Reporting

The final table is produced by putting together the information already contained in model, classification and centrality: one row per selected motif, with its annotation, its positional behaviour and its contribution to the predictor.

report<-cbind(data.frame(motif=rownames(model$coefficients),classification=NA,is_central=FALSE,PE=NA,CE=NA,motif_auroc=NA),coeff=model$coefficients)
report$classification<-TF_ENCODE_KDM_REF$info$target[match(classification$target[match(rownames(model$coefficients),classification$query)],TF_ENCODE_KDM_REF$info$name)]
report<-report[order(report$coeff.central,decreasing=TRUE),]
mm<-match(report$motif,centrality$result$motif)
ss<-which(!is.na(mm))
report$is_central[ss]<-TRUE
report$PE[ss]<-centrality$result$PE[mm[ss]]
report$CE[ss]<-centrality$result$CE[mm[ss]]
report$motif_auroc<-centrality$profiles$aurocs[match(report$motif,colnames(motifs))]
rownames(report)<-1:nrow(report)

Reading the table

Where each column comes from:

NRF1 — example output

In the worked example the top row is the expected one: M-105, classified as NRF1, is centrally enriched, carries by far the largest central coefficient and reaches an AUROC of 0.98 on its own. Several further NRF1 motifs (M-59, M-15, M-1) recover the same signal, while co-occurring factors and chromatin components appear with smaller — and sometimes negative — contributions.

report

The table below is interactive: sort on any column with its header button, and filter the rows with the search box — try NRF1 to collect the motifs that recover the expected factor, or TRUE for the centrally enriched ones.