Cross-Platform Omics Prediction procedure: a statistical machine learning framework for wider implementation of precision medicine.

In this modern era of precision medicine, molecular signatures identified from advanced omics technologies hold great promise to better guide clinical decisions. However, current approaches are often location-specific due to the inherent differences between platforms and across multiple centres, thus limiting the transferability of molecular signatures. We present Cross-Platform Omics Prediction (CPOP), a penalised regression model that can use omics data to predict patient outcomes in a platform-independent manner and across time and experiments. CPOP improves on the traditional prediction framework of using gene-based features by selecting ratio-based features with similar estimated effect sizes. These components gave CPOP the ability to have a stable performance across datasets of similar biology, minimising the effect of technical noise often generated by omics platforms. We present a comprehensive evaluation using melanoma transcriptomics data to demonstrate its potential to be used as a critical part of a clinical screening framework for precision medicine. Additional assessment of generalisation was demonstrated with ovarian cancer and inflammatory bowel disease studies.

Whole-genome landscapes of major melanoma subtypes.

Melanoma of the skin is a common cancer only in Europeans, whereas it arises in internal body surfaces (mucosal sites) and on the hands and feet (acral sites) in people throughout the world. Here we report analysis of whole-genome sequences from cutaneous, acral and mucosal subtypes of melanoma. The heavily mutated landscape of coding and non-coding mutations in cutaneous melanoma resolved novel signatures of mutagenesis attributable to ultraviolet radiation. However, acral and mucosal melanomas were dominated by structural changes and mutation signatures of unknown aetiology, not previously identified in melanoma. The number of genes affected by recurrent mutations disrupting non-coding sequences was similar to that affected by recurrent mutations to coding sequences. Significantly mutated genes included BRAF, CDKN2A, NRAS and TP53 in cutaneous melanoma, BRAF, NRAS and NF1 in acral melanoma and SF3B1 in mucosal melanoma. Mutations affecting the TERT promoter were the most frequent of all; however, neither they nor ATRX mutations, which correlate with alternative telomere lengthening, were associated with greater telomere length. Most melanomas had potentially actionable mutations, most in components of the mitogen-activated protein kinase and phosphoinositol kinase pathways. The whole-genome mutation landscape of melanoma reveals diverse carcinogenic processes across its subtypes, some unrelated to sun exposure, and extends potential involvement of the non-coding genome in its pathogenesis.

A multi-step classifier addressing cohort heterogeneity improves performance of prognostic biomarkers in three cancer types.

Cancer research continues to highlight the extensive genetic diversity that exists both between and within tumors. This intrinsic heterogeneity poses one of the central challenges to predicting patient clinical outcome and the personalization of treatments. Despite progress in some individual tumor types, it is not yet possible to prospectively, accurately classify patients by expected survival. One hypothesis proposed to explain this is that the prognostic classifiers developed to date are insufficiently sensitive and specific; however it is also possible that patients are not equally easy to classify by any given biomarker. We demonstrate in a cohort of 45 AJCC stage III melanoma patients that clinico-pathologic biomarkers can identify those patients that are most likely to be misclassified by a molecular biomarker. The process of modelling the classifiability of patients was then replicated in a cohort of 49 stage II breast cancer patients and 53 stage III colon cancer patients. A multi-step procedure incorporating this information not only improved classification accuracy but also indicated the specific clinical attributes that had made classification problematic in each cohort. These findings show that, even when cohorts are of moderate size, including features that explain the patient-specific performance of a prognostic biomarker in a classification framework can improve the modelling and estimation of survival.

Identification, Review, and Systematic Cross-Validation of microRNA Prognostic Signatures in Metastatic Melanoma.

