scLM: Automatic Detection of Consensus Gene Clusters Across Multiple Single-cell Datasets

In gene expression profiling studies, including single-cell RNA sequencing(sc RNA-seq)analyses, the identification and characterization of co-expressed genes provides critical information on cell identity and function. Gene co-expression clustering in sc RNA-seq data presents certain challenges. We show that commonly used methods for single-cell data are not capable of identifying co-expressed genes accurately, and produce results that substantially limit biological expectations of co-expressed genes. Herein, we present single-cell Latent-variable Model(sc LM), a gene coclustering algorithm tailored to single-cell data that performs well at detecting gene clusters with significant biologic context. Importantly, sc LM can simultaneously cluster multiple single-cell datasets, i.e., consensus clustering, enabling users to leverage single-cell data from multiple sources for novel comparative analysis. sc LM takes raw count data as input and preserves biological variation without being influenced by batch effects from multiple datasets. Results from both simulation data and experimental data demonstrate that sc LM outperforms the existing methods with considerably improved accuracy. To illustrate the biological insights of sc LM, we apply it to our in-house and public experimental sc RNA-seq datasets. sc LM identifies novel functional gene modules and refines cell states, which facilitates mechanism discovery and understanding of complex biosystems such as cancers. A user-friendly R package with all the key features of the sc LM method is available at https://github.com/QSong-github/sc LM.     

Computational discovery of regulatory elements in a continuous expression space

Approaches for regulatory element discovery from gene expression data usually rely on clustering algorithms to partition the data into clusters of co-expressed genes. Gene regulatory sequences are then mined to find overrepresented motifs in each cluster. However, this ad hoc partition rarely fits the biological reality. We propose a novel method called RED² that avoids data clustering by estimating motif densities locally around each gene. We show that RED² detects numerous motifs not detected by clustering-based approaches, and that most of these correspond to characterized motifs. RED² can be accessed online through a user-friendly interface.     

Approaches for extracting practical information from gene co-expression networks in plant biology

Gene co-expression, in many cases, implies the presence of a functional linkage between genes. Co-expression analysis has uncovered gene regulatory mechanisms in model organisms such as Escherichia coli and yeast. Recently, accumulation of Arabidopsis microarray data has facilitated a genome-wide inspection of gene co-expression profiles in this model plant. An approach using network analysis has provided an intuitive way to represent complex co-expression patterns between many genes. Co-expression network analysis has enabled us to extract modules, or groups of tightly co-expressed genes, associated with biological processes. Furthermore, integrated analysis of gene expression and metabolite accumulation has allowed us to hypothesize the functions of genes associated with specific metabolic processes. Co-expression network analysis is a powerful approach for data-driven hypothesis construction and gene prioritization, and provides novel insights into the system-level understanding of plant cellular processes.     

Application of MUTIC to the exploration of gene expression data in prostate cancer

We show here an example of the application of a novel method, MUTIC (model utilization-based clustering), used for identifying complex interactions between genes or gene categories based on gene expression data. The method deals with binary categorical data which consist of a set of gene expression profiles divided into two biologically meaningful categories. It does not require data from multiple time points. Gene expression profiles are represented by feature vectors whose component features are either gene expression values, or averaged expression values corresponding to gene ontology or protein information resource categories. A supervised learning algorithm (genetic programming) is used to learn an ensemble of classification models distinguishing the two categories based on the feature vectors corresponding to their members. Each feature is associated with a "model utilization vector", which has an entry for each high-quality classification model found, indicating whether or not the feature was used in that model. These utilization vectors are then clustered using a variant of hierarchical clustering called Omniclust. The result is a set of model utilization-based clusters, in which features are gathered together if they are often considered together by classification models - which may be because they are co-expressed, or may be for subtler reasons involving multi-gene interactions. The MUTIC method is illustrated here by applying it to a dataset regarding gene expression in prostate cancer and control samples. Compared to traditional expression-based clustering, MUTIC yields clusters that have higher mathematical quality (in the sense of homogeneity and separation) and that also yield novel insights into the underlying biological processes.     

