Machine learning methods have been widely applied to big data analysis in genomics and epigenomics research. Although accuracy and efficiency are common goals in many modeling tasks, model interpretability is especially important to these studies towards understanding the underlying molecular and cellular mechanisms. Deep neural networks (DNNs) have recently gained popularity in various types of genomic and epigenomic studies due to their capabilities in utilizing large-scale high-throughput bioinformatics data and achieving high accuracy in predictions and classifications. However, DNNs are often challenged by their potential to explain the predictions due to their black-box nature. In this review, we present current development in the model interpretation of DNNs, focusing on their applications in genomics and epigenomics. We first describe state-of-the-art DNN interpretation methods in representative machine learning fields. We then summarize the DNN interpretation methods in recent studies on genomics and epigenomics, focusing on current data- and computing-intensive topics such as sequence motif identification, genetic variations, gene expression, chromatin interactions and non-coding RNAs. We also present the biological discoveries that resulted from these interpretation methods. We finally discuss the advantages and limitations of current interpretation approaches in the context of genomic and epigenomic studies. Contact:[email protected], [email protected].

MOTIVATION: MicroRNAs (miRNAs) are small noncoding RNAs that play important roles in gene regulation and phenotype development. The identification of miRNA transcription start sites (TSSs) is critical to understand the functional roles of miRNA genes and their transcriptional regulation. Unlike protein-coding genes, miRNA TSSs are not directly detectable from conventional RNA-Seq experiments due to miRNA-specific process of biogenesis. In the past decade, large-scale genome-wide TSS-Seq and transcription activation marker profiling data have become available, based on which, many computational methods have been developed. These methods have greatly advanced genome-wide miRNA TSS annotation. RESULTS: In this study, we summarized recent computational methods and their results on miRNA TSS annotation. We collected and performed a comparative analysis of miRNA TSS annotations from 14 representative studies. We further compiled a robust set of miRNA TSSs (RSmirT) that are supported by multiple studies. Integrative genomic and epigenomic data analysis on RSmirT revealed the genomic and epigenomic features of miRNA TSSs as well as their relations to protein-coding and long non-coding genes. CONTACT: [email protected], [email protected].

The first version of this text did not label Saidi Wang and Amlan Talukder as equally contributing authors or Xiaoman Li and Haiyan Hu as co-corresponding authors, and also included an incorrect link to the supplementary material. This has now been corrected.

The computational identification of long non-coding RNAs (lncRNAs) is important to study lncRNAs and their functions. Despite the existence of many computation tools for lncRNA identification, to our knowledge, there is no systematic evaluation of these tools on common datasets and no consensus regarding their performance and the importance of the features used. To fill this gap, in this study, we assessed the performance of 17 tools on several common datasets. We also investigated the importance of the features used by the tools. We found that the deep learning-based tools have the best performance in terms of identifying lncRNAs, and the peptide features do not contribute much to the tool accuracy. Moreover, when the transcripts in a cell type were considered, the performance of all tools significantly dropped, and the deep learning-based tools were no longer as good as other tools. Our study will serve as an excellent starting point for selecting tools and features for lncRNA identification.

MOTIVATION: It is a fundamental task to identify microRNAs (miRNAs) targets and accurately locate their target sites. Genome-scale experiments for miRNA target site detection are still costly. The prediction accuracies of existing computational algorithms and tools are often not up to the expectation due to a large number of false positives. One major obstacle to achieve a higher accuracy is the lack of knowledge of the target binding features of miRNAs. The published high-throughput experimental data provide an opportunity to analyze position-wise preference of miRNAs in terms of target binding, which can be an important feature in miRNA target prediction algorithms. RESULTS: We developed a Markov model to characterize position-wise pairing patterns of miRNA-target interactions. We further integrated this model as a scoring method and developed a dynamic programming (DP) algorithm, MDPS (Markov model-scored Dynamic Programming algorithm for miRNA target site Selection) that can screen putative target sites of miRNA-target binding. The MDPS algorithm thus can take into account both the dependency of neighboring pairing positions and the global pairing information. Based on the trained Markov models from both miRNA-specific and general datasets, we discovered that the position-wise binding information specific to a given miRNA would benefit its target prediction. We also found that miRNAs maintain region-wise similarity in their target binding patterns. Combining MDPS with existing methods significantly improves their precision while only slightly reduces their recall. Therefore, position-wise pairing patterns have the promise to improve target prediction if incorporated into existing software tools. AVAILABILITY AND IMPLEMENTATION: The source code and tool to calculate MDPS score is available at http://hulab.ucf.edu/research/projects/MDPS/index.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: The identification of enhancer-promoter interactions (EPIs), especially condition-specific ones, is important for the study of gene transcriptional regulation. Existing experimental approaches for EPI identification are still expensive, and available computational methods either do not consider or have low performance in predicting condition-specific EPIs. RESULTS: We developed a novel computational method called EPIP to reliably predict EPIs, especially condition-specific ones. EPIP is capable of predicting interactions in samples with limited data as well as in samples with abundant data. Tested on more than eight cell lines, EPIP reliably identifies EPIs, with an average area under the receiver operating characteristic curve of 0.95 and an average area under the precision-recall curve of 0.73. Tested on condition-specific EPIPs, EPIP correctly identified 99.26% of them. Compared with two recently developed methods, EPIP outperforms them with a better accuracy. AVAILABILITY AND IMPLEMENTATION: The EPIP tool is freely available at http://www.cs.ucf.edu/˜xiaoman/EPIP/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.