In metastatic melanoma, it is vital to identify and validate biomarkers of prognosis. Previous studies have systematically evaluated protein biomarkers or mRNA-based expression signatures. No such analyses have been applied to microRNA (miRNA)-based prognostic signatures. As a first step, we identified two prognostic miRNA signatures from publicly available data sets (Gene Expression Omnibus/The Cancer Genome Atlas) of global miRNA expression profiling information. A 12-miRNA signature predicted longer survival after surgery for resection of American Joint Committee on Cancer stage III disease (>4 years, no sign of relapse) and outperformed American Joint Committee on Cancer standard-of-care prognostic markers in leave-one-out cross-validation analysis (error rates 34% and 38%, respectively). A similar 15-miRNA biomarker derived from The Cancer Genome Atlas miRNA-seq data performed slightly worse (39%) than these current biomarkers. Both signatures were then assessed for replication in two independent data sets and subjected to systematic cross-validation together with the three other miRNA-based prognostic signatures proposed in the literature to date. Five miRNAs (miR-142-5p, miR-150-5p, miR-342-3p, miR-155-5p, and miR-146b-5p) were reproducibly associated with patient outcome and have the greatest potential for application in the clinic. Our extensive validation approach highlighted among multiple independent cohorts the translational potential and limitations of miRNA signatures, and pointed to future directions in the analysis of this emerging class of markers.

Determination of prognosis in metastatic melanoma through integration of clinico-pathologic, mutation, mRNA, microRNA, and protein information.

In patients with metastatic melanoma, the identification and validation of accurate prognostic biomarkers will assist rational treatment planning. Studies based on "-omics" technologies have focussed on a single high-throughput data type such as gene or microRNA transcripts. Occasionally, these features have been evaluated in conjunction with limited clinico-pathologic data. With the increased availability of multiple data types, there is a pressing need to tease apart which of these sources contain the most valuable prognostic information. We evaluated and integrated several data types derived from the same tumor specimens in AJCC stage III melanoma patients-gene, protein, and microRNA expression as well as clinical, pathologic and mutation information-to determine their relative impact on prognosis. We used classification frameworks based on pre-validation and bootstrap multiple imputation to compare the prognostic power of each data source, both individually as well as integratively. We found that the prognostic utility of clinico-pathologic information was not out-performed by any of the various "-omics" platforms. Rather, a combination of clinico-pathologic variables and mRNA expression data performed best. Furthermore, a patient-based classification analysis revealed that the prognostic accuracy of various data types was not the same for different patients. This indicates that ongoing development in the individualized evaluation of melanoma patients must take account of the value of both traditional and novel "-omics" measurements.

MicroRNA and mRNA expression profiling in metastatic melanoma reveal associations with BRAF mutation and patient prognosis.

The role of microRNAs (miRNAs) in melanoma is unclear. We examined global miRNA expression profiles in fresh-frozen metastatic melanomas in relation to clinical outcome and BRAF mutation, with validation in independent cohorts of tumours and sera. We integrated miRNA and mRNA information from the same samples and elucidated networks associated with outcome and mutation. Associations with prognosis were replicated for miR-150-5p, miR-142-3p and miR-142-5p. Co-analysis of miRNA and mRNA uncovered a network associated with poor prognosis (PP) that paradoxically favoured expression of miRNAs opposing tumorigenesis. These miRNAs are likely part of an autoregulatory response to oncogenic drivers, rather than drivers themselves. Robust association of miR-150-5p and the miR-142 duplex with good prognosis and earlier stage metastatic melanoma supports their potential as biomarkers. miRNAs overexpressed in association with PP in an autoregulatory fashion will not be suitable therapeutic targets.

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

BACKGROUND: Classical approaches to predicting patient clinical outcome via gene expression information are primarily based on differential expression of unrelated genes (single-gene approaches) or genes related by, for example, biologic pathway or function (gene-sets). Recently, network-based approaches utilising interaction information between genes have emerged. An open problem is whether such approaches add value to the more traditional methods of signature modelling. We explored this question via comparison of the most widely employed single-gene, gene-set, and network-based methods, using gene expression microarray data from two different cancers: melanoma and ovarian. We considered two kinds of network approaches. The first of these identifies informative genes using gene expression and network connectivity information combined, the latter drawn from prior knowledge of protein-protein interactions. The second approach focuses on identification of informative sub-networks (small networks of interacting proteins, again from prior knowledge networks). For all methods we performed 100 rounds of 5-fold cross-validation under 3 different classifiers. For network-based approaches, we considered two different protein-protein interaction networks. We quantified resulting patterns of misclassification and discussed the relative value of each relative to ongoing development of prognostic biomarkers. RESULTS: We found that single-gene, gene-set and network methods yielded similar error rates in melanoma and ovarian cancer data. Crucially, however, our novel and detailed patient-level analyses revealed that the different methods were correctly classifying alternate subsets of patients in each cohort. We also found that the network-based NetRank feature selection method was the most stable. CONCLUSIONS: Next-generation methods of gene expression signature modelling harness data from external networks and are foreshadowed as a standard mode of analysis. But what do they add to traditional approaches? Our findings indicate there is value in the way in which different subspaces of the patient sample are captured differently among the various methods, highlighting the possibility of 'combination' classifiers capable of identifying which patients will be more accurately classified by one particular method over another. We have seen this clearly for the first time because of our in-depth analysis at the level of individual patients.