CAFET algorithm reveals Wnt/PCP signature in lung squamous cell carcinoma

We analyzed the gene expression patterns of 138 Non-Small Cell Lung Cancer (NSCLC) samples and developed a new algorithm called Coverage Analysis with Fisher's Exact Test (CAFET) to identify molecular pathways that are differentially activated in squamous cell carcinoma (SCC) and adenocarcinoma (AC) subtypes. Analysis of the lung cancer samples demonstrated hierarchical clustering according to the histological subtype and revealed a strong enrichment for the Wnt signaling pathway components in the cluster consisting predominantly of SCC samples. The specific gene expression pattern observed correlated with enhanced activation of the Wnt Planar Cell Polarity (PCP) pathway and inhibition of the canonical Wnt signaling branch. Further real time RT-PCR follow-up with additional primary tumor samples and lung cancer cell lines confirmed enrichment of Wnt/PCP pathway associated genes in the SCC subtype. Dysregulation of the canonical Wnt pathway, characterized by increased levels of β-catenin and epigenetic silencing of negative regulators, has been reported in adenocarcinoma of the lung. Our results suggest that SCC and AC utilize different branches of the Wnt pathway during oncogenesis.     

Context-specific infinite mixtures for clustering gene expression profiles across diverse microarray dataset

MOTIVATION: Identifying groups of co-regulated genes by monitoring their expression over various experimental conditions is complicated by the fact that such co-regulation is condition-specific. Ignoring the context-specific nature of co-regulation significantly reduces the ability of clustering procedures to detect co-expressed genes due to additional 'noise' introduced by non-informative measurements. RESULTS: We have developed a novel Bayesian hierarchical model and corresponding computational algorithms for clustering gene expression profiles across diverse experimental conditions and studies that accounts for context-specificity of gene expression patterns. The model is based on the Bayesian infinite mixtures framework and does not require a priori specification of the number of clusters. We demonstrate that explicit modeling of context-specificity results in increased accuracy of the cluster analysis by examining the specificity and sensitivity of clusters in microarray data. We also demonstrate that probabilities of co-expression derived from the posterior distribution of clusterings are valid estimates of statistical significance of created clusters. AVAILABILITY: The open-source package gimm is available at http://eh3.uc.edu/gimm.     

GroupRank: Rank Candidate Genes in PPI Network by Differentially Expressed Gene Groups

Many cell activities are organized as a network, and genes are clustered into co-expressed groups if they have the same or closely related biological function or they are co-regulated. In this study, based on an assumption that a strong candidate disease gene is more likely close to gene groups in which all members coordinately differentially express than individual genes with differential expression, we developed a novel disease gene prioritization method GroupRank by integrating gene co-expression and differential expression information generated from microarray data as well as PPI network. A candidate gene is ranked high using GroupRank if it is differentially expressed in disease and control or is close to differentially co-expressed groups in PPI network. We tested our method on data sets of lung, kidney, leukemia and breast cancer. The results revealed GroupRank could efficiently prioritize disease genes with significantly improved AUC value in comparison to the previous method with no consideration of co-exprssed gene groups in PPI network. Moreover, the functional analyses of the major contributing gene group in gene prioritization of kidney cancer verified that our algorithm GroupRank not only ranks disease genes efficiently but also could help us identify and understand possible mechanisms in important physiological and pathological processes of disease.     

