Joint modeling of RNase footprint sequencing profiles for genome-wide inference of RNA structure.

摘要: Recent studies have revealed significant roles of RNA structure in almost every step processing, including transcription, splicing, transport and translation. RNase footprint sequencing (RNase-seq) has emerged to dissect structures at the genome scale. However, it remains challenging analyze RNase-seq data because issues signal sparsity, variability correlations among various RNases. We present a probabilistic framework, joint Poisson-gamma mixture (JPGM), for integrative modeling multiple profiles. Combining JPGM with hidden Markov model allows genome-wide inference structures. apply approach inferring base pairing states on simulated sets profiles double-strand specific V1 single-strand S1 yeast. demonstrate that analysis outputs interpretable states, while approaches each profile separately do not. The predicts all nucleotides 3196 transcripts yeast without compromising accuracy, simple thresholding misses 43% nucleotides. Furthermore, posterior probabilities outputted by our are able resolve structural ambiguity ≈300 000 overlapping cleavage sites. Our also generates accessibilities, which associated three-dimensional conformations.