MOD-scTWAS: Leveraging gene co-expression for single-cell transcriptome-wide association studies

Transcriptome-wide association studies (TWAS) offer a robust approach for pinpointing genes linked to intricate traits. The advent of population-scale single-cell transcriptomic data facilitates genetically informed expression (GReX) prediction and TWAS analyses at a cellular level, however, the efficacy of current single-cell TWAS methods is still suboptimal.

In this study, we introduce MOD-scTWAS, a module-based technique that concurrently models GReX for genes within co-expression clusters to leverage shared information across genes. Building upon a generative model for single-cell gene expression, MOD-scTWAS incorporates heteroscedasticity and inter-gene correlation of individual-level pseudobulk expression in joint GReX prediction.

Through cross-validation on the OneK1K dataset, MOD-scTWAS demonstrated superior mean GReX prediction accuracy compared to scTWAS across all 14 cellular types and expanded the pool of imputable genes. Upon application to TWAS analyses of UK Biobank quantitative hematological traits, MOD-scTWAS unveiled a greater number of significant cell type-gene-trait associations than scTWAS.

These findings underscore the potential of harnessing gene co-expression through joint modeling to enhance cell-type-specific GReX prediction and TWAS discovery.