Department of Computer Engineering
CS 590 SEMINAR
Predicting Survival Rates of Cancer Patients Based on Partial Orders Features
Computer Engineering Department
Predicting the survival of the patient is critical for choosing better treatment strategies and is traditionally based on clinical variables such as age and tumor stage. So far the efforts to identify novel prognostic factors have focused on molecular markers via high throughput molecular profiling of large tumor sets. In most of these approaches, models rely on the individual expression quantities of the molecules in the tumors. However, cancer is a complex disease where molecular mechanisms are dysregulated in various ways. In this study, based on a system level perspective, we incorporate interactions within and between different molecules (mRNA, protein, miRNA) in the tumor. We incorporate partial ordering of the molecules in lieu of individual expression values. Our predictive models are based on the multivariate Cox proportional hazards model with L1 penalized log partial likelihood (LASSO). Testing on different cancer types we demonstrate that partial order representation significantly outperforms the individual quantity based features. Our strategy unveils predictive features with direct relevance to the biological mechanism of cancer.
DATE: 09 November, 2015, Monday @ 16:15