Modeling RNA-binding protein specificity in vivo by precisely registering protein-RNA crosslink sites

Abstract

RNA-binding proteins (RBPs) regulate post-transcriptional gene expression by recognizing short and degenerate sequence elements in their target transcripts. Despite the expanding list of RBPs with in vivo binding sites mapped genomewide using crosslinking and immunoprecipitation (CLIP), defining precise RBP binding specificity remains challenging. We previously demonstrated that the exact protein-RNA crosslink sites can be mapped using CLIP data at single-nucleotide resolution and observed that crosslinking frequently occurs at specific positions in RBP motifs. Here we have developed a computational method, named mCross, to jointly model RBP binding specificity while precisely registering the crosslinking position in motif sites. We applied mCross to 112 RBPs using ENCODE eCLIP data and validated the reliability of the resulting motifs by genomewide analysis of allelic binding sites also detected by CLIP. We found that the prototypical SR protein SRSF1 recognizes GGA clusters to regulate splicing in a much larger repertoire of transcripts than previously appreciated.

I-KCKT allows dissection-free RNA profiling of adultDrosophilaintestinal progenitor cells

“…In addition, we noted that this analysis identified several confirmed direct targets of Drosophila FMRP found by genetic or targeted biochemical means, including CaMKII, chic, Dscam1, futsch, ninaE, rg and tral (Cvetkovska et al, 2013;Kim et al, 2013;Monzo et al, 2010;Reeve et al, 2005;Sears et al, 2019;Sudhakaran et al, 2014;Wang et al, 2017;Zhang et al, 2001). As the first CLIP-based assay to characterize Drosophila FMRP, this analysis also identified CAUUG(A/U) as its top binding motif (Fig 5C), consistent with prior identification of the AUUG sequence in one of the top binding motifs of human FXR1 (Feng et al, 2019).…”

Alternative splicing programming of axon formation

“…Once important RBPs are identified, the regulatory relationships between RBPs and alternative splicing events can be assessed transcriptome‐wide with various experimental and computational tools, including but not limited to CLIP‐Seq, RNA‐Seq, and functional screens of splicing reporters, etc. (Chen & Zheng, ; Feng et al, ; Lee & Ule, ; Ule, Hwang, & Darnell, ; Wheeler, Van Nostrand, & Yeo, ; Zheng, ). Even though not every single target of an RBP is important, genetic interactions between its coordinated targets are presumably more possible than between two random alternative exons, which help prioritize exons of interest.…”

Big data analytics in bioinformatics

“…DeepCLIP produces motifs of varying sizes ranked by the average information content (Figure 2a, b). The top-ranking motifs of DeepCLIP for the analyzed proteins were visually remarkably similar to core binding sites as described in literature (67,68,72,(78)(79)(80)(81)(82) (Figure S6). The motifs depicted in Figure S6 are based on the top-1000 scoring sequences from the positive and negative input dataset and represent the two pseudo-PFMs with the highest mean information score among the five pseudo-PFMs produced.…”

“…We ran 10-fold cross-validation ( Figure S32 and S33) using the same input parameters as previously, and extracted the two best performing models. Their pseudo-PFMs are shown in Figure 2a and Figure 2b and they show high visual agreement with previously published motifs of hnRNP A1 and SRSF1 (60,69,71,72). We used the models to predict binding of hnRNP A1 and SRSF1, respectively.…”

“…The "deep" in deep learning refers to the fact that these algorithms typically have deep network graphs. Depth in this context represents the length of the longest path from input to output [70,72,73]. In the following, deep neural networks will also be referred to as neural networks.…”