Integration of Multi-Omics Data to Identify Cancer Biomarkers

With the growing use of high-throughput technologies, multi-omics data containing various types of high-dimensional omics data is increasingly being generated to explore the association between the molecular mechanism of the host and diseases. In this study, we present an adaptive sparse multi-block partial least square discriminant analysis (asmbPLS-DA), an extension of our previous work, asmbPLS. This integrative approach identifies the most relevant features across different types of omics data while discriminating multiple disease outcome groups. We used simulation data with various scenarios and a real dataset from the TCGA project to demonstrate that asmbPLS-DA can identify key biomarkers from each type of omics data with better biological relevance than existing competitive methods. Moreover, asmbPLS-DA showed comparable performance in the classification of subjects in terms of disease status or phenotypes using integrated multi-omics molecular profiles, especially when combined with other classification algorithms, such as linear discriminant analysis and random forest. We have made the R package called asmbPLS that implements this method publicly available on GitHub. Overall, asmbPLS-DA achieved competitive performance in terms of feature selection and classification. We believe that asmbPLS-DA can be a valuable tool for multi-omics research.

Cell lines used for pre-clinical testing of oncology compounds are not always chosen based on how well they models patient tumors. Instead they are often chosen based on availability and literature prevalence. The advent of high throughput genomic profiling demonstrates a causative relationship between genomic features and drug response, suggesting that cancer drug discovery could be accelerated by using genomics as a criteria to find ideal cell lines for a given cancer type. The overall oncology clinical trial success rate is dismally low, especially in pancreatic cancer. Pancreatic cancer has a five year survival of 5-6% and is predicted to be the second leading cause of cancer by 2030 with a dearth of promising medicines currently in trials. In order to forecast optimal cell lines for drug testing in pancreatic cancer, we leveraged gene expression, mutations, and copy number variation (CNV) data to compare tumors from The Cancer Genome Atlas (TCGA) to cell lines in Cancer Cell Line Encyclopedia (CCLE). To approximate cell line usage, the number of hits for each cell line in PubMed and Google Scholar were combined. Less than 20% of queried pancreatic cancer cell lines represented more than 88% of the total search hits, demonstrating a robust bias towards certain cell lines. We calculated the CNV correlation between each cell line and each tumor. The cell lines that were popular in literature, such as DAN-G (24% of citations), were often ranked worst by CNV correlation with tumors while some cell lines which were rarely cited such as L33 had among the highest CNV correlation. Next, we filtered mutation data using publicly available mutation scoring algorithms to select the most cancer driving mutations. Hierarchical clustering was applied to the tumor samples and cell lines together based on the presence or absence of the top scoring mutations in order to pinpoint cell lines with mutational spectra similar to tumors. In support of observations made in CNV data, popular cell lines such as DAN-G clustered with other cell lines and L33 clustered predominantly amongst tumor samples, providing further evidence that L33 may be an ideal cell line for modeling pancreatic cancer drug response. In order to leverage all available data types, the selected CNVs and mutations were combined into a pathway level event matrix based on the number of relevant mutations or CNVs within a given pathway and then clustered. Unsurprisingly these results show that most cell lines are much more similar to each other than to tumors. However, a few cell lines (including L33) cluster with tumor samples. Overall our results demonstrate that comprehensively L33 shows the best similarity to pancreatic cancer tumors. We believe that selecting preclinical screening methods that best match relevant tumor biology and genomic drivers could help accelerate the development of new medicines for a variety of cancers. Citation Format: Yoonjeong Cha, Andrew Lysaght, Brian Weiner, Sarah Kolitz, Fadi Towfic, Kevin Fowler, Badri Vardarajan, Maxim Artyomov, Benjamin Zeskind, Rebecca Kusko. Leveraging genomics to optimize models for accelerated pancreatic cancer drug development. [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2015 Nov 5-9; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2015;14(12 Suppl 2):Abstract nr A78.

