1 - 19 of 19 results


An approach based on deep convolutional neural networks. DeepBind can discover new patterns even when the locations of patterns within sequences are unknown—a task for which traditional neural networks require an exorbitant amount of training data. DeepBind addresses the above challenges: (i) It can be applied to both microarray and sequencing data; (ii) it can learn from millions of sequences through parallel implementation on a graphics processing unit (GPU); (iii) it generalizes well across technologies, even without correcting for technology-specific biases; (iv) it can tolerate a moderate degree of noise and mislabeled training data; and (v) it can train predictive models fully automatically, alleviating the need for careful and time-consuming hand-tuning.

BEAM / BEAr Motif finder

A method for structural motif discovery from a set of unaligned RNAs. BEAM explores sets of unaligned RNAs sharing a biological property, looking for the most represented local secondary structure motifs, and evaluating their significance with respect to a common background. It takes advantage of a recently developed encoding for RNA secondary structure named BEAR (Brand nEw Alphabet for RNAs) and of evolutionary substitution rates of secondary structure elements. BEAM is successful in retrieving structural motifs even in highly noisy data sets, such as those that can arise in CLIP-Seq or other high-throughput experiments.


A deep learning framework to model the binding preferences of RNA-binding proteins (RBPs) by integrating the primary sequence, predicted secondary and tertiary structural profiles of the target sites. Our framework considered RNA tertiary structure for RBP binding site prediction, and provided strong evidence to support the view that the RNA tertiary structural features can contribute to RBP target recognition. Tests on real CLIP-seq datasets showed that our framework can achieve the comparable or superior performance to the state-of-the-art method for predicting RBP binding sites.


Calculates the probable RNA secondary structure through computational folding. RNApeakFold is an RNA secondary structure prediction tool tailored to CLIP peaks. It first computes base pairing probabilities for the sequence including the peak and +/- 100 flanking nucleotides using RNAfold–p and then, by using those probabilities as energies in an implementation of Nussinov folding, determines the most probable structure of the peak region without the flanking nucleotides. The software is included in the tool SARNAclust, which makes use of it.

MEMERIS / Multiple Em for Motif Elucidation in Rna’s Including secondary Structures

A computational method for searching sequence motifs in a set of RNA sequences and simultaneously integrating information about secondary structures. MEMERIS precomputes values that characterize the single-strandedness of all putative motif occurrences. These values are then used to guide the motif search towards single-stranded regions. Authors conclude that MEMERIS preferably selects single stranded motif occurrences and that it is able to identify a weaker over a stronger motif if the average single strandedness is sufficiently higher.

DeMo / Deep Motif

Provides a convolutional/highway multilayer perceptron (MLP) network. DeMo DeMo is a generic model for visualizing sequence classification tasks. It outperforms the state-of-the-art baseline for 92 different transcription factors binding sites (TFBSs) datasets, as well as generate motifs, or interpretable patterns that represent the important transcription factor binding patterns. This model is applicable to other sequence classification tasks which demand a visual interpretation of the classes.

DeBooster / Deep Boost

A framework to predict the sequence specificities of RNA-binding proteins from high-throughput CLIP-seq data. DeBooster provides an effective index to identify pathogenic mutations from normal sequence variants and study the effects of potential disease-causing mutations in RNA-binding proteins binding sites related to splicing. DeBooster may provide a practically useful tool to analyze high-throughput CLIP-seq data and recover false negatives that are common in current CLIP-seq data.