Benchmarking methods integrating GWAS and single-cell transcriptomic data for mapping trait-cell type associations

Genome-wide association studies (GWAS) have discovered numerous trait-associated variants, but their biological context remains unclear. Integrating GWAS summary statistics with single-cell RNA-sequencing (scRNA-seq) data enables prioritization of cell types in which these variants influence traits. Existing methods broadly follow two strategies: single cell to GWAS, which identifies cell-type-specific genes and tests their enrichment in GWAS signals, and GWAS to single cell, which begins with GWAS-prioritized genes and scores cells or cell types according to their expression profiles. Here, we developed a literature-informed benchmark by integrating PubMed evidence with large language model-assisted literature synthesis to evaluate 20 trait-cell type mapping methods. We identify CATCH, a Cauchy combination of complementary methods, as the most robust overall approach, consistently achieving high statistical power while maintaining effective false-positive control across simulations and real-data benchmarks. We further identify key determinants of performance, including cell-type specificity metrics, GWAS statistical power, and the diversity of scRNA-seq reference datasets, providing practical guidance for the development and application of trait-cell type mapping methods.