The superiority of graph contrastive learning (GCL) has prompted its application to anomaly detection tasks for more powerful risk warning systems. Unfortunately, existing GCL-based models tend to excessively prioritize overall detection performance while neglecting robustness to structural imbalance, which can be problematic for many real-world networks following power-law degree distributions. Particularly, GCL-based methods may fail to capture tail anomalies (abnormal nodes with low degrees). This raises concerns about the security and robustness of current anomaly detection algorithms and therefore hinders their applicability in a variety of realistic high-risk scenarios. To the best of our knowledge, research on the robustness of graph anomaly detection to structural imbalance has received little scrutiny. To address the above issues, this paper presents a novel GCL-based framework named AD-GCL. It devises the neighbor pruning strategy to filter noisy edges for head nodes and facilitate the detection of genuine tail nodes by aligning from head nodes to forged tail nodes. Moreover, AD-GCL actively explores potential neighbors to enlarge the receptive field of tail nodes through anomaly-guided neighbor completion. We further introduce intra- and inter-view consistency loss of the original and augmentation graph for enhanced representation. The performance evaluation of the whole, head, and tail nodes on multiple datasets validates the comprehensive superiority of the proposed AD-GCL in detecting both head anomalies and tail anomalies.

The widespread application of graph data in various high-risk scenarios has increased attention to graph anomaly detection (GAD). Faced with real-world graphs that often carry node descriptions in the form of raw text sequences, termed text-attributed graphs (TAGs), existing graph anomaly detection pipelines typically involve shallow embedding techniques to encode such textual information into features, and then rely on complex self-supervised tasks within the graph domain to detect anomalies. However, this text encoding process is separated from the anomaly detection training objective in the graph domain, making it difficult to ensure that the extracted textual features focus on GAD-relevant information, seriously constraining the detection capability. How to seamlessly integrate raw text and graph topology to unleash the vast potential of cross-modal data in TAGs for anomaly detection poses a challenging issue. This paper presents a novel end-to-end paradigm for text-attributed graph anomaly detection, named CMUCL. We simultaneously model data from both text and graph structures, and jointly train text and graph encoders by leveraging cross-modal and uni-modal multi-scale consistency to uncover potential anomaly-related information. Accordingly, we design an anomaly score estimator based on inconsistency mining to derive node-specific anomaly scores. Considering the lack of benchmark datasets tailored for anomaly detection on TAGs, we release 8 datasets to facilitate future research. Extensive evaluations show that CMUCL significantly advances in text-attributed graph anomaly detection, delivering an 11.13% increase in average accuracy (AP) over the suboptimal.

Most self-supervised graph learning studies typically follow an offline training paradigm, assuming that all data are readily available.This assumption, however, is not always tenable in real-world scenarios as many graph data are generated continuously.Although several continual graph learning models have emerged and achieved empirical success, they almost all rely on external supervision, making it difficult to adapt to applications with a large amount of unlabeled data from the wild.To be honest, research on self-supervised continual graph learning is still surprisingly in its infancy.Therefore, we select several well-known self-supervised graph embedding models as representatives and explore whether they are resistant to catastrophic forgetting in a continual learning setting.Empirical studies find that self-supervised representation models may be potentially better continual learners than supervised counterparts.Driven by this advantage, we propose a self-supervised continual graph representation learning framework based on adaptive spaced replay on node proxies, named Trace.Inspired by the Complementary Learning System theory, Trace employs a dual-system architecture to simulate the functionality and cooperation of the hippocampus and neocortex in the brain.Among them, the fastlearning system efficiently encodes the current input graph to acquire new knowledge and adaptively extracts node proxies from it as important knowledge cached into the memory through progressive clustering.Drawing inspiration from the Ebbinghaus forgetting curve, the slow-learning system implements adaptive spaced replay based on the memory retention rate of each preceding task instead of the widely used consecutive replay scheme for promising flexibility and efficiency.Experiments under task-incremental and class-incremental learning settings on multiple datasets corroborate

