RcisTarget is a new R/Bioconductor implementation of the motif enrichment framework of i-cisTarget and iRegulon. RcisTarget identifies enriched transcription factor binding motifs and candidate transcription factors for a gene list. In brief, RcisTarget is based on two steps. First, it selects DNA motifs that are significantly over-represented in the surroundings of the transcription start site (TSS) of the genes in the gene-set. This is achieved by applying a recovery-based method on a database that contains genome-wide cross-species rankings for each motif. The motifs that are annotated to the corresponding TF and obtain a Normalized Enrichment Score (NES) > 3.0 are retained. Next, for each motif and gene-set, RcisTarget predicts candidate target genes (i.e. genes in the gene-set that are ranked above the leading edge). This method is based on the approach described by Aerts et al. 32 (link) which is also implemented in i-cisTarget (web interface) 33 (link) and iRegulon (Cytoscape plug-in) 34 (link). Therefore, when using the same parameters and databases, RcisTarget provides the same results as i-cisTarget or iRegulon, benchmarked against other TFBS-enrichment tools in Janky et al. 34 (link). More details about the method and its implementation in R are given in the package documentation.
To build the final regulons, we merge the predicted target genes of each TF-module that show enrichment of any motif of the given TF. To detect repression, it is theoretically possible to follow the same approach with the negative-correlated TF modules. However, in the datasets we analyzed, these modules were less numerous and showed very low motif enrichment, suggesting that these are lower quality modules. For this reason, we finally decided to exclude the detection of direct repression from the workflow, and continue only with the positive-correlated targets. The databases used for the analyses presented in this paper are the "18k motif collection" from iRegulon (gene-based motif rankings) for human and mouse. For each species, we used two gene-motif rankings (10kb around the TSS or 500bp upstream the TSS), which determine the search space around the transcription start site.
To build the final regulons, we merge the predicted target genes of each TF-module that show enrichment of any motif of the given TF. To detect repression, it is theoretically possible to follow the same approach with the negative-correlated TF modules. However, in the datasets we analyzed, these modules were less numerous and showed very low motif enrichment, suggesting that these are lower quality modules. For this reason, we finally decided to exclude the detection of direct repression from the workflow, and continue only with the positive-correlated targets. The databases used for the analyses presented in this paper are the "18k motif collection" from iRegulon (gene-based motif rankings) for human and mouse. For each species, we used two gene-motif rankings (10kb around the TSS or 500bp upstream the TSS), which determine the search space around the transcription start site.
