Skip to content

MetaXcan

Summary

MetaXcan is a suite of bioinformatics tools developed by the Im Lab to perform association studies between imputed molecular traits (such as gene expression or protein levels) and clinical phenotypes. By leveraging genetic prediction models from databases like PredictDB, MetaXcan enables researchers to perform Transcriptome-Wide Association Studies (TWAS) and Proteome-Wide Association Studies (PWAS) using individual-level data or GWAS summary statistics (via S-PrediXcan).

Capabilities and Methods

  • S-PrediXcan: Extends the PrediXcan framework to run association analyses using only GWAS summary statistics, avoiding the need for individual-level genotype data.
  • Multi-omic Imputation: Integrates prediction models for transcriptomic, proteomic, and metabolomic traits (such as those hosted on OmicsPred).
  • Multi-ancestry Support: Complements newer multi-ancestry fine-mapping and prediction packages like SuShiE.

Derivation and Validation

The founding paper derives a closed-form expression for computing S-PrediXcan results directly from GWAS summary statistics, phenotype-SNP effect sizes, elastic-net prediction weights, and an LD reference panel — without needing the individual-level genotype/phenotype data the original PrediXcan method required. Applying the framework across 44 GTEx tissues and 100+ GWAS/meta-analysis phenotypes, the authors found S-PrediXcan results were highly concordant with individual-level PrediXcan (r² typically 0.85–0.99+ across simulated, cellular-growth, and disease phenotypes), and robust to mismatches between the GWAS study population and the LD reference set (African, East Asian, and European 1000 Genomes subsets all produced consistent results). Most significant associations were tissue-specific rather than shared across all 44 tissues, and monogenic disease genes were enriched among significant associations for phenotypically related complex traits, consistent with a spectrum of allelic severity from rare Mendelian mutations to common regulatory variants.

Colocalization and Method Comparisons

The MetaXcan framework pairs S-PrediXcan associations with colocalization (COLOC) to filter out associations driven by LD contamination rather than a shared causal variant — the paper's suggested pipeline is: (1) run S-PrediXcan across all tissues with Bonferroni correction, (2) filter by prediction-model significance, (3) discard gene-tissue pairs with high probability of independent (non-colocalized) eQTL/GWAS signals, and (4) optionally retain only pairs with explicit colocalization evidence (COLOC P4 > 0.5). Benchmarked against contemporaneous summary-based methods, S-PrediXcan showed higher concordance with COLOC-estimated colocalization than summary-TWAS (which uses a polygenic BSLMM prediction model rather than sparse elastic net), and avoided a miscalibration the authors identified in Summary Mendelian Randomization (SMR): the SMR χ² statistic's null-hypothesis mean was ≈0.93 rather than the expected 1, and its significance is capped by whichever is weaker of the top eQTL or GWAS association, unnecessarily penalizing well-powered GWAS hits with only modest eQTL support.

A follow-up analysis assembled a broader "blueprint" for interpreting GWAS loci with GTEx resources, benchmarking causal-gene-prioritization approaches (including MetaXcan/S-PrediXcan alongside colocalization methods) against each other and finding that no single approach consistently outperforms combining multiple approaches. Using colocalization together with association methods that account for allelic heterogeneity of gene expression, that study proposed candidate target genes for 47% (2,519 of 5,385) of the GWAS loci examined, and documented widespread dose-dependent effects of RNA expression and splicing across a broad panel of complex traits and diseases.

See Also

  • sPrediXcan and sMultiXcan — the single-tissue and multi-tissue summary-statistics tools within the MetaXcan framework.
  • PhenomeXcan — a large pre-computed resource applying S-PrediXcan/S-MultiXcan across 4,091 traits, refined with the fastENLOC colocalization method.
  • scPrediXcan — extends the S-PrediXcan framework to cell-type-resolution TWAS via a deep-learning cell-type-specific expression model.
  • Multi-INTACT — a Bayesian method combining multiple gene products (expression and protein) with TWAS/colocalization evidence to prioritize causal genes.
  • MetaboXcan — extends the MetaXcan/S-PrediXcan approach from gene expression to genetically predicted plasma metabolites.
  • Polygenicity-Induced Inflation in TWAS — a follow-up finding that highly polygenic target traits inflate MetaXcan/S-PrediXcan false positive rates independent of LD contamination; the correction is now built into the software (v0.8.0+).

Citations

[1] Barbeira, A.N., Dickinson, S.P., Bonazzola, R., Zheng, J., Wheeler, H.E., Torres, J.M., Torstenson, E.S., Shah, K.P., Garcia, T., Edwards, T.L., Stahl, E.A., Huckins, L.M., GTEx Consortium, Nicolae, D.L., Cox, N.J., & Im, H.K. (2018). Exploring the phenotypic consequences of tissue specific gene expression variation inferred from GWAS summary statistics. Nature Communications, 9, 1825. [2] Barbeira, A.N., Bonazzola, R., Gamazon, E.R., Liang, Y., Park, Y., Kim-Hellmuth, S., et al. (2021). Exploiting the GTEx resources to decipher the mechanisms at GWAS loci. Genome Biology, 22, 49.