A Comparative Analysis of Predictive Learning Algorithms on High-Dimensional Microarray Cancer Data

This research evaluates pattern recognition techniques on a subclass of big data where the dimensionality of the input space (p) is much larger than the number of observations (n). Specifically, we evaluate massive gene expression microarray cancer data where the ratio κ is less than one. We explore the statistical and computational challenges inherent in these high dimensional low sample size (HDLSS) problems and present statistical machine learning methods used to tackle and circumvent these difficulties. Regularization and kernel algorithms were explored in this research using seven datasets where κ < 1. These techniques require special attention to tuning necessitating several extensions of cross-validation to be investigated to support better predictive performance. While no single algorithm was universally the best predictor, the regularization technique produced lower test errors in five of the seven datasets studied.

SplicerAV: a tool for mining microarray expression data for changes in RNA processing.

BACKGROUND: Over the past two decades more than fifty thousand unique clinical and biological samples have been assayed using the Affymetrix HG-U133 and HG-U95 GeneChip microarray platforms. This substantial repository has been used extensively to characterize changes in gene expression between biological samples, but has not been previously mined en masse for changes in mRNA processing. We explored the possibility of using HG-U133 microarray data to identify changes in alternative mRNA processing in several available archival datasets. RESULTS: Data from these and other gene expression microarrays can now be mined for changes in transcript isoform abundance using a program described here, SplicerAV. Using in vivo and in vitro breast cancer microarray datasets, SplicerAV was able to perform both gene and isoform specific expression profiling within the same microarray dataset. Our reanalysis of Affymetrix U133 plus 2.0 data generated by in vitro over-expression of HRAS, E2F3, beta-catenin (CTNNB1), SRC, and MYC identified several hundred oncogene-induced mRNA isoform changes, one of which recognized a previously unknown mechanism of EGFR family activation. Using clinical data, SplicerAV predicted 241 isoform changes between low and high grade breast tumors; with changes enriched among genes coding for guanyl-nucleotide exchange factors, metalloprotease inhibitors, and mRNA processing factors. Isoform changes in 15 genes were associated with aggressive cancer across the three breast cancer datasets. CONCLUSIONS: Using SplicerAV, we identified several hundred previously uncharacterized isoform changes induced by in vitro oncogene over-expression and revealed a previously unknown mechanism of EGFR activation in human mammary epithelial cells. We analyzed Affymetrix GeneChip data from over 400 human breast tumors in three independent studies, making this the largest clinical dataset analyzed for en masse changes in alternative mRNA processing. The capacity to detect RNA isoform changes in archival microarray data using SplicerAV allowed us to carry out the first analysis of isoform specific mRNA changes directly associated with cancer survival.

On the simultaneous use of clinical and microarray expression data in the cluster analysis of tissue samples

This paper considers a model-based approach to the clustering of tissue samples of a very large number of genes from microarray experiments. It is a nonstandard problem in parametric cluster analysis because the dimension of the feature space (the number of genes) is typically much greater than the number of tissues. Frequently in practice, there are also clinical data available on those cases on which the tissue samples have been obtained. Here we investigate how to use the clinical data in conjunction with the microarray gene expression data to cluster the tissue samples. We propose two mixture model-based approaches in which the number of components in the mixture model corresponds to the number of clusters to be imposed on the tissue samples. One approach specifies the components of the mixture model to be the conditional distributions of the microarray data given the clinical data with the mixing proportions also conditioned on the latter data. Another takes the components of the mixture model to represent the joint distributions of the clinical and microarray data. The approaches are demonstrated on some breast cancer data, as studied recently in van't Veer et al. (2002).

Prognostic factors in patients with advanced non-small cell lung cancer : data from the phase III FLEX study

