Fig. 3: Comparison of cTWAS with other methods in simulations.

a, Comparison of the P values from standard TWAS and gene PIPs from cTWAS. The results were from one simulation run (parameters: gene PVE 0.052, gene prior 0.025, SNP PVE 0.50, SNP prior 0.00025). Each dot represents a gene and is colored based on whether it is a causal gene. b, Number of genes identified by various methods. We used the following significance thresholds for each method: PIP > 0.8 for cTWAS; Bonferroni-corrected P < 0.05 for FUSION; PP4 > 0.8 for coloc; PIP > 0.8 for FOCUS; FDR < 0.05 for SMR with P < 0.05 for HEIDI; Bonferroni-corrected P < 0.05 for MR-JTI with FDR < 0.05 for FUSION; Bonferroni-corrected P < 0.05 for PMR-Egger; and LFSR < 0.1 for MRLocus. We use different colors for causal genes identified by each method, and the top gray bars indicate noncausal genes. The height of the bar is the mean over five independent simulations; the standard error is shown for each bar in vertical lines from the same five simulations. c,d, Examples of how cTWAS avoided false-positive genes. Top, −log₁₀ P values of genes (from TWAS) and SNPs in a region. Bottom, PIPs of genes and SNPs. Genes are represented by squares, with positions determined by transcription start sites, and SNPs represented by circles. Colors indicate whether the gene or SNP is causal (orange), noncausal but in LD with a causal effect (R² between SNP genotype or imputed expression >0.4, purple), or noncausal and not in LD with causal effect (green). The eQTLs of the genes are plotted in middle tracks. Transcriptome-wide significance cutoff for TWAS was indicated by the red dotted line. Top, values of PP4 (probability of colocalization) from coloc analysis were shown for each gene of interest.