Representation learning-based recommendation models play a dominant role among recommendation techniques. However, most of the existing methods assume both historical interactions and embedding dimensions are independent of each other, and thus regrettably ignore the high-order interaction information among historical interactions and embedding dimensions. In this article, we propose a novel representation learning-based model called COMET (COnvolutional diMEnsion inTeraction), which simultaneously models the high-order interaction patterns among historical interactions and embedding dimensions. To be specific, COMET stacks the embeddings of historical interactions horizontally at first, which results in two “embedding maps”. In this way, internal interactions and dimensional interactions can be exploited by convolutional neural networks (CNN) with kernels of different sizes simultaneously. A fully connected multi-layer perceptron (MLP) is then applied to obtain two interaction vectors. Lastly, the representations of users and items are enriched by the learnt interaction vectors, which can further be used to produce the final prediction. Extensive experiments and ablation studies on various public implicit feedback datasets clearly demonstrate the effectiveness and rationality of our proposed method.

Epigenomics and transcriptomics data from high-throughput sequencing techniques such as RNA-seq and ChIP-seq have been successfully applied in predicting gene transcript expression. However, the locations of chromatin loops in the genome identified by techniques such as Chromatin Interaction Analysis with Paired End Tag sequencing (ChIA-PET) have never been used for prediction tasks. Here, we developed machine learning models to investigate if ChIA-PET could contribute to transcript and exon usage prediction. In doing so, we used a large set of transcription factors as well as ChIA-PET data. We developed different Gradient Boosting Trees models according to the different tasks with the integrated datasets from three cell lines, including GM12878, HeLaS3 and K562. We validated the models via 10-fold cross validation, chromosome-split validation and cross-cell validation. Our results show that both transcript and splicing-derived exon usage can be effectively predicted with at least 0.7512 and 0.7459 of accuracy, respectively, on all cell lines from all kinds of validations. Examining the predictive features, we found that RNA Polymerase II ChIA-PET was one of the most important features in both transcript and exon usage prediction, suggesting that chromatin loop anchors are predictive of both transcript and exon usage.

MOTIVATION: human microbes play a critical role in an extensive range of complex human diseases and become a new target in precision medicine. In silico methods of identifying microbe-disease associations not only can provide a deep insight into understanding the pathogenic mechanism of complex human diseases but also assist pharmacologists to screen candidate targets for drug development. However, the majority of existing approaches are based on linear models or label propagation, which suffers from limitations in capturing nonlinear associations between microbes and diseases. Besides, it is still a great challenge for most previous methods to make predictions for new diseases (or new microbes) with few or without any observed associations. RESULTS: in this work, we construct features for microbes and diseases by fully exploiting multiply sources of biomedical data, and then propose a novel deep learning framework of graph attention networks with inductive matrix completion for human microbe-disease association prediction, named GATMDA. To our knowledge, this is the first attempt to leverage graph attention networks for this important task. In particular, we develop an optimized graph attention network with talking-heads to learn representations for nodes (i.e. microbes and diseases). To focus on more important neighbours and filter out noises, we further design a bi-interaction aggregator to enforce representation aggregation of similar neighbours. In addition, we combine inductive matrix completion to reconstruct microbe-disease associations to capture the complicated associations between diseases and microbes. Comprehensive experiments on two data sets (i.e. HMDAD and Disbiome) demonstrated that our proposed model consistently outperformed baseline methods. Case studies on two diseases, i.e. asthma and inflammatory bowel disease, further confirmed the effectiveness of our proposed model of GATMDA. AVAILABILITY: python codes and data set are available at: https://github.com/yahuilong/GATMDA. CONTACT: [email protected].

The development of deep sequencing technologies has led to the discovery of novel transcripts. Many in silico methods have been developed to assess the coding potential of these transcripts to further investigate their functions. Existing methods perform well on distinguishing majority long noncoding RNAs (lncRNAs) and coding RNAs (mRNAs) but poorly on RNAs with small open reading frames (sORFs). Here, we present DeepCPP (deep neural network for coding potential prediction), a deep learning method for RNA coding potential prediction. Extensive evaluations on four previous datasets and six new datasets constructed in different species show that DeepCPP outperforms other state-of-the-art methods, especially on sORF type data, which overcomes the bottleneck of sORF mRNA identification by improving more than 4.31, 37.24 and 5.89% on its accuracy for newly discovered human, vertebrate and insect data, respectively. Additionally, we also revealed that discontinuous k-mer, and our newly proposed nucleotide bias and minimal distribution similarity feature selection method play crucial roles in this classification problem. Taken together, DeepCPP is an effective method for RNA coding potential prediction.

Long noncoding RNAs (lncRNAs) play significant roles in various physiological and pathological processes via their interactions with biomolecules like DNA, RNA and protein. The existing in silico methods used for predicting the functions of lncRNA mainly rely on calculating the similarity of lncRNA or investigating whether an lncRNA can interact with a specific biomolecule or disease. In this work, we explored the functions of lncRNA from a different perspective: we presented a tool for predicting the interaction biomolecule type for a given lncRNA. For this purpose, we first investigated the main molecular mechanisms of the interactions of lncRNA-RNA, lncRNA-protein and lncRNA-DNA. Then, we developed an ensemble deep learning model: lncIBTP (lncRNA Interaction Biomolecule Type Prediction). This model predicted the interactions between lncRNA and different types of biomolecules. On the 5-fold cross-validation, the lncIBTP achieves average values of 0.7042 in accuracy, 0.7903 and 0.6421 in macro-average area under receiver operating characteristic curve and precision-recall curve, respectively, which illustrates the model effectiveness. Besides, based on the analysis of the collected published data and prediction results, we hypothesized that the characteristics of lncRNAs that interacted with DNA may be different from those that interacted with only RNA.