MicroRNA (miRNA) is not a single sequence, but a series of multiple variants (also termed isomiRs) with sequence and expression heterogeneity. Whether and how these isoforms contribute to functional variation and complexity at the systems and network levels remain largely unknown. To explore this question systematically, we comprehensively analyzed the expression of small RNAs and their target sites to interrogate functional variations between novel isomiRs and their canonical miRNA sequences. Our analyses of the pan-cancer landscape of miRNA expression indicate that multiple isomiRs generated from the same miRNA locus often exhibit remarkable variation in their sequence, expression and function. We interrogated abundant and differentially expressed 5' isomiRs with novel seed sequences via seed shifting and identified many potential novel targets of these 5' isomiRs that would expand interaction capabilities between small RNAs and mRNAs, rewiring regulatory networks and increasing signaling circuit complexity. Further analyses revealed that some miRNA loci might generate diverse dominant isomiRs that often involved isomiRs with varied seeds and arm-switching, suggesting a selective advantage of multiple isomiRs in regulating gene expression. Finally, experimental validation indicated that isomiRs with shifted seed sequences could regulate novel target mRNAs and therefore contribute to regulatory network rewiring. Our analysis uncovers a widespread expansion of isomiR and mRNA interaction networks compared with those seen in canonical small RNA analysis; this expansion suggests global gene regulation network perturbations by alternative small RNA variants or isoforms. Taken together, the variations in isomiRs that occur during miRNA processing and maturation are likely to play a far more complex and plastic role in gene regulation than previously anticipated.

SUMMARY: The R/Bioconductor package primirTSS is a fast and convenient tool that allows implementation of the analytical method to identify transcription start sites of microRNAs by integrating ChIP-seq data of H3K4me3 and Pol II. It further ensures the precision by employing the conservation score and sequence features. The tool showed a good performance when using H3K4me3 or Pol II Chip-seq data alone as input, which brings convenience to applications where multiple datasets are hard to acquire. This flexible package is provided with both R-programming interfaces as well as graphical web interfaces. AVAILABILITY AND IMPLEMENTATION: primirTSS is available at: http://bioconductor.org/packages/primirTSS. The documentation of the package including an accompanying tutorial was deposited at: https://bioconductor.org/packages/release/bioc/vignettes/primirTSS/inst/doc/primirTSS.html. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Identification of microRNA (miRNA) transcriptional start sites (TSSs) is crucial to understand the transcriptional regulation of miRNA. As miRNA expression is highly cell specific, an automatic and systematic method that could identify miRNA TSSs accurately and cell specifically is in urgent requirement. RESULTS: A workflow to identify the TSSs of miRNAs was built by integrating the data of H3K4me3 and DNase I hypersensitive sites as well as combining the conservation level and sequence feature. By applying the workflow to the data for 54 cell lines from the ENCODE project, we successfully identified TSSs for 663 intragenic miRNAs and 620 intergenic miRNAs, which cover 84.2% (1283/1523) of all miRNAs recorded in miRBase 18. For these cell lines, we found 4042 alternative TSSs for intragenic miRNAs and 3186 alternative TSSs for intergenic miRNAs. Our method achieved a better performance than the previous non-cell-specific methods on miRNA TSSs. The cell-specific method developed by Georgakilas et al. gives 158 TSSs of higher accuracy in two cell lines, benefitting from the employment of deep-sequencing technique. In contrast, our method provided a much higher number of miRNA TSSs (7228) for a broader range of cell lines without the limitation of costly deep-sequencing data, thus being more applicable for various experimental cases. Analysis showed that upstream promoters at - 2 kb to - 200 bp of TSS are more conserved for independently transcribed miRNAs, while for miRNAs transcribed with host genes, their core promoters (-200 bp to 200 bp of TSS) are significantly conserved. AVAILABILITY AND IMPLEMENTATION: Predicted miRNA TSSs and promoters can be downloaded from supplementary files. CONTACT: [email protected] or [email protected] or [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Transcription factors (TFs) are important regulatory proteins that govern transcriptional regulation. Today, it is known that in higher organisms different TFs have to cooperate rather than acting individually in order to control complex genetic programs. The identification of these interactions is an important challenge for understanding the molecular mechanisms of regulating biological processes. In this study, we present a new method based on pointwise mutual information, PC-TraFF, which considers the genome as a document, the sequences as sentences, and TF binding sites (TFBSs) as words to identify interacting TFs in a set of sequences. RESULTS: To demonstrate the effectiveness of PC-TraFF, we performed a genome-wide analysis and a breast cancer-associated sequence set analysis for protein coding and miRNA genes. Our results show that in any of these sequence sets, PC-TraFF is able to identify important interacting TF pairs, for most of which we found support by previously published experimental results. Further, we made a pairwise comparison between PC-TraFF and three conventional methods. The outcome of this comparison study strongly suggests that all these methods focus on different important aspects of interaction between TFs and thus the pairwise overlap between any of them is only marginal. CONCLUSIONS: In this study, adopting the idea from the field of linguistics in the field of bioinformatics, we develop a new information theoretic method, PC-TraFF, for the identification of potentially collaborating transcription factors based on the idiosyncrasy of their binding site distributions on the genome. The results of our study show that PC-TraFF can succesfully identify known interacting TF pairs and thus its currently biologically uncorfirmed predictions could provide new hypotheses for further experimental validation. Additionally, the comparison of the results of PC-TraFF with the results of previous methods demonstrates that different methods with their specific scopes can perfectly supplement each other. Overall, our analyses indicate that PC-TraFF is a time-efficient method where its algorithm has a tractable computational time and memory consumption. The PC-TraFF server is freely accessible at http://pctraff.bioinf.med.uni-goettingen.de/.

