Fig. 4
From: Tensorial blind source separation for improved analysis of multi-omic data
Validation of tensorial ICA on multi-tissue smoking EWAS. a Left panel: Scatterplot of the weights of estimated independent components S1,12,i and S2,12,i from the data tensor of dimension 2×152×1062, with mode 1 representing tissue type, mode 2 the different women and mode 3 the CpGs. Red denotes the smkDMCs. Middle panel: As left panel, but now for the rotated tensor, projecting the data onto the whole blood (WB) and buccal (BUC) dimensions, demonstrating the strong correlation between the DNAm variation in whole blood and buccal tissue. Right panel: As left panel, but now for the absolute weights. The green dashed lines represent the cutoff point selecting the 62 CpGs with the largest absolute weights. There are in total 41 smkDMCs among the three larger quadrants, corresponding to a sensitivity of 41/62=0.66, with the enrichment P value given above the plot. b Pearson correlation heat map of the 41 smkDMCs between whole blood (WB) and buccal (BUC) tissue, with correlations computed over the 152 samples. c Plots of the 12th independent component of the mixing matrix in the sample space (y-axis) against smoking exposure for the 152 samples. Left panel: Smoking-pack-years. Right panel: Smoking status (never smokers, ex-smokers and smokers at sample draw). P values are from linear regressions. BUC buccal, DNAm DNA methylation, EWAS epigenome-wide association study, ExSmk ex-smoker, smkDMC smoking-associated differentially methylated CpG, WB whole blood