Integrative Statistical Modelling in Genetics and Genomics

点击翻译按钮获取中文摘要

Well-designed & carefully-analysed scientific studies yield stronger findings, preserve precious resources, and accelerate discovery of biological knowledge. The proposed research will develop new ways of statistical learning from family and patient data. It is motivated by large data collections in two cohorts of women with breast cancer. Susceptibility studies are concerned with inherited genetic variation & early age at breast cancer diagnosis. With the aim of discovering new susceptibility genes that account for the substantial unexplained component in familial breast cancer, we are analysing whole genome sequencing in early onset affected sisters selected from Ontario Familial Breast Cancer Registry (OFBCR) families; the OFBCR is part of large national & international consortia. We propose new methods for data analysis that take into account that siblings can inherit the same genetic mutation from one of their parents. The results will help design well-powered studies in international consortia to confirm candidate variants. Prognostic studies correlate molecular alterations in tumour tissues with variation in the natural history of disease. The axillary node-negative breast cancer study is a prospective cohort of newly diagnosed women followed for disease recurrence/death for 15 years. Arrays of tumour samples have been used to characterize the protein level of various biomarkers in each woman's tumour, but the patterns in the survival data we observed were inconsistent with the mathematical assumptions of conventional analysis of time to disease recurrence. A mixture-cure model that allows for a proportion of "cured" patients explains the observations much better. Although the arrays yield measurements of any one biomarker in most women, when we investigate multiple biomarkers together, some information is lost due to the accumulation of those with at least one missing value. Computational techniques known as "Multiple Imputation" which fill-in the incomplete data are a practical way to address this problem, but need to be specialized for the mixture-cure model.