Co-differential Gene Selection and Clustering Based on Graph Regularized Multi-View NMF in Cancer Genomic Data

Application of non-negative matrix factorization in oncology: one approach for establishing precision medicine

The increase in the expectations of artificial intelligence (AI) technology has led to machine learning technology being actively used in the medical field. Non-negative matrix factorization (NMF) is a machine learning technique used for image analysis, speech recognition, and language processing; recently, it is being applied to medical research. Precision medicine, wherein important information is extracted from large-scale medical data to provide optimal medical care for every individual, is considered important in medical policies globally, and the application of machine learning techniques to this end is being handled in several ways. NMF is also introduced differently because of the characteristics of its algorithms. In this review, the importance of NMF in the field of medicine, with a focus on the field of oncology, is described by explaining the mathematical science of NMF and the characteristics of the algorithm, providing examples of how NMF can be used to establish precision medicine, and presenting the challenges of NMF. Finally, the direction regarding the effective use of NMF in the field of oncology is also discussed.

Multi-view Random-walk Graph regularization Low-Rank Representation for cancer clustering and Differentially Expressed Gene Selection

Cancer genome data generally consists of multiple views from different sources. These views provide different levels of information about gene activity, as well as more comprehensive cancer information. The low-rank representation (LRR) method, as a powerful subspace clustering method, has been extended and applied in cancer data research. However, most methods based on low-rank representation only study single-view data in cancer genome data, such as gene expression data. The methods based on single-view genome data usually ignore the complementary relationship between the views, which is not conducive to further study of cancer. Therefore, this paper proposes a new method named Multi-view Random-walk Graph regularization Low-Rank Representation (MRGLRR) to comprehensively analyze multi-view genomics data. This method uses multi-view model to find the common centroid of view. By constructing a joint affinity matrix to learn the low-rank subspace representation of multiple sets of data, the hidden information of each view is fully obtained. In addition, this method introduces random walk graph regularization constraint to obtain more accurate similarity between samples. Different from the traditional graph regularization constraint, after constructing the KNN graph, we use the random walk algorithm to obtain the weight matrix. The random walk algorithm can retain more local geometric information and better learn the topological structure of the data. What's more, a feature gene selection strategy suitable for multi-view model is proposed to find more differentially expressed genes with research value. Experimental results show that our method is better than other representative methods in terms of clustering and feature gene selection for cancer multi-omics data.

Prognosis of Tumor Microenvironment in Luminal B-Type Breast Cancer

Objective

Tumor microenvironment as an important element of malignancy could help predict cancer prognosis and therapeutic response; thus, a prognostic landscape map of the tumor microenvironment in luminal B breast cancers should be developed.

Methods

The GEO and TCGA databases were employed to retrieve clinical follow-up data and expression profiles of luminal B breast cancer. CIBERSORT was applied to assess the infiltration of the tumor microenvironment of 209 patients and to construct tumor microenvironment-based subtypes of luminal B breast cancer. We also conducted Cox multivariate regression analysis to select features that could be used to develop a microenvironment signature for cancer. Samples were categorized as having low and high TME scores according to the median TME score. The correlations of prognosis and TME score, expression levels of immune factors and genomic variation, and clinical features were further investigated.

Results

We found that high TME scores were correlated with poor prognosis. The current findings showed that the expressions of multiple immune-related genes, including CXCL9, CXCL10, GZMB, and PDCD1LG2, were upregulated in cancer with high TME scores. The high-risk group showed lower TP53 gene mutation frequency as opposed to that of the low-risk group. For the purpose of developing a TME scoring system, the TME infiltration levels of 209 patients with luminal B breast cancer from TCGA were comprehensively analyzed.

Conclusions

Our analysis revealed that the TME score was an indicator of patients' response to immune checkpoint modulators and an effective prognostic biomarker. TME scoring improves current immunotherapy on luminal B breast cancer.

