Dr. Paul C. Boutros, Principal Investigator at Informatics & Biocomputing
Ontario Institute for Cancer Research

Collaborator:

Dr. Fei-Fei Liu, Professor at Department of Radiation Oncology, University of Toronto, Staff Radiation Oncologist at Princess Margaret Hospital/University Health Network, and Senior Scientist at Ontario Cancer Institute

Location:

Ontario Institute for Cancer Research, Toronto

Research Plan:

Cervix cancer is the second most common malignancy in women worldwide, yet to date there are no validated prognostic biomarkers in clinical use. In fact, very few such studies have yet been performed. To fill this gap, we comprehensively profiled miRNA and mRNA expression as a companion biomarker study to our prospective cervix cancer clinical trial evaluating the prognostic impact of hypoxia and other micro‐environmental markers on this disease. To date, over 300 women with cervix cancer have been enrolled since 1994, with both frozen and paraffin‐embedded tumour biopsies banked, all linked with clinical annotations including treatment, outcome, hypoxia and interstitial fluid pressure (IFP) measures. Our hypothesis is that both miRNA and mRNA expression patterns can predict cervix cancer relapse. We further hypothesize that miRNA and mRNA expression profiles are associated with important micro‐environmental characteristics of cervical tumours, such as hypoxia.

This project will involve analysis of some key issues in the development of so‐called “multi‐model biomarkers”. These are biomarkers that integrate diverse and different types of information to make unified predictions about patient characteristics. For example, in this case we will be modeling the relationship between multiple distinct molecular characteristics (tumour hypoxia, miRNA and mRNA levels), clinical covariates (stage, grade, nodal status), and patient survival. The intern will learn about a number of key issues in these types of studies. In particular, focus will be on techniques to reduce the dimensionality of these types of datasets by exploiting covariance amongst the independent variables to increase effective power. Results will be validated on independent patient cohorts from across Canada. This project has the potential to result in real clinical benefit in the near‐term, and will expose the intern to sophisticated biological and biostatistical modeling.

The intern will be treated as a key member of the cervix cancer research team. They will participate in the weekly team meetings, and interact on a regular basis with clinicians, molecular researchers, bioinformaticians, and biostatisticians. Physically, they will be located within the Boutros Lab at OICR, which is a team of biologists, statisticians, engineers, and computer scientists working on techniques for biomarker discovery. They will be able to interact with Dr. Boutros on a regular basis, and will be expected to attend the weekly lab meetings, institutional seminars and journal clubs. OICR has a strong computing facility, including a >5000 core computing cluster and an svn repository for code management. Students will be treated as full members of the lab, and will have the opportunity to present their results in lab‐meetings and, if appropriate, to draft their own manuscripts for publication.

Dr. Paul C. Boutros, Principal Investigator at Informatics & Biocomputing
Ontario Institute for Cancer Research

Collaborator:

Multiple collaborators

Location:

Ontario Institute for Cancer Research, Toronto

Research Plan:

Glioblastoma multiforme (GBM), Ovarian and Colorectal cancers are amongst the most malignant cancers types. In 2010, an estimated 22,000, 21,888 and 142,270 patients were diagnosed with GBM, ovarian and colorectal cancer, respectively. There exist a number of molecular prognostic markers for these cancer types, yet the mortality rate remains very high. For instance, the GBM patients usually survive less than 15 months. The studies focused on deriving prognostic markers of clinical relevance are yet to evaluate the combined impact of multiple somatic mutations as a group, in particular the ones that are less prevalent. The cancer genome atlas (TCGA) project has produced a comprehensive catalogue of somatic mutations across these cancer types. These mutations are further characterized into respective consequence type depending upon the type of genomic alteration they cause, for instance insertion, deletion and inversions. However, these alterations eventually translate into a protein and therefore needs to be further characterized in terms of impact in the corresponding functional domain of the protein. This way, the deregulation impact of somatic mutations can be assessed in a functional context. The functional assessment would not only offer the per‐protein domain dysregulation across various binding sites, it also lends itself towards quantifying the impact of somatic mutations across various subunits of a protein. Following this, we can treat the protein domain‐wise deregulation scores as pseudo‐features to establish a multivariate statistical model to predict patient outcome. The project would involve collection of TCGA datasets, UCSC genomic coordinates of Human genome assembly vHg19 and protein domain coordinates. The intern would map the somatic mutations from genomic to functional coordinates. This would be followed by establishing a novel features selection approach to translate the domain‐wise mutational profiles into impact scores, and subsequent assessment of these scores by a multivariate statistical model. As an end product, we envisage to establish a generic framework that would facilitate integration of multiple somatic mutations and their prognostic evaluation.

