Title: "Helping to cure cancer using topic models."
Speaker: Quaid Morris, Dept. Computer Science, University of Toronto
Location: D.L. Pratt, 290 C
I will talk about using topic models to analyze gene expression profiles from tumour samples. A gene expression profile is an integer-valued vector; the elements of this vector are counts of the number of copies of the corresponding genes' messenger RNAs (mRNAs) present in a sample. These profiles provide a snapshot of the biological state of the cells in the tumour sample and have proved valuable in helping to diagnose the tumor and predict patient prognosis. Because tumor samples are mixtures of different types of tissue, we use topic models to estimate the relative contributions of
each tissue type to the expression profile as well as inferring the distribution of mRNAs within cancerous cells. Our algorithms are called ISOLATE / ISOpure and we've been able to use our methods to improve prognostic accuracy by as much as 75% for lung cancer and prostate cancer by providing a better estimate of the "topic" distribution for cancerous cells.
For additional Information, Contact:
Name: Danny Tarlow