SMRT: Randomized Data Transformation for Cancer Subtyping and Big Data Analysis

Cancer is an umbrella term that includes a range of disorders, from those that are fast-growing and lethal to indolent lesions with low or delayed potential for progression to death. The treatment options, as well as treatment success, are highly dependent on the correct subtyping of individual patients. With the advancement of high-throughput platforms, we have the opportunity to differentiate among cancer subtypes from a holistic perspective that takes into consideration phenomena at different molecular levels (mRNA, methylation, etc.). This demands powerful integrative methods to leverage large multi-omics datasets for a better subtyping. Here we introduce Subtyping Multi-omics using a Randomized Transformation (SMRT), a new method for multi-omics integration and cancer subtyping. SMRT offers the following advantages over existing approaches: (i) the scalable analysis pipeline allows researchers to integrate multi-omics data and analyze hundreds of thousands of samples in minutes, (ii) the ability to integrate data types with different numbers of patients, (iii) the ability to analyze un-matched data of different types, and (iv) the ability to offer users a convenient data analysis pipeline through a web application. We also improve the efficiency of our ensemble-based, perturbation clustering to support analysis on machines with memory constraints. In an extensive analysis, we compare SMRT with eight state-of-the-art subtyping methods using 37 TCGA and two METABRIC datasets comprising a total of almost 12,000 patient samples from 28 different types of cancer. We also performed a number of simulation studies. We demonstrate that SMRT outperforms other methods in identifying subtypes with significantly different survival profiles. In addition, SMRT is extremely fast, being able to analyze hundreds of thousands of samples in minutes. The web application is available at http://SMRT.tinnguyen-lab.com. The R package will be deposited to CRAN as part of our PINSPlus software suite.

Dual Hyper-Graph Regularized Supervised NMF for Selecting Differentially Expressed Genes and Tumor Classification

Non-negative matrix factorization (NMF) is a dimensionality reduction technique based on high-dimensional mapping. It can learn part-based representations effectively. In this paper, we propose a method called Dual Hyper-graph Regularized Supervised Non-negative Matrix Factorization (HSNMF). To encode the geometric information of the data, the hyper-graph is introduced into the model as a regularization term. The advantage of hyper-graph learning is to find higher order data relationship to enhance data relevance. This method constructs the data hyper-graph and the feature hyper-graph to find the data manifold and the feature manifold simultaneously. The application of hyper-graph theory in cancer datasets can effectively find pathogenic genes. The discrimination information is further introduced into the objective function to obtain more information about the data. Supervised learning with label information greatly improves the classification effect. Furthermore, the real datasets of cancer usually contain sparse noise, so the L_2,1-norm is applied to enhance the robustness of HSNMF algorithm. Experiments under The Cancer Genome Atlas (TCGA) datasets verify the feasibility of the HSNMF method.

Clonal Architectures Predict Clinical Outcome in Gastric Adenocarcinoma Based on Genomic Variation, Tumor Evolution, and Heterogeneity

Stomach adenocarcinoma (STAD) is a highly heterogeneous disease. Due to the lack of effective molecular markers and personalized treatment, the prognosis of gastric cancer patients is still very poor. The ABSOLUTE algorithm and cancer cell fraction were used to evaluate the clonal and subclonal status of 349 TCGA (The Cancer Genome Cancer Atlas)-STAD patients. Non-negative matrix factorization was used to identify the mutation characteristics of the samples. Univariate Cox regression analysis was used to determine the relationship between clonal/subclonal events and prognosis, and the Spearman correlation was used to evaluate the relationship of clonal/subclonal events to tumor mutation burden (TMB) and neoantigens. The evolution pattern of STAD demonstrated great tumor heterogeneity. TP53, USH2A, and GLI3 appeared earliest in STAD and may drive STAD. CTNNB1, LRP1B, and ERBB4 appeared the latest in STAD, and may be related to STAD’s progress. Univariate Cox regression analysis identified four early genes, eight intermediate genes, and seven late genes significantly associated with overall survival. The number of subclonal events in the T stage was significantly different. The N stage, gender, and histological type were significantly different for clonal events, and there was a significant correlation between clonal/subclonal events and TMB/neoantigens. Our results highlight the importance of systematic evaluation of evolutionary models in the clinical management of STAD and personalized gastric cancer treatment.