The intern will be treated as a key member of the cancer bioinformatics research team. They will participate in the weekly team meetings, and interact on a regular basis with clinicians, molecular researchers, bioinformaticians, and biostatisticians. Physically, they will be located within the Boutros Lab at OICR, which is a team of biologists, statisticians, engineers, and computer scientists working on techniques for biomarker discovery. They will be able to interact with Dr. Boutros on a regular basis, and will be expected to attend the weekly lab meetings, institutional seminars and journal clubs. OICR has a strong computing facility, including a >5000 core computing cluster and an svn repository for code management. Students will be treated as full members of the lab, and will have the opportunity to present their results in lab‐meetings and, if appropriate, to draft their own manuscripts for publication.

Dr. Keyue Ding, Senior Biostatistician, NCIC-Clinical Trials Group, Associate Professor, Department of Community Health and Epidemiology, Queen’s University

Location:

Queen’s University, Kingston, Ontario

Research Plan:

In some biomarker studies accompanying clinical trials, serum/blood samples were collected at baseline and at certain time points after treatments. Usually based on the biological/medical background, those interested markers derived from the samples were found either relevant to the disease, i.e., the marker (baseline) may be the prognostic/predictive factor, or based on drug’s mechanism of action that the drug works on the tumor may first affect the markers, and the marker’s level of change lies in the causal pathway that the drug working on the tumor. And we can check if the markers’ changes after treatment can be predictive of outcome either for both arms (surrogate markers) or predictive of differential outcome between 2 treatment arms (predictive markers). Based on those early signs, we may predict which patients would be benefit from the drug, and who would not and could start for other alternative therapy earlier.

Statistical issues relevant to the question are how to model marker’s changes? Relative changes, absolute changes or use the marker values themselves in the statistical model? Addison and Ding et al (2010) model the relative changes. The relevant issue arises as to use the relative changes or to check some threshold effect in correlation to study outcome and determine the cutoffs? For convenient to use, and easy explanation of the results, cutoffs based on the threshold effect in correlation to outcome are common practice. Due to the outcome driven method, statistical issues of control the false discovery rate, and adjust bias in effect estimation would be key for common practice. Approximations of the false discovery rate, and bias in estimation will be developed. Simulation study and apply to real trial data will also be performed.

Dr. Wei Xu, Principal Biostatistician at the Princess Margaret Hospital, Scientist at the Ontario Cancer Institute, Assistant Professor of Biostatistics in School of Public Health, University of Toronto

Collaborator:

Dr. Geoffrey Liu, Alan B. Brown Chair in Molecular Genomics, University of Toronto, Cancer Care Ontario Research Chair in Experimental Therapeutics and Population Studies, medical oncologist at Princess Margaret Hospital, a pharmacogenetic epidemiologist, a Scientist at the Ontario Cancer Institute, and a Visiting Scientist at the Harvard School of Public Health

Location:

Princess Margaret Hospital, Toronto

Research Plan:

This study will delve into several of the most interesting aspects of translational cancer research, using human gastro-esophageal and lung cancer models. The uniqueness of these applications relates to the combined analyses of biomarkers identified from human patient tumours, and which are, in parallel, applied to analyses of human patient tumours implanted into immunocompromised mice. The ability to utilize such mouse model systems allows for analyses of multiple tumours from one person, which is then treated similarly or differently. Further, a large translational high impact clinical trials program has just started at Princess Margaret Hospital that will utilize such combined approaches in multiple tumour sites. The ability to work with both human and animal model systems, all in the context of clinical and observational trials, is a novel feature of this internship.

