sPrediXcan

Summary

sPrediXcan (Summary-PrediXcan) is a command-line tool implemented in Python for performing Transcriptome-Wide Association Studies (TWAS) using genome-wide association study (GWAS) summary statistics. It is part of the MetaXcan software suite and utilizes pre-computed gene expression prediction models (such as those trained on GTEx data) to estimate the association between predicted gene expression levels and a phenotype of interest. By analyzing summary statistics rather than individual-level genotype data, sPrediXcan enables fast, computationally efficient, and widely applicable gene-level association analyses.

Methodology and Use Cases

sPrediXcan extends the original PrediXcan framework, which requires individual-level genotype data, to work directly with summary-level datasets: - Gene Expression Prediction: It utilizes weights from prediction models (trained on reference transcriptome panels using elastic net or similar methods) to impute the genetic component of gene expression. - Summary Statistics Integration: It mathematically infers the association between genetically-predicted gene expression and the phenotype using GWAS effect sizes and linkage disequilibrium (LD) reference panels, ignoring and removing SNPs that are present in the models but missing from the GWAS. - Benchmarking Performance: A systematic evaluation benchmarked sPrediXcan alongside FUSION for TWAS signature-matching-based drug prioritisation, demonstrating that sPrediXcan outperformed FUSION by yielding more robust drug class enrichment for first-line treatments like HMGCR inhibitors.

Derivation

sPrediXcan (S-PrediXcan) is derived as a closed-form summary-statistics equivalent of individual-level PrediXcan: given GWAS SNP effect sizes, elastic-net expression-prediction weights, and an LD reference panel's SNP covariance matrix, the gene-level Z-score can be computed analytically without ever accessing individual genotypes or phenotypes. Validated against true PrediXcan across a null-hypothesis simulated phenotype, a cellular growth phenotype, and WTCCC bipolar disorder and type 1 diabetes GWAS, S-PrediXcan and PrediXcan Z-scores correlated at r² > 0.99 when reference and study populations matched, and remained highly concordant (r² generally 0.85–0.93) even under substantial ancestry mismatch between the GWAS study population and the LD reference panel — though a handful of false positives emerged in poorly-matched cases, underscoring the value of closely matched LD reference data.

sPrediXcan differs from SNP-level GWAS by testing genes rather than variants: for each gene it correlates genetically-predicted expression (from a reference transcriptome panel such as GTEx) with the phenotype, rather than testing individual SNP-phenotype associations directly. Compared to two related summary-based methods, the original paper found S-TWAS (which uses a polygenic BSLMM prediction model) shows a higher proportion of colocalization-estimated independent (non-causal) signals among its significant genes than S-PrediXcan's sparse elastic-net models, and SMR's association statistic is miscalibrated under the null (mean ≈0.93 rather than the expected 1) and its significance is bounded by whichever of the eQTL or GWAS association is weaker — a property that unnecessarily penalizes genes with strong GWAS signal but only modest eQTL power.

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] 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