Identification of an Immune Gene Signature Based on Tumor Microenvironment Characteristics in Colon Adenocarcinoma

Tumor microenvironment (TME) changes are related to the occurrence and development of colon adenocarcinoma (COAD). This study aimed to analyze the characteristics of the immune microenvironment in CC, as well as the microenvironment's relationship with the clinical features of CC. Based on The Cancer Genome Atlas (TCGA) and GSE39582 cohorts, the scores of 22 tumor infiltrating lymphocytes (TILs) were calculated using CIBERSORT. ConsensusClusterPlus was used for unsupervised clustering. Three TME subtypes (TMEC1, TMEC2, and TME3) were identified based on TIL scores. TMEC2 was associated with the worst prognosis. Random forest, k-means clustering, and principal component analysis were used to construct the TME score risk signature. The median TME score was used to divide the samples into high- and low-risk groups. The prognoses of the patients with high TME scores were worse than those of the patients with low TME scores. A high TME score was an independent prognostic risk factor for patients with colon cancer. The Gene Set Enrichment Analysis (GSEA) results showed that those with high TME scores were enriched in FOCAL_ADHESION, ECM_RECEPTOR_INTERACTION, and PATHWAYS_IN_CANCER. Our findings will provide a new strategy for immunotherapy in patients with CC.

Integrative analysis of genomic, epigenomic and transcriptomic data identified molecular subtypes of esophageal carcinoma

Esophageal cancer (EC) involves many genomic, epigenetic and transcriptomic disorders, which play key roles in the heterogeneous progression of cancer. However, the study of EC with multi-omics has not been conducted. This study identified a high consistency between DNA copy number variations and abnormal methylations in EC by analyzing genomics, epigenetics and transcriptomics data and investigating mutual correlations of DNA copy number variation, methylation and gene expressions, and stratified copy number variation genes (CNV-Gs) and methylation genes (MET-Gs). The methylation, CNVs and expression profiles of CNV-Gs and MET-Gs were analyzed by consistent clustering using iCluster integration, here, we determined three subtypes (iC1, iC2, iC3) with different molecular traits, prognostic characteristics and tumor immune microenvironment features. We also identified 4 prognostic genes (CLDN3, FAM221A, GDF15 and YBX2) differentially expressed in the three subtypes, and could therefore be used as representative biomarkers for the three subtypes of EC. In conclusion, by performing comprehensive analysis on genomic, epigenetic and transcriptomic regulations, the current study provided new insights into the multilayer molecular and pathological traits of EC, and contributed to the precision medication for EC patients.

Development and validation of prognostic markers in sarcomas base on a multi-omics analysis

BACKGROUND

In sarcomas, the DNA copy number and DNA methylation exhibit genomic aberrations. Transcriptome imbalances play a driving role in the heterogeneous progression of sarcomas. However, it is still unclear whether abnormalities of DNA copy numbers are systematically related to epigenetic DNA methylation, thus, a comprehensive analysis of sarcoma occurrence and development from the perspective of epigenetic and genomics is required.

METHODS

RNASeq, copy number variation (CNV), methylation data, clinical follow-up information were obtained from The Cancer Genome Atlas (TCGA) and GEO database. The association between methylation and CNV was analyzed to further identify methylation-related genes (MET-Gs) and CNV abnormality-related genes (CNV-Gs). Subsequently DNA copy number, methylation, and gene expression data associated with the MET-Gs and CNV-Gs were integrated to determine molecular subtypes and clinical and molecular characteristics of molecular subtypes. Finally, key biomarkers were determined and validated in independent validation sets.

RESULTS

A total of 5354 CNV-Gs and 4042 MET-Gs were screened and showed a high degree of consistency. Four molecular subtypes (iC1, iC2, iC3, and iC4) with different prognostic significances were identified by multiomics cluster analysis, specifically, iC2 had the worst prognosis and iC4 indicated an immune-enhancing state. Three potential prognostic markers (ENO1, ACVRL1 and APBB1IP) were determined after comparing the molecular characteristics of the four molecular subtypes. The expression of ENO1 gene was significantly correlated with CNV, and was noticeably higher in iC2 subtype with the worst prognosis than any other subtypes. The expressions of ACVRL1 and APBB1IP were negatively correlated with methylation, and were high-expressed in the iC4 subtype with the most favorable prognosis. In addition, the number of silent/nonsilent mutations and neoantigens in iC2 subtype were significantly more than those in iC1/iC3/iC4 subtype, and the same trend was also observed in CNV Gain/Loss.