(1) Using germline samples directly from patients, pathway analytical approaches of pharmacogenetic and prognostic genetic sequence variants will be developed to identify the role of these variants in predicting tumour response and survival. These pathway biostatistical approaches represent one of the recent advances in analyzing and identifying candidate genes for validation. Data from a large genome-wide association study (of germline DNA) has recently been generated in human esophageal and lung cancer patients, matching the individuals above, and await analysis of their relationships with survival. These data from our Princess Margaret Hospital patients come from BEACON, the Barrett`s Esophageal and Adenocarcinoma Consortium (http://bea.tlvcloud.org/).

(2) Using parallel approaches in gene expression, methylation (epigenetics), copy number variant and mutation analyses will identify key genes altered during the treatment of human-tumour mouse model systems. Drugs used include both chemotherapy and biological agents-molecularly targeted agents, alone or in combination with radiation. Animals are treated multiple times and followed over time. Analyses of gene expression, methylation, copy number variation, and growth retardation over time will be modeled, using pure biostatistical and combined biostatistical and bioinformatical approaches.

(3) Matched to these xenografted mouse models are the corresponding human primary tumours, which will also be evaluated for gene expression, epigenetics, copy number variation and mutational analyses. Corresponding analyses of the human patient treatment and survival outcomes will also be evaluated. Protein evaluation through immunohistochemical staining of tissue microarrays will also be performed.

What is unique about this coordinated effort? The same patient will have tissue available for profiling across multiple platforms (germline DNA, tumour DNA, RNA, epigenetics, protein). These same patients have their clinical course and outcome completely charted. The same patient`s cancer will also have been implanted in immunocompromised mice. These mice are then subjected to various different treatment regimens, which are then compared for their efficacy against the tumours. Throughout the course of these mouse experiments, samples are taken and are analyzed across these same profiling platforms to identify markers of response to therapy, and their relationship to treatment responses in mice and in the original patient.

The uniqueness of this evaluation is the ability to simultaneously model, using biostatistical and bioinformatical approaches, clinical and pre-clinical outcomes, treatments, and biological high dimensionality data across multiple platforms. Some of the data will be coming from ASSET, a Phase II clinical trial of gastroesophageal cancers treated with chemotherapy and a biological agent. Further, the ability to integrate pre-clinical and clinical aspects of drug development will provide a broad introduction to the intern of the possibilities that lie in molecular and pharmacostatistical methodologies.

The intern will work in an interdisciplinary setting to develop methodologies to analyze, under Drs. Liu and Xu`s direction, the following specific aims:

Aim 1: Determine what clinico-pathological and molecular characteristics correlate with the engraftment of tumours in mice.

Aim 2: Determine how xenograft engraftment is connected to the survival of patients and how time-to-engraftment of xenografts is related to survival or to the tumour’s response to treatment in the patient.

Aim 3: Characterize the model by determining the profile differences between xenografts and the original patient tumour, early passage and late passage xenografts, and xenografts at two different time points of the same passage. A passage is defined as a generation of a mouse carrying the tumour. When the tumour grows too large, the mouse is sacrificed, and the tumour is re-implanted onto another mouse, where the new mouse is considered the next generation or passage of the tumour.

Aim 4: Determine the profile patterns and differences between untreated and treated xenografts. Treatments include, chemotherapy, biological therapy, radiotherapy, and their combinations.

Aim 5: Determine what profile (germline and tumour) are associated with treatment responses, and clinical outcomes in the patients with esophageal and lung cancer.

The training involved will allow the intern exposure to cross-disciplinary approaches to the spectrum of high dimensionality data analyses that intersects between biostatistics, molecular biology, and bioinformatics. The internship also seeks to provide the intern with experience with written and oral presentation, novel methodological development, and scientific writing.

