DeeReCT-PolyA: a robust and generic deep learning method for PAS identification

Abstract

Motivation Polyadenylation is a critical step for gene expression regulation during the maturation of mRNA. An accurate and robust method for poly(A) signals (PASs) identification is not only desired for the purpose of better transcripts’ end annotation, but can also help us gain a deeper insight of the underlying regulatory mechanism. Although many methods have been proposed for PAS recognition, most of them are PAS motif- and human-specific, which leads to high risks of overfitting, low generalization power, and inability to reveal the connections between the underlying mechanisms of different mammals. Results In this work, we propose a robust, PAS motif agnostic, and highly interpretable and transferrable deep learning model for accurate PAS recognition, which requires no prior knowledge or human-designed features. We show that our single model trained over all human PAS motifs not only outperforms the state-of-the-art methods trained on specific motifs, but can also be generalized well to two mouse datasets. Moreover, we further increase the prediction accuracy by transferring the deep learning model trained on the data of one species to the data of a different species. Several novel underlying poly(A) patterns are revealed through the visualization of important oligomers and positions in our trained models. Finally, we interpret the deep learning models by converting the convolutional filters into sequence logos and quantitatively compare the sequence logos between human and mouse datasets. Availability and implementation https://github.com/likesum/DeeReCT-PolyA Supplementary information Supplementary data are available at Bioinformatics online.

A deep dense inception network for protein beta‐turn prediction

“…The pretrained network is effective in learning some more general beta‐turn features; then the transfer learning technique can transfer the base network to some more specific models that can classify nine‐class beta‐turns. We have also demonstrated some techniques for tuning deep neural networks on small data classification problems, which may be useful in other areas of biological sequence analyses with imbalanced data sets, such as genomic analysis, poly‐signal identification, post‐translational modification prediction, and so forth.…”

Terminitor: Cleavage Site Prediction Using Deep Learning Models

“…Another angle to address the prediction problem is through machine learning (ML) approaches, and several such models have been proposed. These include more traditional ML methods like support vector machine and Hidden Markov Model, but also the latest deep learning models, such as DeeReCT-PolyA and DeepGSR (23)(24)(25)(26). Deep learning models generally outperform traditional classifiers because they abandon manually selected features.…”

“…Raw sequences are directly fed in and hidden features may be learnt and modeled. The latter two tools mentioned above have been reported to show around 90% accuracy on test sequences (23,24). However, these models are only trained to distinguish whether a poly(A) signal (PAS, a conserved hexamer motif) is real, but in reality, a small proportion of functional human poly(A) site does not require PAS (27,28).…”

Deep learning in bioinformatics: Introduction, application, and perspective in the big data era

“…There are also a number of breakthroughs in using deep learning to perform biomedical image processing and biomedical diagnosis. For example, [35] proposes a method based on deep neural networks, which can reach dermatologist-level performance in classifying skin cancer; [66] uses transfer learning to solve the data-hungry problem to promote the automatic medical diagnosis; [22] proposes a deep learning method to automatically predict fluorescent labels from transmitted-light images of unlabeled biological samples; [41,160] also propose deep learning methods to analyze 1D data CNN, RNN [198,3,25,156,87,6,80,157,158,175,169] Structure prediction and reconstruction MRI images, Cryo-EM images, fluorescence microscopy images, protein contact map 2D data CNN, GAN, VAE [167,90,38,168,180,196,170] Biomolecular property and function prediction Sequencing data, PSSM, structure properties, microarray gene expression 1D data, 2D data, structured data DNN, CNN, RNN [85,204,75,4] Biomedical image processing and diagnosis CT images, PET images, MRI images 2D data CNN, GAN [35,66,41,22,160] Biomolecule interaction prediction and systems biology Microarray gene expression, PPI, gene-disease interaction, diseasedisease similarity network, diseasevariant network 1D data, 2D data, structured data, graph data CNN, GCN [95,201,203,165,71,…”