FUN-PROSE: A deep learning approach to predict condition-specific gene expression in fungi
Fig 3
Discovery of sequence motifs by the FUN-PROSE model.
(a)-(c) Examples of sequence motifs extracted from the convolutional kernels of the FUN-PROSE model trained on the respective species: S. cerevisiae (a), N crassa (b), I. orientalis (c). The best matching S. cerevisiae motif in the YEASTRACT database is shown in the middle column and motif’s positional activation profile—in the right column. (d) The heatmap showing the positional distribution of the top 0.5% of activations for each motif, i.e. kernel in the first convolutional layer, over all S. cerevisiae promoter sequences. The rows of this heatmap are sorted by the average activation level within 300bp from the transcription start site. Note that most motifs exhibit non-random positional preferences indicative of biological function.