Dr. Wei Xu, Principal Biostatistician at the Princess Margaret Hospital, Scientist at the Ontario Cancer Institute, Assistant Professor of Biostatistics in School of Public Health, University of Toronto

Collaborators:

Dr. Geoffrey Liu, Clinical Scientist, Alan B. Brown Chair in Molecular Genomics, University of Toronto, Cancer Care Ontario Research Chair in Experimental Therapeutics and Population Studies, medical oncologist at Princess Margaret Hospital, a pharmacogenetic epidemiologist, a Scientist at the Ontario Cancer Institute, and a Visiting Scientist at the Harvard School of Public Health

Dr. Fei-Fei Liu, Director of Head and Neck Translational Research Group, Princess Margaret Hospital

Location:

Princess Margaret Hospital, Toronto

Research Plan:

The objective of this study is to evaluate molecular prognostic markers in head and neck cancer (HNC) outcomes (1), as clinical prognostic factors are imprecise, and improving our ability to predict clinical outcomes is the cornerstone to individualized patient management. Inherited genetic variations (typically single nucleotide polymorphisms, SNPs) are being studied as prognostic factors in many cancers (2-12). SNPs can be measured by a simple blood test using available technologies in most clinical diagnostic laboratories (13). The overall objective is to test and validate a comprehensive list of SNPs as predictors of HNC outcomes, utilizing the latest SNP selection, bioinformatic and multivariate modeling strategies, multiple replication datasets, multistage pathway approaches, large well characterized patient populations, and clearly defined outcomes.

There are three datasets for this multi-stage study including screening cohort of 540 early stage Quebec HNC patients from a multi-center, double-blind, placebo-controlled randomized chemoprevention trial for stage I or II HNC “QUE” (14, 15, 16, 17), replication cohort of 389 early stage HNC patients from Princess Margaret Hospital/University Health Network (PMH/UHN), “PMH1”; and validation cohort of 279 late stage HNC patients from PMH “PMH2”.

There are multiple aims for this proposed study. Aim 1: To validate the association between HNC outcomes and 30 SNPs/SNP pathways identified from the published literature in the three datasets (. We hypothesize that each SNP/SNP pathway is independently associated with HNC outcome. Aim 2: To apply a genome-wide association (GWA) analysis between HNC outcomes and 600,000 SNPs genotyped for all QUE patients (Aim 2A). From Aim 2A, we will then identify the genome wide significant (p<5x10-7) SNPs to validate in PMH1 (Aim 2B) and explore in PMH2 (Aim 2C). In Aims 2B/2C, we hypothesize that each SNP is independently associated with HNC outcome. Aim 3: DNA repair dominates the published literature for SNPs and outcome in both HNCs and other aerodigestive tract cancers. We hypothesize that DNA repair capacity, as measured by the COMET assay, is associated with HNC outcomes in a prospective dataset (n=156), PMH3. Aim 4: To perform exploratory prognostic modeling. We will test the combined role of SNPs and serum prognostic markers (being validated in a separate NCIC grant, co-PIs Meyer/Bairati) in QUE, hypothesizing that SNPs of genes associated with these serum markers modify, interact, or are correlated with each other in HNC outcomes (Aim 4A). We will also test the combined role of significant SNPs identified in Project 4 and significant RNA signatures identified in Projects 1 & 2, as independent predictors of survival outcomes in a subset of PMH1/PMH2 (Aim 4B). Finally, we will test the association between various DNA repair gene SNPs and DNA repair capacity, as measured by the COMET assay (Aim 4C).

