Predictive modeling of brain cancer outcomes based on molecular profiling data

Rebecca Fuchs

Mentor: Dr. Yuriy Gusev, Innovation Center for Biomedical Informatics, Department of Oncology, Georgetown University

Date/Time: August 27, 2019 at 2pm

Location: Room 1300, Harris Building

Brain cancer is deadly, the incidence rate of brain cancer is 22.64 cases per 100,000 Americans every year. 5 year life expectancy varies with age of patient and progression of the disease, but averages to 35% mortality. In this project we utilized REMBRANT data set – a large collection of clinical and molecular profiling data as well as newly generated data on chromosomal instability index of more than 600 brain cancer patients. We used this data as an input for machine learning algorithm Random Forest with Recursive Feature Elimination (RF RFE). Using this approach we generated different predictive models of survival for male and female patients with brain cancer and determined most informative molecular features based on importance score. Top ranked genes and chromosomal regions called cytobands were further analysed using systems biology tools to determine main biological pathways affected by changes in genes expression and copy number variation. The results shows that it is feasible to build predictive models with nearly 80% accuracy based on expression data of small subsets of 20-40 top ranked expressed genes or 10-20 genomically unstable cytobands.