ORFeus: a computational method to detect programmed ribosomal frameshifts and other non-canonical translation events

Figures

• 

  Fig. 1. Processing ribo-seq data. A Ribo-seq involves experimentally generating reads (top), then computationally mapping sequenced reads to precise ribosome positions (bottom). Ideal data shows a clear triplet periodicity and can be offset to correspond to a position within the P-site of the ribosome by shifting the 5’ or 3’ end of all reads. B Realigned ribo-seq data from Wu et al [38] shows clear triplet periodicity in aggregate across all annotated ORFs in S. cerevisiae (top). This periodic pattern is not as obvious in individual ORFs (bottom). The ribo-seq signal shown here for PDA1, a housekeeping gene commonly used as a reference in gene expression studies, is representative of the signal observed across other genes in S. cerevisiae

• 

  Fig. 2. Inferring canonical ORFs from ribo-seq data. A ORFeus takes in mapped ribo-seq data, genome annotations, and genome sequences and returns predicted ORFs calculated using an HMM. B A simple HMM to model canonical translation includes states corresponding to the 5’UTR and 3’UTR as well as nucleotide 1 (blue), 2 (red), and 3 (grey) of each start codon (diamonds), sense codon (circles), and stop codon (squares). All nonzero transition probabilities are denoted with arrows. A schematic representation of the emission probabilities for each state are plotted below the corresponding states. C A more complex HMM includes states for each individual start, stop, and sense codon sequence. Idealized ribo-seq density for a transcript that undergoes canonical translation of a single ORF is depicted below the model. The input relative ribo-seq density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _i^t$$\end{document}ρit and nucleotide sequence \documentclass[12pt]{minimal} \usepackage{amsmath}…

• 

  Fig. 3. Inferring alternative ORFs from ribo-seq data. A A more complex HMM to model non-canonical translation includes additional states and transitions to represent upstream ORFs and downstream ORFs (yellow), programmed ribosomal frameshifting (blue and red), and stop codon read-through (purple). The probability of each type of event (denoted with a Greek letter) is set to reflect its expected frequency. B The full model can infer both canonical ORFs and alternative ORFs, including recoding events

• 

  Fig. 4. Model performance on known and simulated altORFs. A–E Well-characterized examples of alternative translation events are correctly inferred by the model. The annotated and predicted ORF are shown schematically above the real \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _i^t$$\end{document}ρit signal for each transcript. Sensitivity (F) and specificity (G) are shown for sequences simulated using parameters estimated from the Wu et al. S. cerevisiae data set [38]. Each point represents the mean sensitivity or specificity over 100 simulated sequences with the given mean ribo-seq coverage. H The number of ORFs with at least the given ribo-seq coverage drops off rapidly as the coverage threshold increases. The data shown are for the Wu et al. S. cerevisiae data set [38]