CONCLUSION

The current comprehensive analysis of genomic and epigenomic regulation provides new insights into multilayered pathobiology of sarcomas. Four molecular subtypes and three prognostic markers developed in this study improve the current understanding of the molecular mechanisms underlying sarcoma.

Correntropy induced loss based sparse robust graph regularized extreme learning machine for cancer classification

Background

As a machine learning method with high performance and excellent generalization ability, extreme learning machine (ELM) is gaining popularity in various studies. Various ELM-based methods for different fields have been proposed. However, the robustness to noise and outliers is always the main problem affecting the performance of ELM.

Results

In this paper, an integrated method named correntropy induced loss based sparse robust graph regularized extreme learning machine (CSRGELM) is proposed. The introduction of correntropy induced loss improves the robustness of ELM and weakens the negative effects of noise and outliers. By using the L_2,1-norm to constrain the output weight matrix, we tend to obtain a sparse output weight matrix to construct a simpler single hidden layer feedforward neural network model. By introducing the graph regularization to preserve the local structural information of the data, the classification performance of the new method is further improved. Besides, we design an iterative optimization method based on the idea of half quadratic optimization to solve the non-convex problem of CSRGELM.

Conclusions

The classification results on the benchmark dataset show that CSRGELM can obtain better classification results compared with other methods. More importantly, we also apply the new method to the classification problems of cancer samples and get a good classification effect.

Identification of prognostic and immune-related gene signatures in the tumor microenvironment of endometrial cancer

Uterine corpus endometrial cancer (UCEC) is one of the most prevalent female malignancies in clinical practice. Due to the lack of effective biomarkers and personalized treatments, the prognosis of advanced-stage EC remains unfavorable. Modulation of the immune microenvironment is closely related to the onset and development of endometrial cancer. In the present study, we attempt to systematically analyze the characteristics of the immune microenvironment of endometrial cancer and investigate its association with clinical features by applying bioinformatics. RNA-Seq in TCGA (The Cancer Genome Atlas) and clinical follow-up information of patents were used for analysis. The Tumor Microenvironment (TME) score infiltration patterns of 523 endometrial cancer patients were evaluated using CIBERSORT. Random forest, multivariable cox analysis were used to build the TME score. Fisher's exact test was used to compare the genes that show significant differences in the frequency of mutations between groups. Two TME phenotypes were defined. There is a significant relationship between the TME score and grade. High TME score samples are highly expressed in immune activation, TGF pathway activation and immune checkpoint genes, and low TME score samples have high frequency mutations of PTEN, CSE1L and ITGB3. Therefore, describing the comprehensive landscape of UCEC's TME characteristics may help explain patients' response to immunotherapy and provide new strategies for cancer treatment.

Identification of an energy metabolism‑related gene signature in ovarian cancer prognosis

Changes in energy metabolism may be potential biomarkers and therapeutic targets for cancer as they frequently occur within cancer cells. However, basic cancer research has failed to reach a consistent conclusion on the function(s) of mitochondria in energy metabolism. The significance of energy metabolism in the prognosis of ovarian cancer remains unclear; thus, there remains an urgent need to systematically analyze the characteristics and clinical value of energy metabolism in ovarian cancer. Based on gene expression patterns, the present study aimed to analyze energy metabolism‑associated characteristics to evaluate the prognosis of patients with ovarian cancer. A total of 39 energy metabolism‑related genes significantly associated with prognosis were obtained, and three molecular subtypes were identified by nonnegative matrix factorization clustering, among which the C1 subtype was associated with poor clinical outcomes of ovarian cancer. The immune response was enhanced in the tumor microenvironment. A total of 888 differentially expressed genes were identified in C1 compared with the other subtypes, and the results of the pathway enrichment analysis demonstrated that they were enriched in the 'PI3K‑Akt signaling pathway', 'cAMP signaling pathway', 'ECM‑receptor interaction' and other pathways associated with the development and progression of tumors. Finally, eight characteristic genes (tolloid‑like 1 gene, type XVI collagen, prostaglandin F2α, cartilage intermediate layer protein 2, kinesin family member 26b, interferon inducible protein 27, growth arrest‑specific gene 1 and chemokine receptor 7) were obtained through LASSO feature selection; and a number of them have been demonstrated to be associated with ovarian cancer progression. In addition, Cox regression analysis was performed to establish an 8‑gene signature, which was determined to be an independent prognostic factor for patients with ovarian cancer and could stratify sample risk in the training, test and external validation datasets (P<0.01; AUC >0.8). Gene Set Enrichment Analysis results revealed that the 8‑gene signature was involved in important biological processes and pathways of ovarian cancer. In conclusion, the present study established an 8‑gene signature associated with metabolic genes, which may provide new insights into the effects of energy metabolism on ovarian cancer. The 8‑gene signature may serve as an independent prognostic factor for ovarian cancer patients.

