RBPamp: Quantitative Modeling of Protein-RNA Interactionsin vitroPredictsin vivoBinding

Abstract

RNA-binding proteins (RBPs) control the processing and function of cellular transcripts to effect post-transcriptional gene regulation. Sequence-specific binding of RBPs to millions of synthetic RNAs has been probed in vitro by RNA Bind-n-Seq (RBNS). Here we describe RBPamp, a bio-physically-based model of protein-RNA interactions and associated algorithm that inferred affinity spectra of 79 diverse human RBPs from RBNS data. RBPamp supports multiple motifs per RBP, models RBP concentration and binding site saturation, and accounts for the effects of RNA sec-ondary structure. RBPamp affinities along transcripts are predictive of in vivo binding, as meas-ured by eCLIP density. For many RBPs, average local eCLIP density increases monotonically with predicted affinity, and the shape of this relationship can suggest free protein concentrations and potential cooperativity. Together, these analyses demonstrate a powerful integrative approach for the quantitative dissection of RBP function.

Quantifying negative selection in human 3’ UTRs uncovers constrained targets of RNA-binding proteins

“…We also find that RBPamp affinity models (Jens et al 2022) derived from in vitro RBNS data (Lambert et al 2014; Dominguez et al 2018) are insufficient to independently identify constrained RBP binding sites (Fig. 2E).…”

“…We first focused on general RBP binding sites in 3’ UTRs, with the goal of identifying precise/short RBP binding sites of high confidence using complementary orthogonal methods: for each RBP with available data, we identified the highest affinity RBPamp motif (based on high throughput in vitro RNA Bind-n-Seq (RBNS) data (Dominguez et al 2018; Jens et al 2022) in the vicinity of each of the more than 25,000 ENCODE eCLIP peaks (Van Nostrand et al 2020) and termed these sites “RBPamp eCLIP-Proximal” or “ReP” sites (Fig. 2A).…”

“…Position-specific affinity matrices (PSAMs) of length k = 10 or k = 11 generated by RBPamp (Jens et al 2022) were used to score the affinity of a given RBP for potential target RNA sequences relative to the RBP's ideal binding site (affinity = 1.0). We used PSAMs that were generated by considering both the sequence and predicted secondary structure of random sequence target RNA oligos in RBNS experiments.…”

Systematic identification of disease-causing promoter and untranslated region variants in 8,040 undiagnosed individuals with rare disease

“…RNA binding protein predictions were generated using the methods detailed in Findlay et al ( 38 ) for all possible variants within motifs that are proximal to ENCODE eCLIP sites, that are also high affinity sites as predicted by RBPamp(39). These were intersected with our variants and filtered to retain only those with a reference affinity of ≥ 0.1 and with an impact of ‘loss of binding’ predicted by the RBP binding affinity model (defined as alternative allele affinity / reference allele affinity < ⅓).…”

An interpretable model of pre-mRNA splicing for animal and plant genes

“…Likewise, dozens of SFs have been extensively studied biochemically and genetically (14). In vitro binding preferences of RNA-binding proteins (RBPs), including many SFs, have been mapped (15,16), and tools for modeling binding have been developed (17,18). Recently, in vivo binding and splicing activity following RNA interference knockdown have been assessed for dozens of SFs (19).…”