Molecular interaction networks for the analysis of human disease: utility, limitations, and considerations.

High-throughput '-omics' data can be combined with large-scale molecular interaction networks, for example, protein-protein interaction networks, to provide a unique framework for the investigation of human molecular biology. Interest in these integrative '-omics' methods is growing rapidly because of their potential to understand complexity and association with disease; such approaches have a focus on associations between phenotype and "network-type." The potential of this research is enticing, yet there remain a series of important considerations. Here, we discuss interaction data selection, data quality, the relative merits of using data from large high-throughput studies versus a meta-database of smaller literature-curated studies, and possible issues of sociological or inspection bias in interaction data. Other work underway, especially international consortia to establish data formats, quality standards and address data redundancy, and the improvements these efforts are making to the field, is also evaluated. We present options for researchers intending to use large-scale molecular interaction networks as a functional context for protein or gene expression data, including microRNAs, especially in the context of human disease.

VAN: an R package for identifying biologically perturbed networks via differential variability analysis.

BACKGROUND: Large-scale molecular interaction networks are dynamic in nature and are of special interest in the analysis of complex diseases, which are characterized by network-level perturbations rather than changes in individual genes/proteins. The methods developed for the identification of differentially expressed genes or gene sets are not suitable for network-level analyses. Consequently, bioinformatics approaches that enable a joint analysis of high-throughput transcriptomics datasets and large-scale molecular interaction networks for identifying perturbed networks are gaining popularity. Typically, these approaches require the sequential application of multiple bioinformatics techniques - ID mapping, network analysis, and network visualization. Here, we present the Variability Analysis in Networks (VAN) software package: a collection of R functions to streamline this bioinformatics analysis. FINDINGS: VAN determines whether there are network-level perturbations across biological states of interest. It first identifies hubs (densely connected proteins/microRNAs) in a network and then uses them to extract network modules (comprising of a hub and all its interaction partners). The function identifySignificantHubs identifies dysregulated modules (i.e. modules with changes in expression correlation between a hub and its interaction partners) using a single expression and network dataset. The function summarizeHubData identifies dysregulated modules based on a meta-analysis of multiple expression and/or network datasets. VAN also converts protein identifiers present in a MITAB-formatted interaction network to gene identifiers (UniProt identifier to Entrez identifier or gene symbol using the function generatePpiMap) and generates microRNA-gene interaction networks using TargetScan and Microcosm databases (generateMicroRnaMap). The function obtainCancerInfo is used to identify hubs (corresponding to significantly perturbed modules) that are already causally associated with cancer(s) in the Cancer Gene Census database. Additionally, VAN supports the visualization of changes to network modules in R and Cytoscape (visualizeNetwork and obtainPairSubset, respectively). We demonstrate the utility of VAN using a gene expression data from metastatic melanoma and a protein-protein interaction network from the Human Protein Reference Database. CONCLUSIONS: Our package provides a comprehensive and user-friendly platform for the integrative analysis of -omics data to identify disease-associated network modules. This bioinformatics approach, which is essentially focused on the question of explaining phenotype with a 'network type' and in particular, how regulation is changing among different states of interest, is relevant to many questions including those related to network perturbations across developmental timelines.

Disturbed protein-protein interaction networks in metastatic melanoma are associated with worse prognosis and increased functional mutation burden.

For disseminated melanoma, new prognostic biomarkers and therapeutic targets are urgently needed. The organization of protein-protein interaction networks was assessed via the transcriptomes of four independent studies of metastatic melanoma and related to clinical outcome and MAP-kinase pathway mutations (BRAF/NRAS). We also examined patient outcome-related differences in a predicted network of microRNAs and their targets. The 32 hub genes with the most reproducible survival-related disturbances in co-expression with their protein partner genes included oncogenes and tumor suppressors, previously known correlates of prognosis, and other proteins not previously associated with melanoma outcome. Notably, this network-based gene set could classify patients according to clinical outcomes with 67-80% accuracy among cohorts. Reproducibly disturbed networks were also more likely to have a higher functional mutation burden than would be expected by chance. The disturbed regions of networks are therefore markers of clinically relevant, selectable tumor evolution in melanoma which may carry driver mutations.