Breast cancer comprises several molecular subtypes with distinct clinical features and treatment responses, and a substantial portion of each subtype remains incurable. A comprehensive analysis of multi-omics data and clinical profiles is required in order to better understand the biological complexity of this cancer type and to identify new prognostic and therapeutic markers. Thus, there arises a need for useful analytical tools to assist in the investigation and clinical management of the disease. We developed Cancer Target Gene Screening (CTGS), a web application that provides rapid and user-friendly analysis of multi-omics data sets from a large number of primary breast tumors. It allows the investigation of genomic and epigenomic aberrations, evaluation of transcriptomic profiles and performance of survival analyses and of bivariate correlations between layers of omics data. Notably, the genome-wide screening function of CTGS prioritizes candidate genes of clinical and biological significance among genes with copy number alteration, DNA methylation and dysregulated expression by the integrative analysis of different types of omics data in customized subgroups of breast cancer patients. These features may help in the identification of druggable cancer driver genes in a specific subtype or the clinical condition of human breast cancer. CTGS is available at http://ctgs.biohackers.net.

BackgroundGenomic profiling of solid human tumors by projects such as The Cancer Genome Atlas (TCGA) has provided important information regarding the somatic alterations that drive cancer progression and patient survival. Although researchers have successfully leveraged TCGA data to build prognostic models, most efforts have focused on specific cancer types and a targeted set of gene-level predictors. Less is known about the prognostic ability of pathway-level variables in a pan-cancer setting. To address these limitations, we systematically evaluated and compared the prognostic ability of somatic point mutation (SPM) and copy number variation (CNV) data, gene-level and pathway-level models for a diverse set of TCGA cancer types and predictive modeling approaches.ResultsWe evaluated gene-level and pathway-level penalized Cox proportional hazards models using SPM and CNV data for 29 different TCGA cohorts. We measured predictive accuracy as the concordance index for predicting survival outcomes. Our comprehensive analysis suggests that the use of pathway-level predictors did not offer superior predictive power relative to gene-level models for all cancer types but had the advantages of robustness and parsimony. We identified a set of cohorts for which somatic alterations could not predict prognosis, and a unique cohort LGG, for which SPM data was more predictive than CNV data and the predictive accuracy is good for all model types. We found that the pathway-level predictors provide superior interpretative value and that there is often a serious collinearity issue for the gene-level models while pathway-level models avoided this issue.ConclusionOur comprehensive analysis suggests that when using somatic alterations data for cancer prognosis prediction, pathway-level models are more interpretable, stable and parsimonious compared to gene-level models. Pathway-level models also avoid the issue of collinearity, which can be serious for gene-level somatic alterations. The prognostic power of somatic alterations is highly variable across different cancer types and we have identified a set of cohorts for which somatic alterations could not predict prognosis. In general, CNV data predicts prognosis better than SPM data with the exception of the LGG cohort.

Integrating multi-omics data through deep learning for accurate cancer prognosis prediction

G-protein-coupled receptor-associated sorting protein 1 (GPRASP1) plays an important role in tumorigenesis. However, GPRASP1 specific role has not been clarified in head and neck cancer (HNC). HNC RNA sequencing (RNA-seq) datasets, DNA methylation data, somatic mutation data, copy number variation (CNV) data, and corresponding clinicopathologic information were acquired from the Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. A comprehensive evaluation was performed to explore the relationship of GPRASP1 expression with clinicopathologic characteristics, CNV, and DNA methylation. Additionally, we employed HNC tissue microarray (TMA) to further confirm the relation between GPRASP1 expression and clinical features. Then, we systematically associated the GPRASP1 with immunological properties from numerous perspectives, such as immune cell infiltration, immune-related pathways, immune checkpoint inhibitors (ICIs), immunomodulators, immunogenicity, and immunotherapy. Analyzing TCGA, GEO, and TMA datasets, GPRASP1 is significantly down-regulated in HNC compared to normal tissues. The expression of GPRASP1 is significantly negatively correlated with clinical features (perineural invasion, histologic grade, T stage, and TNM stage), and is an independent predictor of favorable prognosis, regardless of other clinicopathological features (HR: 0.42, 95% CI 0.20-0.91, p=0.028). The etiological investigation found that the abnormal expression of GPRASP1 was related to DNA methylation, not CMV. Subsequently, the high expression of GPRASP1 was significantly correlated with immune cell infiltration (CD8+ T cell, tumor infiltrating lymphocyte), immune-related pathways (cytolytic activity, check-point, human leukocyte antigen), ICIs (CTLA4, HAVCR2, LAG3, PDCD1, and TIGIT), immunomodulators (CCR4/5, CXCL9, CXCR3/4/5), and immunogenicity (immune score, neoantigen, tumor mutation burden). Finally, immunophenoscore and tumor immune dysfunction and exclusion analysis demonstrated that GPRASP1 expression levels can accurately predict the immunotherapeutic response. GPRASP1 is a promising candidate biomarker that plays a role in the occurrence, development, and prognosis of HNC. Evaluating GPRASP1 expression will aid in the characterization of tumor microenvironment infiltration and orient more efficient immunotherapy strategies.