There are several research topics for the intern student to work based on this study depending on the student’s research interest. He/she can work on exploratory multivariate prognostic modeling of time to event data (18, 19). A long term goal of the study is to identify a prognostic tool that incorporates clinical and molecular factors, firstly by identifying the best individual markers within each discipline (RNA, SNP, dosimetry), with a future goal to validate across disciplines (20). The study integration provides opportunity to perform initial exploratory multi-disciplinary analyses. QUE will have both serologic and SNP data, and we will explore different serologic markers and associated SNPs within the same pathway. Likewise, a subset of PMH1 and all of PMH2 will have mRNA and miRNA data, thus allowing exploration of different SNP-gene expression signatures in relation to HNC outcomes. Finally, we will explore the relationship between the COMET assay and DNA repair SNPs that are associated with outcomes on validation or replication.

Besides that, he/she can work on the utility of bioinformatics & integrative databases. Using protein structure predictive software and sequence analysis programs, we can evaluate polymorphic variants in EGFR pathway genes. Hundreds of pathway SNPs can be chosen for further evaluation in an upcoming Phase III Intergroup study of advanced non-small cell lung cancer. In a separate analysis of colon cancer, OPHID (U of Toronto protein-protein interaction database) was used to identify Met-signaling targets that were then verified experimentally (21). The same database has also been used to help narrow gene expression-related prognostic biomarkers in ovarian cancer (22). Adapted bioinformatics and higher order statistical approaches can be applied for continuous variable outcomes, including cluster analysis to detect gene-gene interactions, neural networks/artificial intelligence, and adaptations of classification and regression tree (CART) analysis (23-25).

Another topic is the analysis of survival and toxicity. The impact of single SNP on each of the time to event outcomes (FFS, OS, recurrence-free rate, time to second primary event) will be analyzed independently, using Kaplan-Meier method to estimate survival curves; Log-rank test will be used to detect SNP effect on survival. Cox Proportional- Hazards (CPH) models will be developed for multivariate analysis. Logistic regression will be applied on toxicity rate. Covariates being considered for adjustment will include age, gender, disease stage, histologic site, radiation dosimetry metric/total radiation dose, performance status, treatment and ethnicity. Some variables can be treated as stratification variables instead of adjustment variables, as appropriate. For example, QUE treatments fell into 3 groups: α-tocopherol + β-carotene supplementation; α-tocopherol supplementation; and placebo (14, 15). To adjust for population heterogeneity, we can apply a stratified proportional hazard model to each SNP, which allows the baseline hazards to differ between strata and tests whether our results are simply driven by strata- specific baseline hazards (26).

For more information, click here …

Dr. Bingshu Chen, Senior Biostatistician at the NCIC Clinical Trials Group, Assistant Professor in the Department of Community Health and Epidemiology, Queen’s University

Location:

Queen’s University, Kingston

Research Plan:

NCIC CTG LY12 is a multi-centre, randomized phase III study coordinated by the NCIC Clinical Trials Group. The primary objective of this study is to compare objective response rates of patients with relapsed or refractory aggressive histology non-Hodgkin’s lymphoma after salvage treatment with two cycles of either gemcitabine, cisplatin, and dexamethasone (GDP) or dexamethasone, cytarabine, and cisplatin (DHAP) to assess non-inferiority of GDP in efficacy. Recent primary analysis of this trial showed that GDP arm was non-inferior to DHAP arm in term of objective response rate. Patients on the GDP arm also experienced less adverse events (toxicity) and reported to have a better quality of life.

During the training period, the intern will be able to work on the analysis of LY.12 database in the following three topics:

1) Health economic analysis: to study the resource utilization associated with GDP and DHAP treatments. The resource utilization includes drug and administration cost, hospitalization cost, emergency room cost, inpatient costs, and outpatient costs etc.

2) Tissue micro array (TMA) analysis: to investigate how the response rate related to patient’s genetic profile.

3) Subset analyses for patients in different age group (age>65 vs. age <= 65) and patients with different histology type (Transformed lymphoma (TRS) vs. Denovo diffused large B cell lymphoma (DLBCL)): to detect any potential different outcomes (response rate, overall survival) among different subset of patients.

