Network-based biomarkers enhance classical approaches to prognostic gene expression signatures

NetRank Recovers Known Cancer Hallmark Genes as Universal Biomarker Signature for Cancer Outcome Prediction

Gene expression can serve as a powerful predictor for disease progression and other phenotypes. Consequently, microarrays, which capture gene expression genome-wide, have been used widely over the past two decades to derive biomarker signatures for tasks such as cancer grading, prognosticating the formation of metastases, survival, and others. Each of these signatures was selected and optimized for a very specific phenotype, tissue type, and experimental set-up. While all of these differences may naturally contribute to very heterogeneous and different biomarker signatures, all cancers share characteristics regardless of particular cell types or tissue as summarized in the hallmarks of cancer. These commonalities could give rise to biomarker signatures, which perform well across different phenotypes, cell and tissue types. Here, we explore this possibility by employing a network-based approach for pan-cancer biomarker discovery. We implement a random surfer model, which integrates interaction, expression, and phenotypic information to rank genes by their suitability for outcome prediction. To evaluate our approach, we assembled 105 high-quality microarray datasets sampled from around 13,000 patients and covering 13 cancer types. We applied our approach (NetRank) to each dataset and aggregated individual signatures into one compact signature of 50 genes. This signature stands out for two reasons. First, in contrast to other signatures of the 105 datasets, it is performant across nearly all cancer types and phenotypes. Second, It is interpretable, as the majority of genes are linked to the hallmarks of cancer in general and proliferation specifically. Many of the identified genes are cancer drivers with a known mutation burden linked to cancer. Overall, our work demonstrates the power of network-based approaches to compose robust, compact, and universal biomarker signatures for cancer outcome prediction.

Identification of subtypes correlated with tumor immunity and immunotherapy in cutaneous melanoma

Because immune checkpoint inhibitors (ICIs) are effective for a subset of melanoma patients, identification of melanoma subtypes responsive to ICIs is crucial. We performed clustering analyses to identify immune subtypes of melanoma based on the enrichment levels of 28 immune cells using transcriptome datasets for six melanoma cohorts, including four cohorts not treated with ICIs and two cohorts treated with ICIs. We identified three immune subtypes (Im-H, Im-M, and Im-L), reproducible in these cohorts. Im-H displayed strong immune signatures, low stemness and proliferation potential, genomic stability, high immunotherapy response rate, and favorable prognosis. Im-L showed weak immune signatures, high stemness and proliferation potential, genomic instability, low immunotherapy response rate, and unfavorable prognosis. The pathways highly enriched in Im-H included immune, MAPK, apoptosis, calcium, VEGF, cell adhesion molecules, focal adhesion, gap junction, and PPAR. The pathways highly enriched in Im-L included Hippo, cell cycle, and ErbB. Copy number alterations correlated inversely with immune signatures in melanoma, while tumor mutation burden showed no significant correlation. The molecular features correlated with favorable immunotherapy response included immune-promoting signatures and pathways of PPAR, MAPK, VEGF, calcium, and glycolysis/gluconeogenesis. Our data recapture the immunological heterogeneity in melanoma and provide clinical implications for the immunotherapy of melanoma.

Clinical and Molecular Correlates of NLRC5 Expression in Patients With Melanoma

NLRC5 is an important regulator in antigen presentation and inflammation, and its dysregulation promotes tumor progression. In melanoma, the impact of NLRC5 expression on molecular phenotype, clinical characteristics, and tumor features is largely unknown. In the present study, public datasets from the Cancer Cell Line Encyclopedia (CCLE), Gene Expression Omnibus (GEO), The Cancer Genome Atlas (TCGA), and cBioPortal were used to address these issues. We identify that NLRC5 is expressed in both immune cells and melanoma cells in melanoma samples and its expression is regulated by SPI1 and DNA methylation. NLRC5 expression is closely associated with Breslow thickness, Clark level, recurrence, pathologic T stage, and ulceration status in melanoma. Truncating/splice mutations rather than missense mutations in NLRC5 could compromise the expression of downstream genes. Low expression of NLRC5 is associated with poor prognosis, low activity of immune-related signatures, low infiltrating level of immune cells, and low cytotoxic score in melanoma. Additionally, NLRC5 expression correlates with immunotherapy efficacy in melanoma. In summary, these findings suggest that NLRC5 acts as a tumor suppressor in melanoma via modulating the tumor immune microenvironment. Targeting the NLRC5 related pathway might improve efficacy of immunotherapy for melanoma patients.