In recent biomedical studies, multidimensional profiling, which collects proteomics as well as other types of omics data on the same subjects, is getting increasingly popular. Proteomics, transcriptomics, genomics, epigenomics, and other types of data contain overlapping as well as independent information, which suggests the possibility of integrating multiple types of data to generate more reliable findings/models with better classification/prediction performance. In this chapter, a selective review is conducted on recent data integration techniques for both unsupervised and supervised analysis. The main objective is to provide the "big picture" of data integration that involves proteomics data and discuss the "intuition" beneath the recently developed approaches without invoking too many mathematical details. Potential pitfalls and possible directions for future developments are also discussed.

BackgroundPan-cancer studies have disclosed many commonalities and differences in mutations, copy number variations, and gene expression alterations among cancers. Some of these features are significantly associated with clinical outcomes, and many prognosis-predictive biomarkers or biosignatures have been proposed for specific cancer types. Here, we systematically explored the biological functions and the distribution of survival-related genes (SRGs) across cancers.ResultsWe carried out two different statistical survival models on the mRNA expression profiles in 33 cancer types from TCGA. We identified SRGs in each cancer type based on the Cox proportional hazards model and the log-rank test. We found a large difference in the number of SRGs among different cancer types, and most of the identified SRGs were specific to a particular cancer type. While these SRGs were unique to each cancer type, they were found mostly enriched in cancer hallmark pathways, e.g., cell proliferation, cell differentiation, DNA metabolism, and RNA metabolism. We also analyzed the association between cancer driver genes and SRGs and did not find significant over-representation amongst most cancers.ConclusionsIn summary, our work identified all the SRGs for 33 cancer types from TCGA. In addition, the pan-cancer analysis revealed the similarities and the differences in the biological functions of SRGs across cancers. Given the potential of SRGs in clinical utility, our results can serve as a resource for basic research and biotech applications.

Pancreatic cancer is the third leading cause of cancer-related deaths in the United States. Disease biomarkers quantified from blood-based assays may help reduce mortality by enabling early detection, treatment selection, or response and resistance assessment. PrognomiQ has developed a multi-omics assay and analysis platform that comprehensively profiles blood samples to detect proteins, metabolites, lipids, mRNA, miRNA, cfDNA fragments, and methylation at CpG sites. This platform can provide deep insights into the biology of pancreatic cancer and could enable the development of high sensitivity and specificity tests for the early detection of pancreatic cancer. We conducted a case-control study comprising 196 subjects: 92 with untreated pancreatic cancer and 104 matched controls without pancreatic cancer. For each subject, blood was collected in assay-specific tubes and processed to provide 7 different `omics readouts. cfDNA and mRNA were isolated from samples and assayed following standard NGS protocols. cfDNA fragments were processed to estimate fragment-length disorder and copy-number variation along with CpG site methylation. In addition, targeted and untargeted LCMS were used to detect and quantify proteins, metabolites, and lipids. After normalization, non-parametric univariate analyses of cases versus controls were performed to identify differentially abundant features on all available samples for each assay. Unsupervised learning was used to investigate the separation of subjects into groups based on disease status for the subset of 157 subjects for which complete data on all 7 readouts were available. We detected 2,812 proteins, 811 lipids, 373 metabolites, and 110,864 mRNA transcripts in all samples where data for each assay was available. Of these, 275 proteins, 232 lipids, 49 metabolites, and 3385 mRNA transcripts were significantly different (FWER < 0.05) in cases versus controls. From cfDNA data, we identified 35 non-contiguous genomic regions associated with fragment-length disorder, 557 with copy-number variation, and 5 with multiple, differentially methylated CpGs (FWER < 0.05) that aggregately span 307 protein-coding genes; of these, the overlap with the differentially expressed proteins included E-cadherin (tumor suppressor) and N-cadherin (involved in epithelial-to-mesenchymal transition). Statistically significant genes and proteins were found to be associated with processes including Wnt signaling, regulation of focal adhesion assembly, and actin cytoskeleton organization. Multi-omics, unsupervised learning showed separation of early- and late-stage cases and controls. High-dimensional bioinformatics analyses systematically decomposed each `omics data type into joint and orthogonal components associated with pancreatic cancer. Ongoing multivariate analyses, including supervised machine learning, will further elucidate the biology of pancreatic cancer development, and serve as the foundation for high-sensitivity blood tests for the early detection and monitoring of pancreatic cancer. Citation Format: Ehdieh Khaledian, Preethi Prasad, John Blume, Ghristine Bundalian, Connie Chao-Shern, Jinlyung Choi, Rea Cuaresma, Jared Deyarmin, Jun Heok Jang, Manoj Khadka, Thidar Khin, Yuya Kodama, Ajinkya Kokate, Joon-Yong Lee, Manway Liu, Nithya Mudaliar, Madhuvanthi Ramaiah, Saividya Ramaswamy, Peter Spiro, Kavya Swaminathan, Preston Williams, Mi Yang, James Yee, Brian Young, Robert Zawada, Susan Zhang, Chinmay Belthangady, Bruce Wilcox, Philip Ma. High-dimensional, multi-omics analyses of proteins, metabolites, transcripts, and genes enable biomarker discovery in early- and late-stage pancreatic cancer [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer; 2022 Sep 13-16; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2022;82(22 Suppl):Abstract nr A038.

