A pseudotime-dependent TWAS framework identifies disease genes along cell developmental paths.

Transcriptome-wide association studies (TWASs) link genes to disease risk by integrating gene expression with genome-wide association study (GWAS) data. The growing availability of single-cell expression data offers the opportunity to dissect these associations at finer cellular resolution and uncover effects masked in bulk TWAS analyses. Existing single-cell TWAS methods often map associations to discrete cell types, potentially overlooking the continuous nature of cellular processes and misidentifying the causal cell stages where genes exert their effects. To address this limitation, we developed the pseudotime-dependent TWAS (pt-TWAS), a framework that models gene expression as a continuous function of pseudotime to capture dynamic gene effects along developmental trajectories. By flexibly modeling and utilizing shared genetic effects across cell stages, this approach achieved higher statistical power than existing single-cell TWAS methods in our extensive simulations. pt-TWAS further enables identification of causal cell stages underlying disease risk by constructing confidence bands for gene effect curves. Applied to a GWAS of B cell acute lymphoblastic leukemia using single-cell data from OneK1K, pt-TWASs replicated known risk genes and pinpointed their relevant cell stages, demonstrating its utility for revealing fine-grained, cell-stage-specific genetic mechanisms. An R package implementing pt-TWASs is available on GitHub.