Single-cell melanoma transcriptomes depicting functional versatility and clinical implications of STIM1 in the tumor microenvironment

Rationale: Previous studies have implicated the functions of stromal interaction molecule 1 (STIM1) in immunity and malignancy, however, the specificity and effects of STIM1 expression in malignant and non-malignant cells in the tumor microenvironment are unclear.

Methods: In the current study, we posed two central questions: (1) does STIM1 expression elicit different cellular programs in cell types within the melanoma tumor microenvironment (2) whether the expression of STIM1 and STIM1-coexpressed genes (SCGs) serve as prognostic indicators of patient's outcomes? To answer these questions, we dissected cell-specific STIM1-associated cellular programs in diverse cell types within the melanoma tumor microenvironment by measuring cell-type specificity of STIM1 expression and SCGs.

Results: A distinct set of SCGs was highly affected in malignant melanoma cells, but not in the other cell types, suggesting the existence of malignant-cell-specific cellular programs reflected by STIM1 expression. In contrast to malignant cells, STIM1 expression appeared to trigger universal and non-specific biological functions in non-malignant cell types, as exemplified by the transcriptomes of macrophages and CD4⁺ T regulatory cells. Results from bioinformatic analyses indicated that SCGs in malignant cells may alter cell-cell interactions through cytokine/chemokine signaling and/or orchestrate immune infiltration into the tumor. Moreover, a prognostic association between SCGs in CD4⁺ T regulatory cells and patient's outcomes was identified. However, we didn't find any correlation between SCGs and responsiveness of immunotherapy.

Conclusions: Overall, our results provide an integrated biological framework for understanding the functional and clinical consequences of cell-specific STIM1 expression in melanoma.

Improved cancer biomarkers identification using network-constrained infinite latent feature selection

Identifying biomarkers that are associated with different types of cancer is an important goal in the field of bioinformatics. Different researcher groups have analyzed the expression profiles of many genes and found some certain genetic patterns that can promote the improvement of targeted therapies, but the significance of some genes is still ambiguous. More reliable and effective biomarkers identification methods are then needed to detect candidate cancer-related genes. In this paper, we proposed a novel method that combines the infinite latent feature selection (ILFS) method with the functional interaction (FIs) network to rank the biomarkers. We applied the proposed method to the expression data of five cancer types. The experiments indicated that our network-constrained ILFS (NCILFS) provides an improved prediction of the diagnosis of the samples and locates many more known oncogenes than the original ILFS and some other existing methods. We also performed functional enrichment analysis by inspecting the over-represented gene ontology (GO) biological process (BP) terms and applying the gene set enrichment analysis (GSEA) method on selected biomarkers for each feature selection method. The enrichments analysis reports show that our network-constraint ILFS can produce more biologically significant gene sets than other methods. The results suggest that network-constrained ILFS can identify cancer-related genes with a higher discriminative power and biological significance.

Computational models of melanoma

Genes, proteins, or cells influence each other and consequently create patterns, which can be increasingly better observed by experimental biology and medicine. Thereby, descriptive methods of statistics and bioinformatics sharpen and structure our perception. However, additionally considering the interconnectivity between biological elements promises a deeper and more coherent understanding of melanoma. For instance, integrative network-based tools and well-grounded inductive in silico research reveal disease mechanisms, stratify patients, and support treatment individualization. This review gives an overview of different modeling techniques beyond statistics, shows how different strategies align with the respective medical biology, and identifies possible areas of new computational melanoma research.

Network modeling of patients' biomolecular profiles for clinical phenotype/outcome prediction

Methods for phenotype and outcome prediction are largely based on inductive supervised models that use selected biomarkers to make predictions, without explicitly considering the functional relationships between individuals. We introduce a novel network-based approach named Patient-Net (P-Net) in which biomolecular profiles of patients are modeled in a graph-structured space that represents gene expression relationships between patients. Then a kernel-based semi-supervised transductive algorithm is applied to the graph to explore the overall topology of the graph and to predict the phenotype/clinical outcome of patients. Experimental tests involving several publicly available datasets of patients afflicted with pancreatic, breast, colon and colorectal cancer show that our proposed method is competitive with state-of-the-art supervised and semi-supervised predictive systems. Importantly, P-Net also provides interpretable models that can be easily visualized to gain clues about the relationships between patients, and to formulate hypotheses about their stratification.