Integrative analysis of genomic and epigenetic regulation of endometrial cancer

Endometrial carcinomas (EC) are characterized by high DNA copy numbers and DNA methylation aberrations. In this study, we sought to comprehensively explore the effect of these two factors on development and progression of EC by analyzing integrated genomic and epigenetic analysis to. We found high DNA copy number and DNA methylation abnormalities in EC, with 6308 copy-number variation genes (CNV-G) and 4376 methylation genes (MET-G). We used these CNV-G and MET-G to subcategorize the samples for prognostic analysis, and identified three molecular subtypes (iC1, iC2, iC3). Moreover, the subtypes exhibited different tumor immune microenvironment characteristics. A further analysis of their molecular characteristics revealed three potential prognostic markers (KIAA1324, nonexpresser of pathogenesis-related genes1 (NPR1) and idiopathic hypogonadotropic hypogonadism (IHH)). Notably, all three markers showed distinct CNV, DNA methylation, and gene expression profiles. Analysis of mutations among the three subtypes revealed that iC2 had fewer mutations than the other subtypes. Conversely, iC2 showed significantly higher CNV levels than other subtypes. This comprehensive analysis of genomic and epigenetic profiles identified three prognostic markers, therefore, provides new insights into the multi-layered pathology of EC. These can be utilized for accurate treatment of EC patients.

Immune gene signature delineates a subclass of thyroid cancer with unfavorable clinical outcomes

Thyroid cancer (THCA) is a heterogeneous disease with multiple clinical outcomes Immune cells regulate its progression. Three immunomolecular subtypes (C1, C2, C3) were identified in gene expression data sets from TCGA and GEO databases. Among them, subtype C3 had highest frequency of BRAF mutations, lowest frequency of RAS mutations, highest mutation load and shorter progression-free survival. Functional enrichment analysis for the genes revealed that C1 was up-regulated in the ERK cascade pathway, C2 was up-regulated in cell migration and proliferation pathways, while C3 was enriched in body fluid, protein regulation and response to steroid hormones functions. Notably, the three molecular subtypes exhibit differences in immune microenvironments as shown by timer database and analysis of immune expression signatures. The abundance of B_cell, CD4_Tcell, Neutrophil, Macrophage and Dendritic cells in C2 subtype were lower than in C1 and C3 subtypes Leukocyte fraction, proliferation macrophage regulation, lymphocyte infiltration, IFN gamma response and TGF beta response scores were significantly higher in C3 compared with C1 and C2 subtypes. Unlike C3 subtype, it was observed that C1 and C2 subtypes were significantly negatively correlated with most immune checkpoint genes in two different cohorts. The characteristic genes were differentially expressed between cancer cells, adjacent tissues, and metastatic tissues in different cohorts. In summary, THCA can be subclassified into three molecular subtypes with distinct histological types, genetic and transcriptional phenotypes, all of which have potential clinical implications.

