Figure 6: Detected associations by simulated univariate (as in trQTLs) and multivariate (as in sQTLs) approaches.

Gain in the proportion of detected association (y axis) when using multivariate approach versus univariate approach. We simulated two populations for genes expressing 3, 4, 7 or 10 transcripts (columns) when the transcript ratios are shifted by a certain value (effect size, x axis) from one population to the other following four different scenarios (rows): ‘all transcripts’: the splicing ratios of all transcripts change with the same intensity in the second population compared with the first. ‘first and second major transcripts only’: only the splicing ratios of the first two major transcripts, are shifted in the second population. ‘second and third major transcripts only’: only the splicing ratios of the second and third major isoforms are changed in the second population. ‘first transcript strong, others weak’: the splicing ratios of all transcripts are shifted but the value of the change in the major isoform is distributed equally among the rest of the isoforms, that is, the major transcript changes strongly while the other transcripts change slightly. For each configuration, 500 genes with different splicing ratios were pooled with 4,500 non-associated genes. After multiple testing correction, we compute the true positive rate (that is, proportion of true association detected) using a FDR threshold of 1%. The plot displays the difference between multivariate TPR and univariate TPR. (Positive values correspond to higher TPR in the multivariate approach). The curves are obtained using a LOESS model.