Dr. George Rodrigues, Clinician Scientist at London Health Sciences Center,
Associate Professor at the Department of Epidemiology and Biostatistics, Western University

Location:

London Regional Cancer Program, London

Research Plan:

The Genitourinary Radiation Oncologists of Canada (GUROC) published a consensus-based three-group (i.e. low-, intermediate-, and high-risk) risk stratification system for the management of prostate cancer in 2001. Since the publication of this consensus statement, an expanding literature on classical and novel prognostic factors, risk stratification schema, and other pre- or post-treatment models have been published. In 2009, the GUROC commissioned a critical review of the evidence related to novel prognostic factor and stratification approaches prior to final discussions regarding any alteration of the GUROC risk stratification scheme. This review affirmed that the existing GUROC classification system should be modified utilizing an evidence-based consensus approach, consistent with common treatment management practices, coherent with the current classification system, and employing available statistically robust prognostic factors. Specifically, the assessment of sev eral specific changes to the stratification system were recommended including: the creation of a very low-risk category, division of the intermediate-risk into two distinct groups, reassessment of the interface between intermediate- and high-risk, and the integration of novel prognostic factors (positive core percentage, Gleason 3+4 vs. 4+3) into the stratification system.

The Prostate Cancer Risk Stratification (ProCaRS) radiotherapy database (n=7974) was commissioned by GUROC to support the investigation of hypothesis related to prostate cancer risk stratification optimization. An initial report has detailed the American Society of Radiation Oncology (ASTRO) biochemical failure-free survival (BFFS) and overall survival (OS) outcomes of the ProCaRS database in terms of different radiotherapy subgroups (e.g. external beam vs. brachytherapy). A second report has investigated on the definition of a new GUROC ProCaRS risk stratification system based on a recursive partitioning analysis of available pre-treatment prognostic factors to predict for ASTRO BFFS. Ongoing projects using the ProCaRS database includes the prediction of outcomes using neural networks as well a propensity score matched-pair analysis of brachytherapy versus external-beam radiation therapy.

The project associated with the Oncology Research and Methods Training Program placement will focus on two areas:

1. Investigation of the relationship between androgen ablation utilization with clinical outcome.

2. Investigation of the relationship of radiotherapy dose with clinical outcome.

For more information, click here …

Dr. Paul C. Boutros, Principal Investigator at Informatics & Biocomputing
Ontario Institute for Cancer Research

Collaborators:

Multiple collaborators

Location:

Ontario Institute for Cancer Research, Toronto

Research Plan:

Glioblastoma multiforme (GBM), Ovarian and Colorectal cancers are amongst the most malignant cancers types. In 2010, an estimated 22,000, 21,888 and 142,270 patients were diagnosed with GBM, ovarian and colorectal cancer, respectively. There exist a number of molecular prognostic markers for these cancer types, yet the mortality rate remains very high. For instance, the GBM patients usually survive less than 15 months. The studies focused on deriving prognostic markers of clinical relevance are yet to evaluate the combined impact of multiple somatic mutations as a group, in particular the ones that are less prevalent. The cancer genome atlas (TCGA) project has produced a comprehensive catalogue of somatic mutations across these cancer types. These mutations are further characterized into respective consequence type depending upon the type of genomic alteration they cause, for instance insertion, deletion and inversions. However, these alterations eventually translate into a protein and therefore needs to be further characterized in terms of impact in the corresponding functional domain of the protein. This way, the deregulation impact of somatic mutations can be assessed in a functional context. The functional assessment would not only offer the per‐protein domain dysregulation across various binding sites, it also lends itself towards quantifying the impact of somatic mutations across various subunits of a protein. Following this, we can treat the protein domain‐wise deregulation scores as pseudo‐features to establish a multivariate statistical model to predict patient outcome. The project would involve collection of TCGA datasets, UCSC genomic coordinates of Human genome assembly vHg19 and protein domain coordinates. The intern would map the somatic mutations from genomic to functional coordinates. This would be followed by establishing a novel features selection approach to translate the domain‐wise mutational profiles into impact scores, and subsequent assessment of these scores by a multivariate statistical model. As an end product, we envisage to establish a generic framework that would facilitate integration of multiple somatic mutations and their prognostic evaluation.

