Fig. 5: Integrative motifs derived from PrismNet are more specific in capturing transcriptome-wide binding site than canonical sequence motifs and cluster together with similar post-transcriptional regulatory functions.
From: Predicting dynamic cellular protein–RNA interactions by deep learning using in vivo RNA structures
a Motifs of different RBPs. Left, the PrismNet integrative motifs and sequence motifs downloaded from the ATtRACT database for indicated RBPs. Q-values were calculated using TOMTOM112 for motif similarity between the sequence component of PrismNet integrative motifs and ATtRACT motifs. A PrismNet integrative motif comprises a sequence component (upper region of each logo) and a structural component (lower regions of each logo), where “U” stands for an unpaired nucleotide and “P” stands for a paired nucleotide. Right, paired with each RBP at the left; the number of true positives among all matched binding sites for the indicated RBP based on transcriptome motif scanning using the PrismNet integrative motifs and the ATtRACT motifs. True positives were determined by comparing with eCLIP experiments. b Hierarchical clustering of integrative motifs of human RBPs with available functional annotations.113 Colors represent functions in the inner tree, structural preferences in the circle of PrismNet integrative motifs, and RBDs in the outer circles. RBPs with red fonts are discussed in the main text. For each RBP, usually multiple motifs were defined (STAR Methods), which are indicated with the number after each RBP.