The FLEX study demonstrated that the addition of cetuximab to chemotherapy significantly improved overall survival in the first-line treatment of patients with advanced non-small cell lung cancer (NSCLC). In the FLEX intention to treat (ITT) population, we investigated the prognostic significance of particular baseline characteristics. Individual patient data from the treatment arms of the ITT population of the FLEX study were combined. Univariable and multivariable Cox regression models were used to investigate variables with potential prognostic value. The ITT population comprised 1125 patients. In the univariable analysis, longer median survival times were apparent for females compared with males (12.7 vs 9.3 months); patients with an Eastern Cooperative Oncology Group performance status (ECOG PS) of 0 compared with 1 compared with 2 (13.5 vs 10.6 vs 5.9 months); never smokers compared with former smokers compared with current smokers (14.6 vs 11.1 vs 9.0); Asians compared with Caucasians (19.5 vs 9.6 months); patients with adenocarcinoma compared with squamous cell carcinoma (12.4 vs 9.3 months) and those with metastases to one site compared with two sites compared with three or more sites (12.4 months vs 9.8 months vs 6.4 months). Age (<65 vs ≥65 years), tumor stage (IIIB with pleural effusion vs IV) and percentage of tumor cells expressing EGFR (<40% vs ≥40%) were not identified as possible prognostic factors in relation to survival time. In multivariable analysis, a stepwise selection procedure identified age (<65 vs ≥65 years), gender, ECOG PS, smoking status, region, tumor histology, and number of organs involved as independent factors of prognostic value. In summary, in patients with advanced NSCLC enrolled in the FLEX study, and consistent with previous analyses, particular patient and disease characteristics at baseline were shown to be independent factors of prognostic value. The FLEX study is registered with ClinicalTrials.gov, number NCT00148798. © 2012 Elsevier Ireland Ltd.

Spatio-temporal modelling of cancer data in Queensland using Bayesian methods

Cancer is the leading contributor to the disease burden in Australia. This thesis develops and applies Bayesian hierarchical models to facilitate an investigation of the spatial and temporal associations for cancer diagnosis and survival among Queenslanders. The key objectives are to document and quantify the importance of spatial inequalities, explore factors influencing these inequalities, and investigate how spatial inequalities change over time. Existing Bayesian hierarchical models are refined, new models and methods developed, and tangible benefits obtained for cancer patients in Queensland. The versatility of using Bayesian models in cancer control are clearly demonstrated through these detailed and comprehensive analyses.

Investigating local relationships between trace elements in soils and cancer data

Medical geology research has recognised a number of potentially toxic elements (PTEs), such as arsenic, cobalt, chromium, copper, nickel, lead, vanadium, uranium and zinc, known to influence human disease by their respective deficiency or toxicity. As the impact of infectious diseases has decreased and the population ages, so cancer has become the most common cause of death in developed countries including Northern Ireland. This research explores the relationship between environmental exposure to potentially toxic elements in soil and cancer disease data across Northern Ireland. The incidence of twelve different cancer types (lung, stomach, leukaemia, oesophagus, colorectal, bladder, kidney, breast, mesothelioma, melanoma and non melanoma(NM) both basal and squamous, were examined in the form of twenty-five coded datasets comprising aggregates over the 12 year period from 1993 to 2006. A local modelling technique,geographically weighted regression (GWR) is usedto explore the relationship between environmental exposure and cancer disease data. The results show comparisons of the geographical incidence of certain cancers (stomach and NM squamous skin cancer) in relation to concentrations of certain PTEs (arsenic levels in soils and radon were identified). Findings from the research have implications for regional human health risk assessments.

Model selection for prognostic time-to-event gene signature discovery with applications in early breast cancer data

Model selection between competing models is a key consideration in the discovery of prognostic multigene signatures. The use of appropriate statistical performance measures as well as verification of biological significance of the signatures is imperative to maximise the chance of external validation of the generated signatures. Current approaches in time-to-event studies often use only a single measure of performance in model selection, such as logrank test p-values, or dichotomise the follow-up times at some phase of the study to facilitate signature discovery. In this study we improve the prognostic signature discovery process through the application of the multivariate partial Cox model combined with the concordance index, hazard ratio of predictions, independence from available clinical covariates and biological enrichment as measures of signature performance. The proposed framework was applied to discover prognostic multigene signatures from early breast cancer data. The partial Cox model combined with the multiple performance measures were used in both guiding the selection of the optimal panel of prognostic genes and prediction of risk within cross validation without dichotomising the follow-up times at any stage. The signatures were successfully externally cross validated in independent breast cancer datasets, yielding a hazard ratio of 2.55 [1.44, 4.51] for the top ranking signature.

Failure-Free Survival and Radiotherapy in Patients With Newly Diagnosed Nonmetastatic Prostate Cancer: Data From Patients in the Control Arm of the STAMPEDE Trial

Importance: The natural history of patients with newly diagnosed high-risk nonmetastatic (M0) prostate cancer receiving hormone therapy (HT) either alone or with standard-of-care radiotherapy (RT) is not well documented. Furthermore, no clinical trial has assessed the role of RT in patients with node-positive (N+) M0 disease. The STAMPEDE Trial includes such individuals, allowing an exploratory multivariate analysis of the impact of radical RT.

Objective: To describe survival and the impact on failure-free survival of RT by nodal involvement in these patients.