The intern will be treated as a key member of the cancer bioinformatics research team. They will participate in the weekly team meetings, and interact on a regular basis with clinicians, molecular researchers, bioinformaticians, and biostatisticians. Physically, they will be located within the Boutros Lab at OICR, which is a team of biologists, statisticians, engineers, and computer scientists working on techniques for biomarker discovery. They will be able to interact with Dr. Boutros on a regular basis, and will be expected to attend the weekly lab meetings, institutional seminars and journal clubs. OICR has a strong computing facility, including a >5000 core computing cluster and an svn repository for code management. Students will be treated as full members of the lab, and will have the opportunity to present their results in lab‐meetings and, if appropriate, to draft their own manuscripts for publication.

Dr. Paul C. Boutros, Principal Investigator at Informatics & Biocomputing, Ontario Institute for Cancer Research

Collaborators:

Dr. Ben Neel, Professor at Department of Medical Biophysics, University of Toronto, and Senior Scientist at Ontario Cancer Institute

Dr. Bradly Wouters, Professor at Department of Medical Biophysics, University of Toronto, and Senior Scientist at Ontario Cancer Institute

Location:

Ontario Institute for Cancer Research, Toronto

Research Plan:

Current cancer drug development typically targets oncogenes and their signalling pathways, based on the premise that tumours become “addicted” to these genes/pathways. Due to these mutations, tumours can also become unusually dependent on pathways that are not directly affected by mutation; we refer to this concept as emergent synthetic lethality. Recent evidence indicates that drugs exploiting synthetic lethality can be highly potent and selective; e.g., PARP inhibitors kill cancer cells with mutations in the BRCA1/2 genes. Yet there has been no systematic, assessment of synthetic lethality relationships in relevant human tumour models.

The Ontario Research Fund has recently funded us to undertake a large, systematic study of this question. In particular, we are sequencing the entire genomes of 200 xenografts – these are primary human cancers (ovarian, colon, lung, pancreas) that have been implanted into immunodeficient mice. For each tumour a diverse array of molecular data is being collected, including exome sequencing, transcriptome sequencing, and genome‐wide copy‐number variation. Two major challenges in analyzing this data will be: 1) the presence of contaminating sequence (noise) from the host mouse genome and 2) the large number of rare‐frequency events. These will be the topics focused on by the intern.

The intern student can be involved in both of these sub‐projects (contaminating noise and rare‐frequency events), but will focus on one of them. To overcome the presence of contaminating noise, we will employ a variety of deconvolution techniques. For example, one approach would be to do a competitive alignment process, attaching a likelihood to each data‐point and using these to determine the weight of evidence for a specific mutation.

To tackle the rare‐frequency event problem may require Bayesian techniques, along with data‐reduction. In particular, if we hypothesize that there is allelic heterogeneity (e.g. as for BRCA1), then individual loci will have insufficient power so nested modeling will be required to collapse across regions or functional groupings. Our proposed plan is to aggregate the rare variants in each protein‐coding gene as well as non‐coding genomic area to improve the statistical power of detection. For each type of variation, we implement methods to assess the functional impact of variations (e.g. a score) on the protein function (based on typical indicators such as evolutionary conservation of amino‐acid residues). We use these scores along with genomic arrangements of each individual as inputs for GWA analyses. For GWA analysis we use statistical signal processing methods to pinpoint and refine the impact of rare events in the noise created by common variants. We propose to investigate methods such as recursive least square filtering, principle component analysis and latent‐variable dimensionality reduction algorithms. We will finally improve our detection capabilities by incorporating information from pathways known to contribute to prostate cancer. This results in a network‐based approach to identify collective contribution of multiple biomarkers in cancer. These are exciting challenges in modern biostatistics, and will provide the intern with exposure to cutting edge techniques.