BACKGROUND: Long non-coding RNAs (lncRNAs) play significant roles in varieties of physiological and pathological processes.The premise of the lncRNA functional study is that the lncRNAs are identified correctly. Recently, deep learning method like convolutional neural network (CNN) has been successfully applied to identify the lncRNAs. However, the traditional CNN considers little relationships among samples via an indirect way. RESULTS: Inspired by the Siamese Neural Network (SNN), here we propose a novel network named Class Similarity Network in coding RNA and lncRNA classification. Class Similarity Network considers more relationships among input samples in a direct way. It focuses on exploring the potential relationships between input samples and samples from both the same class and the different classes. To achieve this, Class Similarity Network trains the parameters specific to each class to obtain the high-level features and represents the general similarity to each class in a node. The comparison results on the validation dataset under the same conditions illustrate the superiority of our Class Similarity Network to the baseline CNN. Besides, our method performs effectively and achieves state-of-the-art performances on two test datasets. CONCLUSIONS: We construct Class Similarity Network in coding RNA and lncRNA classification, which is shown to work effectively on two different datasets by achieving accuracy, precision, and F1-score as 98.43%, 0.9247, 0.9374, and 97.54%, 0.9990, 0.9860, respectively.

Coronavirus Disease-19 (COVID-19) has led to global epidemics with high morbidity and mortality. However, there are currently no proven effective drugs targeting COVID19. Identifying drug-virus associations can not only provide insights into the understanding of drug-virus interaction mechanism, but also guide and facilitate the screening of compound candidates for antiviral drug discovery. In this work, we propose a novel framework of Heterogeneous Graph Attention Networks for Drug-Virus Association predictions, named HGATDVA. First, we fully incorporate multiple sources of biomedical data to construct abundant features for drugs and viruses. Second, we construct two drug-virus heterogeneous graphs. For each graph, we design a self-enhanced graph attention network (SGAT) to explicitly model the dependency between a node and its local neighbors and derive the graph-specific representations for nodes. Third, we further develop a neural network architecture with tri-aggregator to aggregate the graph-specific representations to generate the final node representations. Experiments on two datasets were conducted to demonstrate the effectiveness of our proposed method in identifying candidate drugs for viruses.

MOTIVATION: Influenza viruses are persistently threatening public health, causing annual epidemics and sporadic pandemics. The evolution of influenza viruses remains to be the main obstacle in the effectiveness of antiviral treatments due to rapid mutations. The goal of this work is to predict whether mutations are likely to occur in the next flu season using historical glycoprotein hemagglutinin sequence data. One of the major challenges is to model the temporality and dimensionality of sequential influenza strains and to interpret the prediction results. RESULTS: In this article, we propose an efficient and robust time-series mutation prediction model (Tempel) for the mutation prediction of influenza A viruses. We first construct the sequential training samples with splittings and embeddings. By employing recurrent neural networks with attention mechanisms, Tempel is capable of considering the historical residue information. Attention mechanisms are being increasingly used to improve the performance of mutation prediction by selectively focusing on the parts of the residues. A framework is established based on Tempel that enables us to predict the mutations at any specific residue site. Experimental results on three influenza datasets show that Tempel can significantly enhance the predictive performance compared with widely used approaches and provide novel insights into the dynamics of viral mutation and evolution. AVAILABILITY AND IMPLEMENTATION: The datasets, source code and supplementary documents are available at: https://drive.google.com/drive/folders/15WULR5__6k47iRotRPl3H7ghi3RpeNXH. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Long non-coding RNAs (lncRNAs) can exert functions via forming triplex with DNA. The current methods in predicting the triplex formation mainly rely on mathematic statistic according to the base paring rules. However, these methods have two main limitations: (1) they identify a large number of triplex-forming lncRNAs, but the limited number of experimentally verified triplex-forming lncRNA indicates that maybe not all of them can form triplex in practice, and (2) their predictions only consider the theoretical relationship while lacking the features from the experimentally verified data. RESULTS: In this work, we develop an integrated program named TriplexFPP (Triplex Forming Potential Prediction), which is the first machine learning model in DNA:RNA triplex prediction. TriplexFPP predicts the most likely triplex-forming lncRNAs and DNA sites based on the experimentally verified data, where the high-level features are learned by the convolutional neural networks. In the fivefold cross validation, the average values of Area Under the ROC curves and PRC curves for removed redundancy triplex-forming lncRNA dataset with threshold 0.8 are 0.9649 and 0.9996, and these two values for triplex DNA sites prediction are 0.8705 and 0.9671, respectively. Besides, we also briefly summarize the cis and trans targeting of triplexes lncRNAs. CONCLUSIONS: The TriplexFPP is able to predict the most likely triplex-forming lncRNAs from all the lncRNAs with computationally defined triplex forming capacities and the potential of a DNA site to become a triplex. It may provide insights to the exploration of lncRNA functions.