SUMMARY: The great variety of human cell types in morphology and function is due to the diverse gene expression profiles that are governed by the distinctive regulatory networks in different cell types. It is still a challenging task to explain how the regulatory networks achieve the diversity of different cell types. Here, we report on our studies of the design principles of the tissue regulatory system by constructing the regulatory networks of eight human tissues, which subsume the regulatory interactions between transcription factors (TFs), microRNAs (miRNAs) and non-TF target genes. The results show that there are in-/out-hubs of high in-/out-degrees in tissue networks. Some hubs (strong hubs) maintain the hub status in all the tissues where they are expressed, whereas others (weak hubs), in spite of their ubiquitous expression, are hubs only in some tissues. The network motifs are mostly feed-forward loops. Some of them having no miRNAs are the common motifs shared by all tissues, whereas the others containing miRNAs are the tissue-specific ones owned by one or several tissues, indicating that the transcriptional regulation is more conserved across tissues than the post-transcriptional regulation. In particular, a common bow-tie framework was found that underlies the motif instances and shows diverse patterns in different tissues. Such bow-tie framework reflects the utilization efficiency of the regulatory system as well as its high variability in different tissues, and could serve as the model to further understand the structural adaptation of the regulatory system to the specific requirements of different cell functions. CONTACT: [email protected]; [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Current approaches for identifying transcriptional regulatory elements are mainly via the combination of two properties, the evolutionary conservation and the overrepresentation of functional elements in the promoters of co-regulated genes. Despite the development of many motif detection algorithms, the discovery of conserved motifs in a wide range of phylogenetically related promoters is still a challenge, especially for the short motifs embedded in distantly related gene promoters or very closely related promoters, or in the situation that there are not enough orthologous genes available. RESULTS: A mutation degree model is proposed and a new word counting method is developed for the identification of transcriptional regulatory elements from a set of co-expressed genes. The new method comprises two parts: 1) identifying overrepresented oligo-nucleotides in promoters of co-expressed genes, 2) estimating the conservation of the oligo-nucleotides in promoters of phylogenetically related genes by the mutation degree model. Compared with the performance of other algorithms, our method shows the advantages of low false positive rate and higher specificity, especially the robustness to noisy data. Applying the method to co-expressed gene sets from Arabidopsis, most of known cis-elements were successfully detected. The tool and example are available at http://mcube.nju.edu.cn/jwang/lab/soft/ocw/OCW.html. CONCLUSIONS: The mutation degree model proposed in this paper is adapted to phylogenetic data of different qualities, and to a wide range of evolutionary distances. The new word-counting method based on this model has the advantage of better performance in detecting short sequence of cis-elements from co-expressed genes of eukaryotes and is robust to less complete phylogenetic data.