The intern will be treated as a key member of the ORF/GL2 research team. They will participate in the weekly team meetings, and interact on a regular basis with clinicians, molecular researchers, bioinformaticians, and biostatisticians. Physically, they will be located within the Boutros Lab at OICR, which is a team of biologists, statisticians, engineers, and computer scientists working on techniques for biomarker discovery. They will be able to interact with Dr. Boutros on a regular basis, and will be expected to attend the weekly lab meetings, institutional seminars and journal clubs. OICR has a strong computing facility, including a >5000 core computing cluster and an svn repository for code management. Students will be treated as full members of the lab, and will have the opportunity to present their results in lab‐meetings and, if appropriate, to draft their own manuscripts for publication.

Dr. Paul C. Boutros, Principal Investigator at Informatics & Biocomputing, Ontario Institute for Cancer Research

Collaborators:

Dr. Robert Bristow, Professor at Department of Radiation Oncology, University of Toronto, Staff Radiation Oncologist at Princess Margaret Hospital/University Health Network, Senior Scientist at Ontario Cancer Institute

Dr. Theo can der Kwast, Department of Laboratory Medicine and Pathology, University of Toronto, Staff Pathologist at University Health Network

Location:

Ontario Institute for Cancer Research, Toronto

Research Plan:

Prostate cancer (CaP) is the most commonly diagnosed malignancy in Canadian men. In 2011, it is predicted that almost 25,000 men will be diagnosed with CaP, and more than 4,000 are expected to die from the disease. It is now the third most common cause of cancer death after lung and colorectal cancer in males. For example, in western populations, 1 in 6 men are diagnosed with prostate cancer while 1 in 34 die of metastatic disease. Treatment options for CaP depend on the TNM-staging of the disease. Using the prognostic variables of T-category, the serum prostate specific antigen (PSA), and the pathologic Gleason score (GS), men with localized prostate cancer are placed in low, intermediate and high-risk groupings. Yet within each of these groups, some men have good outcome with current therapies, while others will die of aggressive metastatic disease.

Understanding the basis of clinical heterogeneity in outcome is of fundamental importance. If detected while still confined to the organ, prostate cancer can often be cured by either surgery (radical prostatectomy) or radiotherapy. The introduction of large scale prostate-specific antigen (PSA) testing for prostate cancer has led to a two-fold increase in diagnosis of prostate cancer in North America. Of these, it is estimated that only half would have led to morbidity and mortality if left undetected. To tackle this challenge, Prostate Cancer Canada and OICR have funded The Canadian Prostate Cancer Genome Network (CPC-GENE). CPC-GENE is a 5-year project to sequence the complete genomes of 500 prostate cancers, with a focus on improving the accuracy of risk-grouping of prostate cancer.

One key aspect to this project will be validation of results in independent patient cohorts. As a result, there is a position for an intern student focused on the meta-analysis of multiple studies of prostate cancer genomic data. Each existing study has looked at the status of up to 20,000 genes, along with a series of clinical covariates, using a variety of different technical platforms. The intern will be focused on developing appropriate techniques for integrating these studies, including standardization of data across different platforms, estimating and adjusting for study-specific effects, handling varying levels of missing data, estimating heterogeneity, and using permutation techniques to identify bias and influential studies. The successful completion of this project will not only develop a key resource for CPC-GENE, but will also generate a methodology useful to a large number of on-going cancer genome-sequencing projects.

The intern will be treated as a key member of the CPC-GENE research team. They will participate in the weekly team meetings, and interact on a regular basis with clinicians, molecular researchers, bioinformaticians, and biostatisticians. Physically, they will be located within the Boutros Lab at OICR, which is a team of biologists, statisticians, engineers, and computer scientists working on techniques for biomarker discovery. They will be able to interact with Dr. Boutros on a regular basis, and will be expected to attend the weekly lab meetings, institutional seminars and journal clubs. OICR has a strong computing facility, including a >5000 core computing cluster and an svn repository for code management. Students will be treated as full members of the lab, and will have the opportunity to present their results in lab-meetings and, if appropriate, to draft their own manuscripts for publication.