CoGA: An R Package to Identify Differentially Co-Expressed Gene Sets by Analyzing the Graph Spectra

Gene set analysis aims to identify predefined sets of functionally related genes that are differentially expressed between two conditions. Although gene set analysis has been very successful, by incorporating biological knowledge about the gene sets and enhancing statistical power over gene-by-gene analyses, it does not take into account the correlation (association) structure among the genes. In this work, we present CoGA (Co-expression Graph Analyzer), an R package for the identification of groups of differentially associated genes between two phenotypes. The analysis is based on concepts of Information Theory applied to the spectral distributions of the gene co-expression graphs, such as the spectral entropy to measure the randomness of a graph structure and the Jensen-Shannon divergence to discriminate classes of graphs. The package also includes common measures to compare gene co-expression networks in terms of their structural properties, such as centrality, degree distribution, shortest path length, and clustering coefficient. Besides the structural analyses, CoGA also includes graphical interfaces for visual inspection of the networks, ranking of genes according to their “importance” in the network, and the standard differential expression analysis. We show by both simulation experiments and analyses of real data that the statistical tests performed by CoGA indeed control the rate of false positives and is able to identify differentially co-expressed genes that other methods failed.     

Is optimal gene order impossible?

Recent evidence suggests that yeast genes encoding proteins that are present in the same protein complex tend to be linked and to be co-expressed. More generally, we found that genes that are close to each other in the protein interaction network tend to be linked more often than expected and are often co-expressed. Unexpectedly, we found that linked genes in network proximity have unusually high recombination rates. Because high recombination rates are associated with high rates of genome re-organization, our findings might explain why the clustering of genes in proximity in the network is such a weak effect: there could be a co-evolutionary cycle of physical linkage for co-expression, upwards modification of the recombination rate and concomitant break-up of a cluster. Under such a model an "optimal" gene order is never stable.     

Identifying candidate genes for wood formation in poplar based on microarray network analysis and graph theory

In poplar, genetic research on wood properties is very important for the improvement of wood quality. Studies of wood formation genes at each developmental stage using modern biotechnology have often been limited to several genes or gene families. Because of the complex regulatory network involved in the co-expression and interactions of thousands of genes, however, the genetic mechanisms of wood formation must be surveyed on a genome-wide scale. In this study, we identified wood formation-related genes using a differentially co-expressed (DCE) gene subset approach based on biological networks inferred from microarray data. Gene co-expression networks in leaf, root, and wood tissues were first constructed and topologically analyzed using microarray data collected from the Gene Expression Omnibus. The DCE gene modules in wood-forming tissue were then detected based on graph theory, which was followed by gene ontology (GO) enrichment analysis and GO annotation of probe sets. Finally, 72 probe sets were identified in the largest cohesive subgroup of the DCE gene network in wood tissue, with most of the probe sets associated with wood formation-related biological processes and GO cellular component categories. The approach described in this paper provides an effective strategy to identify wood formation genes in poplar and should contribute to the better understanding of the genetic and molecular mechanisms underlying wood properties in trees.     

Whole miRNome-wide Differential Co-expression of MicroRNAs

Co-regulation of genes has been extensively analyzed, however, rather limited knowledge is available on co-regulations within the miRNome. We investigated differential co-expression of microRNAs (miRNAs) based on miRNome profiles of whole blood from 540 individuals. These include patients suffering from different cancer and non-cancer diseases, and unaffected controls. Using hierarchi-cal clustering, we found 9 significant clusters of co-expressed miRNAs containing 2-36 individual miRNAs. Through analyzing multiple sequencing alignments in the clusters, we found that co-expression of miRNAs is associated with both sequence similarity and genomic co-localization. We calculated correlations for all 371,953 pairs of miRNAs for all 540 individuals and identified 184 pairs of miRNAs with high correlation values. Out of these 184 pairs of miRNAs, 16 pairs (8.7%) were differentially co-expressed in unaffected controls, cancer patients and patients with non-cancer diseases. By computing correlated and anti-correlated miRNA pairs, we constructed a network with 184 putative co-regulations as edges and 100 miRNAs as nodes. Thereby, we detected specific clusters of miRNAs with high and low correlation values. Our approach represents the most comprehensive co-regulation analysis based on whole miRNome-wide expression profiling. Our findings further decrypt the interactions of miRNAs in normal and human pathological processes.