Copy number variants (CNVs) are among the main genetic factors identified in schizophrenia (SZ) through genome-scale studies conducted mostly in Caucasian populations. However, to date, there have been no genome-scale CNV reports on patients from India. To address this shortcoming, we generated, for the first time, genome-scale CNV data for 168 SZ patients and 168 controls from South India. In total, 63 different CNVs were identified in 56 patients and 46 controls with a significantly higher proportion of medium-sized deletions (100 kb-1 Mb) after multiple testing (FDR = 2.7E-4) in patients. Of these, 13 CNVs were previously reported; however, when searched against GWAS, transcriptome, exome, and DNA methylation studies, another 17 CNVs with candidate genes were identified. Of the total 30 CNVs, 28 were present in 38 patients and 12 in 27 controls, indicating a significantly higher representation in the former (p = 1.87E-5). Only 4q35.1-q35.2 duplications were significant (p = 0.020) and observed in 11 controls and 2 patients. Among the others that are not significant, a few examples of patient-specific and previously reported CNVs include deletions of 11q14.1 (DLG2), 22q11.21, and 14q21.1 (LRFN5). 16p13.3 deletion (RBFOX1), 3p14.2 duplication (CADPS), and 7p11.2 duplication (CCT6A) were some of the novel CNVs containing candidate genes. However, these observations need to be replicated in a larger sample size. In conclusion, this report constitutes an important foundation for future CNV studies in a relatively unexplored population. In addition, the data indicate that there are advantages in using an integrated approach for better identification of candidate CNVs for SZ and other mental health disorders.

BackgroundThrough the systematic large-scale profiling of metabolites, metabolomics provides a tool for biomarker discovery and improving disease monitoring, diagnosis, prognosis, and treatment response, as well as for delineating disease mechanisms and etiology. As a downstream product of the genome and epigenome, transcriptome, and proteome activity, the metabolome can be considered as being the most proximal correlate to the phenotype. Integration of metabolomics data with other -omics data in multi-omics analyses has the potential to advance understanding of human disease development and treatment.Aim of reviewTo understand the current funding and potential research opportunities for when metabolomics is used in human multi-omics studies, we cross-sectionally evaluated National Institutes of Health (NIH)-funded grants to examine the use of metabolomics data when collected with at least one other -omics data type. First, we aimed to determine what types of multi-omics studies included metabolomics data collection. Then, we looked at those multi-omics studies to examine how often grants employed an integrative analysis approach using metabolomics data.Key scientific concepts of reviewWe observed that the majority of NIH-funded multi-omics studies that include metabolomics data performed integration, but to a limited extent, with integration primarily incorporating only one other -omics data type. Some opportunities to improve data integration may include increasing confidence in metabolite identification, as well as addressing variability between -omics approach requirements and -omics data incompatibility.