Network Properties of Cancer Prognostic Gene Signatures in the Human Protein Interactome

Prognostic gene signatures are critical in cancer prognosis assessments and their pinpoint treatments. However, their network properties remain unclear. Here, we obtained nine prognostic gene sets including 1439 prognostic genes of different cancers from related publications. Four network centralities were used to examine the network properties of prognostic genes (PG) compared with other gene sets based on the Human Protein Reference Database (HPRD) and String networks. We also proposed three novel network measures for further investigating the network properties of prognostic gene sets (PGS) besides clustering coefficient. The results showed that PG did not occupy key positions in the human protein interaction network and were more similar to essential genes rather than cancer genes. However, PGS had significantly smaller intra-set distance (IAD) and inter-set distance (IED) in comparison with random sets (p-value < 0.001). Moreover, we also found that PGS tended to be distributed within network modules rather than between modules (p-value < 0.01), and the functional intersection of the modules enriched with PGS was closely related to cancer development and progression. Our research reveals the common network properties of cancer prognostic gene signatures in the human protein interactome. We argue that these are biologically meaningful and useful for understanding their molecular mechanism.

DCARS: differential correlation across ranked samples

MOTIVATION

Genes act as a system and not in isolation. Thus, it is important to consider coordinated changes of gene expression rather than single genes when investigating biological phenomena such as the aetiology of cancer. We have developed an approach for quantifying how changes in the association between pairs of genes may inform the outcome of interest called Differential Correlation across Ranked Samples (DCARS). Modelling gene correlation across a continuous sample ranking does not require the dichotomisation of samples into two distinct classes and can identify differences in gene correlation across early, mid or late stages of the outcome of interest.

RESULTS

When we evaluated DCARS against the typical Fisher Z-transformation test for differential correlation, as well as a typical approach testing for interaction within a linear model, on real TCGA data, DCARS significantly ranked gene pairs containing known cancer genes more highly across several cancers. Similar results are found with our simulation study. DCARS was applied to 13 cancers datasets in TCGA, revealing several distinct relationships for which survival ranking was found to be associated with a change in correlation between genes. Furthermore, we demonstrated that DCARS can be used in conjunction with network analysis techniques to extract biological meaning from multi-layered and complex data.

AVAILABILITY AND IMPLEMENTATION

DCARS R package and sample data are available at https://github.com/shazanfar/DCARS. Publicly available data from The Cancer Genome Atlas (TCGA) was used using the TCGABiolinks R package. Supplementary Files and DCARS R package is available at https://github.com/shazanfar/DCARS.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

A data-driven, knowledge-based approach to biomarker discovery: application to circulating microRNA markers of colorectal cancer prognosis

Recent advances in high-throughput technologies have provided an unprecedented opportunity to identify molecular markers of disease processes. This plethora of complex-omics data has simultaneously complicated the problem of extracting meaningful molecular signatures and opened up new opportunities for more sophisticated integrative and holistic approaches. In this era, effective integration of data-driven and knowledge-based approaches for biomarker identification has been recognised as key to improving the identification of high-performance biomarkers, and necessary for translational applications. Here, we have evaluated the role of circulating microRNA as a means of predicting the prognosis of patients with colorectal cancer, which is the second leading cause of cancer-related death worldwide. We have developed a multi-objective optimisation method that effectively integrates a data-driven approach with the knowledge obtained from the microRNA-mediated regulatory network to identify robust plasma microRNA signatures which are reliable in terms of predictive power as well as functional relevance. The proposed multi-objective framework has the capacity to adjust for conflicting biomarker objectives and to incorporate heterogeneous information facilitating systems approaches to biomarker discovery. We have found a prognostic signature of colorectal cancer comprising 11 circulating microRNAs. The identified signature predicts the patients' survival outcome and targets pathways underlying colorectal cancer progression. The altered expression of the identified microRNAs was confirmed in an independent public data set of plasma samples of patients in early stage vs advanced colorectal cancer. Furthermore, the generality of the proposed method was demonstrated across three publicly available miRNA data sets associated with biomarker studies in other diseases.

Sparse Pathway-Induced Dynamic Network Biomarker Discovery for Early Warning Signal Detection in Complex Diseases

In many complex diseases, the transition process from the healthy stage to the catastrophic stage does not occur gradually. Recent studies indicate that the initiation and progression of such diseases are comprised of three steps including healthy stage, pre-disease stage, and disease stage. It has been demonstrated that a certain set of trajectories can be observed in the genetic signatures at the molecular level, which might be used to detect the pre-disease stage and to take necessary medical interventions. In this paper, we propose two optimization-based algorithms for extracting the dynamic network biomarkers responsible for catastrophic transition into the disease stage, and to open new horizons to reverse the disease progression at an early stage through pinpointing molecular signatures provided by high-throughput microarray data. The first algorithm relies on meta-heuristic intelligent search to characterize dynamic network biomarkers represented as a complete graph. The second algorithm induces sparsity on the adjacency matrix of the genes by taking into account the biological signaling and metabolic pathways, since not all the genes in the ineractome are biologically linked. Comprehensive numerical and meta-analytical experiments verify the effectiveness of the results of the proposed approaches in terms of network size, biological meaningfulness, and verifiability.

