicted expression between tissues. - Statistical Power: By pooling eQTL data, it increases the overall statistical power of the TWAS, enabling the detection of genes with weaker, shared regulatory signals. - Drug Repurposing Benchmark: In a benchmarking study by the Shah Lab (as of 2026-07-19), sMultiXcan models were evaluated for TWAS signature-matching based drug prioritisation. The study found that while sMultiXcan has higher power for gene-trait associations, single-tissue models in trait-relevant tissues (e.g., liver for LDL cholesterol) performed better for prioritizing known first-line treatments, as multi-tissue models may dilute the tissue-specific signatures critical for drug action.

Method and Validation

MultiXcan jointly regresses a phenotype on a gene's predicted expression across all available tissue models simultaneously (rather than testing each tissue separately), using principal components of the predicted-expression matrix to avoid multicollinearity from the substantial cross-tissue correlation (median pairwise correlation r ≈ 0.56 across GTEx tissue models for a given gene). S-MultiXcan is the summary-statistics-only extension, derived to be highly concordant with the individual-level version when LD is well matched.

Applied to 222 UK Biobank traits, MultiXcan detected more significant gene associations than single-tissue PrediXcan (scanned across all 44 GTEx tissues) in 103 traits, versus 21 traits where PrediXcan detected more, with an average 162.7% increase in significant associations and ~48% overlap between the two methods' hits. For self-reported high cholesterol (50,497 cases, 100,994 controls), MultiXcan found 251 significant genes versus 196 for PrediXcan across all tissues and only 33 for the single best tissue (whole blood), including lipid-metabolism genes (APOM, PAFAH1B2) and a glucose-transport gene (SLC5A6) missed by single/all-tissue PrediXcan. Simulations confirmed the expected trade-off: when a trait has one true causal tissue, single-tissue PrediXcan with that tissue outperforms MultiXcan in 99.9% of cases, but when multiple or all tissues are causal, MultiXcan wins in 84.4–99.5% of cases.

Citations

[1] Barbeira, A.N., Pividori, M., Zheng, J., Wheeler, H.E., Nicolae, D.L., & Im, H.K. (2019). Integrating predicted transcriptome from multiple tissues improves association detection. PLoS Genetics, 15(1), e1007889. [2] Chauquet, S., Jiang, J. C., Barker, L. F., Hunter, Z. L., Singh, G., Wray, N. R., McRae, A. F., & Shah, S. (2026). From GWAS to drug: A framework for drug candidate prioritisation using a gene expression signature matching approach. medRxiv. Source paper: 2026.04.22.26349470v2.full.pdf