BRAF mutation, NRAS mutation, and the absence of an immune-related expressed gene profile predict poor outcome in patients with stage III melanoma.

Prediction of outcome for melanoma patients with surgically resected macroscopic nodal metastases is very imprecise. We performed a comprehensive clinico-pathologic assessment of fresh-frozen macroscopic nodal metastases and the preceding primary melanoma, somatic mutation profiling, and gene expression profiling to identify determinants of outcome in 79 melanoma patients. In addition to disease stage 4 years, 90% confidence interval): the presence of a nodular component in the primary melanoma (6.8, 0.6-76.0), and small cell size (11.1, 0.8-100.0) or low pigmentation (3.0, 0.8-100.0) in the nodal metastases. Absence of BRAF mutation (20.0, 1.0-1000.0) or NRAS mutation (16.7, 0.6-1000.0) were both favorable prognostic factors. A 46-gene expression signature with strong overrepresentation of immune response genes was predictive of better survival (10.9, 0.4-325.6); in the full cohort, median survival was >100 months in those with the signature, but 10 months in those without. This relationship was validated in two previously published independent stage III melanoma data sets. We conclude that the presence of BRAF mutation, NRAS mutation, and the absence of an immune-related expressed gene profile predict poor outcome in melanoma patients with macroscopic stage III disease.

Review and cross-validation of gene expression signatures and melanoma prognosis.

In melanoma, there is an urgent need to identify novel biomarkers with prognostic performance superior to traditional clinical and histological parameters. Gene expression-based prognostic signatures offer promise, but studies have been challenged by sample scarcity, cohort heterogeneity, and doubts about the efficacy of such signatures relative to current clinical practices. Motivated by new studies that have begun to address these challenges, we reviewed prognostic signatures derived from gene expression microarray analysis of human melanoma tissue. We used REMARK-based criteria to select the most relevant studies and directly compared their signature gene lists. Through functional ontology enrichment analysis, we observed that these independent data sets converge in part upon immune response processes and the G-protein signaling NRAS-regulation pathway, both important in melanoma development and progression. The signatures correctly predicted patient outcome in independent gene expression data sets with some notably low misclassification rates, particularly among studies involving more advanced-stage tumors. This successful cross-validation indicates that gene expression analysis-based signatures are becoming translationally relevant to care of melanoma patients, as well as improving understanding of the aspects of melanoma biology that determine patient outcome.

Melanoma prognosis: a REMARK-based systematic review and bioinformatic analysis of immunohistochemical and gene microarray studies.

Despite intensive research efforts, within-stage survival rates for melanoma vary widely. Pursuit of molecular biomarkers with improved prognostic significance over clinicohistological measures has produced extensive literature. Reviews have synthesized these data, but none have systematically partitioned high-quality studies from the remainder across different molecular methods nor examined system properties of that output. Databases were searched for studies analyzing protein expression by immunohistochemistry (n = 617, extending the only systematic review to date by 102 studies) or for gene expression microarray studies (n = 45) in melanoma in relation to outcome. REMARK-derived criteria were applied to identify high-quality studies. Biomarkers and pathways were functionally assessed by using gene ontology software. Most manuscripts did not meet REMARK-based criteria, an ongoing trend that can impede translational research. Across REMARK-compliant literature, 41 proteins were significantly associated with outcome. Multimarker tests consistently emerged among the most promising potential biomarkers, indicating a need to continue assessing candidates in that composite setting. Twenty-one canonical pathways were populated by outcome-related proteins but not by those that failed to show such an association; we propose that this set of pathways warrants closer investigation to understand drivers of poor outcome in melanoma. Two-gene expression microarray studies met REMARK-based criteria reflecting a genuine paucity of literature in the area. The 254 outcome-related genes were examined for correspondences with the systematically identified protein signature. This analysis highlighted proliferating cell nuclear antigen and survivin as priorities for further examination as biomarkers in melanoma prognosis, and illustrated ongoing need to integrate alternative approaches to biomarker discovery in melanoma translational research.