DNA methylation plays a critical role in breast cancer development. Previous studies have identified DNA methylation marks in white blood cells as promising biomarkers for breast cancer. However, these studies were limited by low statistical power and potential biases. Using a new methodology, we investigated DNA methylation marks for their associations with breast cancer risk. Statistical models were built to predict levels of DNA methylation marks using genetic data and DNA methylation data from HumanMethylation450 BeadChip from the Framingham Heart Study (n = 1595). The prediction models were validated using data from the Women's Health Initiative (n = 883). We applied these models to genomewide association study (GWAS) data of 122 977 breast cancer patients and 105 974 controls to evaluate if the genetically predicted DNA methylation levels at CpG sites (CpGs) are associated with breast cancer risk. All statistical tests were two-sided. Of the 62 938 CpG sites CpGs investigated, statistically significant associations with breast cancer risk were observed for 450 CpGs at a Bonferroni-corrected threshold of P less than 7.94 × 10-7, including 45 CpGs residing in 18 genomic regions, that have not previously been associated with breast cancer risk. Of the remaining 405 CpGs located within 500 kilobase flaking regions of 70 GWAS-identified breast cancer risk variants, the associations for 11 CpGs were independent of GWAS-identified variants. Integrative analyses of genetic, DNA methylation, and gene expression data found that 38 CpGs may affect breast cancer risk through regulating expression of 21 genes. Our new methodology can identify novel DNA methylation biomarkers for breast cancer risk and can be applied to other diseases.

The multiomics data are heterogeneous and come from different biological levels such as epigenetics, genomics, transcriptomics and proteomics. The development of high-throughput technologies has enabled researchers not only to study all the entities together but also to utilize information from different levels spanning DNA methylation, copy number variation (CNV), mutation, gene expression, and miRNA expression. With the recent advancement in image informatics, the field of radiomics is rapidly emerging. It can be expected that the information from microscopic images of the tissue will soon be part of many multiomics studies. Meanwhile, integration of different kinds of multiomics data to extract relevant biological information is currently a big challenge. This study is our ongoing effort to develop a model that properly integrates multiomics data and allows easy retrieval of information relevant to biological processes. In this article, we have enriched our previous graph database model to store gene expression, miRNA expression, DNA methylation, mutation, CNV, clinical data, including information of the image of tissue slides. To show that the model is working, we used data from the Cancer Genome Atlas for three cancer types.

Purpose To develop a molecular based classification of non-small cell lung carcinomas based on genome wide copy number variations (CNVs). Background For a number of clinical and biologic reasons, the accurate classification of non small cell lung carcinoma (NSCLC) into adenocarcinoma (ADC) and squamous cell carcinoma (SCC) is essential. DNA based tests, which are not currently used, are more robust when applied to formalin fixed paraffin embedded tissues. Materials and Methods TCGA Dataset: The Cancer Genome Atlas project (TCGA) level 3 CNVs data (Affymetrix Genome-Wide Human SNP Array 6.0) of resected ADC (n = 241) and SCC (n = 210) patients were utilized as a Training set along with 1091 non malignant lung samples. SPORE Patient Dataset: The UT Lung SPORE cohort consists of 248 resected lung cancers (168 ADC and 74 SCC samples) run with the Agilent 244K Human Genome CGH Microarray. Molecular signature identification based on the morphologic subclassification of NSCLC: We used statistical algorithms to identify potential CNV biomarkers and combined them with known amplified oncogenes to identify 28 CNV signature genes that were highly correlated with histological classification. Identification of CNV classifier genes. The CNV signature genes were identified through the following four sequential steps: 1) Paired t-test. 2) Elastic Net. 3) Partial lease squares algorithm. 4) Naive Bayes classifier. Results: The 28 gene CNV signature accurately separated squamous cell carcinomas from adenocarcinomas in the Training and Validation sets as well as distinguished tumors from non-malignant tissues (Table 1). Table 1. The classification results between ADC and SCC DataSensitivitySpecificityAccuracyADC vs SCCTraining set (TCGA)0.940.870.91Validation set (SPORE)0.870.840.86Tumor vs Non-malignantADC0.910.960.94SCC0.970.980.98 Conclusions: A 28 gene CNV signature distinguished lung adenocarcinomas from squamous cell carcinomas with great accuracy (86-91%) and the lung tumor samples can be distinguished from the non-malignant lung samples with an accuracy of 94-98%. Citation Format: Kai Song, Guangqiang Zheng, Luc Girard, Ignacio I. Wistuba, Jack A. Roth, Carmen Behrens, Milind B. Suraokar, John D. Minna, Adi F. Gazdar. Copy number variations distinguish lung adenocarcinomas from squamous cell carcinomas. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 613. doi:10.1158/1538-7445.AM2015-613

MLDeCNV: A machine learning approach for predicting copy number variation types in plant genomes.