Design, Setting, and Participants: Cohort study using data collected for patients allocated to the control arm (standard-of-care only) of the STAMPEDE Trial between October 5, 2005, and May 1, 2014. Outcomes are presented as hazard ratios (HRs) with 95% CIs derived from adjusted Cox models; survival estimates are reported at 2 and 5 years. Participants were high-risk, hormone-naive patients with newly diagnosed M0 prostate cancer starting long-term HT for the first time. Radiotherapy is encouraged in this group, but mandated for patients with node-negative (N0) M0 disease only since November 2011.

Exposures: Long-term HT either alone or with RT, as per local standard. Planned RT use was recorded at entry.

Main Outcomes and Measures: Failure-free survival (FFS) and overall survival.

Results: A total of 721 men with newly diagnosed M0 disease were included: median age at entry, 66 (interquartile range [IQR], 61-72) years, median (IQR) prostate-specific antigen level of 43 (18-88) ng/mL. There were 40 deaths (31 owing to prostate cancer) with 17 months' median follow-up. Two-year survival was 96% (95% CI, 93%-97%) and 2-year FFS, 77% (95% CI, 73%-81%). Median (IQR) FFS was 63 (26 to not reached) months. Time to FFS was worse in patients with N+ disease (HR, 2.02 [95% CI, 1.46-2.81]) than in those with N0 disease. Failure-free survival outcomes favored planned use of RT for patients with both N0M0 (HR, 0.33 [95% CI, 0.18-0.61]) and N+M0 disease (HR, 0.48 [95% CI, 0.29-0.79]).

Conclusions and Relevance: Survival for men entering the cohort with high-risk M0 disease was higher than anticipated at study inception. These nonrandomized data were consistent with previous trials that support routine use of RT with HT in patients with N0M0 disease. Additionally, the data suggest that the benefits of RT extend to men with N+M0 disease.

Trial Registration: clinicaltrials.gov Identifier: NCT00268476; ISRCTN78818544.

Mining tissue microarray data to uncover combinations of biomarker expression patterns that improve intermediate staging and grading of clear cell renal cell cancer

PURPOSE: Tumor stage and nuclear grade are the most important prognostic parameters of clear cell renal cell carcinoma (ccRCC). The progression risk of ccRCC remains difficult to predict particularly for tumors with organ-confined stage and intermediate differentiation grade. Elucidating molecular pathways deregulated in ccRCC may point to novel prognostic parameters that facilitate planning of therapeutic approaches. EXPERIMENTAL DESIGN: Using tissue microarrays, expression patterns of 15 different proteins were evaluated in over 800 ccRCC patients to analyze pathways reported to be physiologically controlled by the tumor suppressors von Hippel-Lindau protein and phosphatase and tensin homologue (PTEN). Tumor staging and grading were improved by performing variable selection using Cox regression and a recursive bootstrap elimination scheme. RESULTS: Patients with pT2 and pT3 tumors that were p27 and CAIX positive had a better outcome than those with all remaining marker combinations. A prolonged survival among patients with intermediate grade (grade 2) correlated with both nuclear p27 and cytoplasmic PTEN expression, as well as with inactive, nonphosphorylated ribosomal protein S6. By applying graphical log-linear modeling for over 700 ccRCC for which the molecular parameters were available, only a weak conditional dependence existed between the expression of p27, PTEN, CAIX, and p-S6, suggesting that the dysregulation of several independent pathways are crucial for tumor progression. CONCLUSIONS: The use of recursive bootstrap elimination, as well as graphical log-linear modeling for comprehensive tissue microarray (TMA) data analysis allows the unraveling of complex molecular contexts and may improve predictive evaluations for patients with advanced renal cancer.

Harnessing the complexity of gene expression data from cancer: from single gene to structural pathway methods

High-dimensional gene expression data provide a rich source of information because they capture the expression level of genes in dynamic states that reflect the biological functioning of a cell. For this reason, such data are suitable to reveal systems related properties inside a cell, e.g., in order to elucidate molecular mechanisms of complex diseases like breast or prostate cancer. However, this is not only strongly dependent on the sample size and the correlation structure of a data set, but also on the statistical hypotheses tested. Many different approaches have been developed over the years to analyze gene expression data to (I) identify changes in single genes, (II) identify changes in gene sets or pathways, and (III) identify changes in the correlation structure in pathways. In this paper, we review statistical methods for all three types of approaches, including subtypes, in the context of cancer data and provide links to software implementations and tools and address also the general problem of multiple hypotheses testing. Further, we provide recommendations for the selection of such analysis methods.