A novel method to identify differential pathways in uterine leiomyomata based on network strategy

The aim of the present study was to identify differential pathways in uterine leiomyomata (UL) using a novel method based on protein-protein interaction networks and pathway analysis. The pathway networks were constructed by examining the intersections of the Reactome database and the Search Tool for the Retrieval of Interacting Genes/proteins (STRING) protein-protein interaction (PPI) networks. The Objective network was defined as the differential expressed genes (DEGs) associated with the interactions identified by STRING. Topological centrality (degree) analysis was performed for the Objective network to explore the hub genes and hub networks. Subsequent to isolating the intersections between the Pathway and Objective networks, randomization tests were conducted to identify the differential pathways. There were 559,598 interactions in the Pathway networks. A total of 657 genes with 3,835 interactions were mapped in the Objective network, which included 20 hub genes. It was identified that 358 pathways demonstrated interaction with the Objective network, such as Signal Transduction, Immune System and Signaling by G-protein-coupled receptor (GPCR). By accessing the randomization tests, P-values of these pathways were close to 0, which indicated that they were significantly different. The present study successfully identified differential pathways (such as signal transduction, immune system and signaling by GPCR) in UL, which may be potential biomarkers in the detection and treatment of UL.

Construction of protein interaction network involved in lung adenocarcinomas using a novel algorithm

Studies that only assess differentially-expressed (DE) genes do not contain the information required to investigate the mechanisms of diseases. A complete knowledge of all the direct and indirect interactions between proteins may act as a significant benchmark in the process of forming a comprehensive description of cellular mechanisms and functions. The results of protein interaction network studies are often inconsistent and are based on various methods. In the present study, a combined network was constructed using selected gene pairs, following the conversion and combination of the scores of gene pairs that were obtained across multiple approaches by a novel algorithm. Samples from patients with and without lung adenocarcinoma were compared, and the RankProd package was used to identify DE genes. The empirical Bayesian (EB) meta-analysis approach, the search tool for the retrieval of interacting genes/proteins database (STRING), the weighted gene coexpression network analysis (WGCNA) package and the differentially-coexpressed genes and links package (DCGL) were used for network construction. A combined network was also constructed with a novel rank-based algorithm using a combined score. The topological features of the 5 networks were analyzed and compared. A total of 941 DE genes were screened. The topological analysis indicated that the gene interaction network constructed using the WGCNA method was more likely to produce a small-world property, which has a small average shortest path length and a large clustering coefficient, whereas the combined network was confirmed to be a scale-free network. Gene pairs that were identified using the novel combined method were mostly enriched in the cell cycle and p53 signaling pathway. The present study provided a novel perspective to the network-based analysis. Each method has advantages and disadvantages. Compared with single methods, the combined algorithm used in the present study may provide a novel method to analyze gene interactions, with increased credibility.

Using Multi-objective Optimization to Identify Dynamical Network Biomarkers as Early-warning Signals of Complex Diseases

Biomarkers have gained immense scientific interest and clinical value in the practice of medicine. With unprecedented advances in high-throughput technologies, research interest in identifying novel and customized disease biomarkers for early detection, diagnosis, or drug responses is rapidly growing. Biomarkers can be identified in different levels of molecular biomarkers, networks biomarkers and dynamical network biomarkers (DNBs). The latter is a recently developed concept which relies on the idea that a cell is a complex system whose behavior is emerged from interplay of various molecules, and this network of molecules dynamically changes over time. A DNB can serve as an early-warning signal of disease progression, or as a leading network that drives the system into the disease state, and thus unravels mechanisms of disease initiation and progression. It is therefore of great importance to identify DNBs efficiently and reliably. In this work, the problem of DNB identification is defined as a multi-objective optimization problem, and a framework to identify DNBs out of time-course high-throughput data is proposed. Temporal gene expression data of a lung injury with carbonyl chloride inhalation exposure has been used as a case study, and the functional role of the discovered biomarker in the pathogenesis of lung injury has been thoroughly analyzed.

InCoB2014: Systems Biology update from the Asia-Pacific. Introduction

Selected papers from the 13th International Conference on Bioinformatics (InCoB2014), July 31-2 August, 2014 in Sydney, Australia have been compiled in this supplement. These range from network analysis and gene regulatory networks to systems level biological analysis, providing the 2014 update to InCoB's computational systems biology research.