Multiview learning for understanding functional multiomics

The molecular mechanisms and functions in complex biological systems currently remain elusive. Recent high-throughput techniques, such as next-generation sequencing, have generated a wide variety of multiomics datasets that enable the identification of biological functions and mechanisms via multiple facets. However, integrating these large-scale multiomics data and discovering functional insights are, nevertheless, challenging tasks. To address these challenges, machine learning has been broadly applied to analyze multiomics. This review introduces multiview learning-an emerging machine learning field-and envisions its potentially powerful applications to multiomics. In particular, multiview learning is more effective than previous integrative methods for learning data's heterogeneity and revealing cross-talk patterns. Although it has been applied to various contexts, such as computer vision and speech recognition, multiview learning has not yet been widely applied to biological data-specifically, multiomics data. Therefore, this paper firstly reviews recent multiview learning methods and unifies them in a framework called multiview empirical risk minimization (MV-ERM). We further discuss the potential applications of each method to multiomics, including genomics, transcriptomics, and epigenomics, in an aim to discover the functional and mechanistic interpretations across omics. Secondly, we explore possible applications to different biological systems, including human diseases (e.g., brain disorders and cancers), plants, and single-cell analysis, and discuss both the benefits and caveats of using multiview learning to discover the molecular mechanisms and functions of these systems.

Hyper-Graph Regularized Constrained NMF for Selecting Differentially Expressed Genes and Tumor Classification

Non-negative Matrix Factorization (NMF) is a dimensionality reduction approach for learning a parts-based and linear representation of non-negative data. It has attracted more attention because of that. In practice, NMF not only neglects the manifold structure of data samples, but also overlooks the priori label information of different classes. In this paper, a novel matrix decomposition method called Hyper-graph regularized Constrained Non-negative Matrix Factorization (HCNMF) is proposed for selecting differentially expressed genes and tumor sample classification. The advantage of hyper-graph learning is to capture local spatial information in high dimensional data. This method incorporates a hyper-graph regularization constraint to consider the higher order data sample relationships. The application of hyper-graph theory can effectively find pathogenic genes in cancer datasets. Besides, the label information is further incorporated in the objective function to improve the discriminative ability of the decomposition matrix. Supervised learning with label information greatly improves the classification effect. We also provide the iterative update rules and convergence proofs for the optimization problems of HCNMF. Experiments under The Cancer Genome Atlas (TCGA) datasets confirm the superiority of HCNMF algorithm compared with other representative algorithms through a set of evaluations.

Tumor Microenvironment Characterization in Glioblastoma Identifies Prognostic and Immunotherapeutically Relevant Gene Signatures

Tumor microenvironment (TME) cells are important elements in tumor tissue. There is increasing evidence that they have important clinical pathological significance in predicting tumor clinical outcomes and therapeutic effects. However, no systematic analysis of TME cell interactions in glioblastoma (GBM) has been reported. We systematically analyzed the transcriptional sequencing data of GBM to find an immune gene marker to predict the clinical results of GBM. First, we downloaded the expression profiles and clinical follow-up information of GBM from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO). CIBERSORT was used to evaluate the infiltration mode of TME in 757 patients, systematically correlated TME phenotype with genomic characteristics and clinicopathological characteristics of GBM, defined four TME phenotypes, and TMEScore was constructed using algorithms such as random forest and principal component analysis. There is a significant correlation between TMEScore and age of onset. High TMEScore samples are characterized by immune activation, TGF pathway activation, and high expression of immune checkpoint genes, while low TMEScore samples are characterized by high-frequency IDH1 and MET mutations. Therefore, a comprehensive landscape depicting the TME characteristics of GBM may help explain GBM's response to immunotherapy and provide new strategies for cancer treatment. In this study, TMEScore can be used as a new prognostic marker to predict the survival of GBM patients, and as a potential predictor of immune checkpoint inhibitor response.

RCMF: a robust collaborative matrix factorization method to predict miRNA-disease associations

Background

Predicting miRNA-disease associations (MDAs) is time-consuming and expensive. It is imminent to improve the accuracy of prediction results. So it is crucial to develop a novel computing technology to predict new MDAs. Although some existing methods can effectively predict novel MDAs, there are still some shortcomings. Especially when the disease matrix is processed, its sparsity is an important factor affecting the final results.

Results

A robust collaborative matrix factorization (RCMF) is proposed to predict novel MDAs. The L_2,1-norm are introduced to our method to achieve the highest AUC value than other advanced methods.

Conclusions

5-fold cross validation is used to evaluate our method, and simulation experiments are used to predict novel associations on Gold Standard Dataset. Finally, our prediction accuracy is better than other existing advanced methods. Therefore, our approach is effective and feasible in predicting novel MDAs.