Practicum

Microarray data identification, annotation, and quality assessment In this practicum, students will identify appropriate breast cancer data sets, determine the nature and structure of these datasets, download and install local copies, assess the quality of the data and its annotation, and prepare consistently annotated and normalized data for subsequent analysis by others.

The data include information on (i) clinical study design (e.g., inclusion/exclusion criteria, comparability of patient populations across data sets, key clinical and associated data required for later analysis, HIPAA implications for access to clinical data); (ii) microarray experimental design, (e.g., different microarray platforms, general MIAME requirements, (iii) data quality and quantity of clinical and microarray data (e.g., are the data adequately annotated, is the correct data available to allow analysis, are there missing data and how might these preclude or limit the use of a data set). Students will learn the key clinical characteristics of breast cancer in the context of its biology, histopathology, diagnosis, and clinical course as needed to understand the nature, quality and utility of the clinical data collected, what these represent to oncologists, and which clinical data are likely to be useful for later data analysis. The end-product will be the collection and annotation of breast cancer data sets that can be used by others to extract other prognostic signatures and/or to explore the possible use of individual genes as independent biomarkers. It is likely that the students will be given a list of individual biomarkers of interest to explore towards the end of their practicum.

Building and testing probabilistic models of breast cancer prognosis In this practicum, students will explore existing (publicly available) breast cancer data sets to find gene expression patterns associated with prognosis. Initial datasets are already available and downloaded, other data sets will eventually be available from the efforts of students working in practicum #1.

The goal will be to use one data set as a training set to find differentially expressed genes that can be used to build an in silico predictor of breast cancer prognosis. These genes will then be used as the input to several statistical models to train a predictor of prognosis (model output will be an appropriate measure of prognosis, e.g., disease recurrence or overall survival). Using one or more independent data sets also publicly available), the performance of the predictors will be evaluated and compared with the performance of known classifiers, e.g., MammaPrint. Performance will include ROC, NPV and PPV measures and the generation and interpretation of standard Kaplan-Meier plots. For data interpretation ad for model building, students will learn the key clinical characteristics of breast cancer in the context of its biology, histopathology, diagnosis, and clinical course as needed to understand the nature, quality and utility of the prognostic model.