J. Sunil Rao, Ph.D.
Professor of Public Health Sciences
Description of Research
Dr. Rao is the co-leader of the Cancer Control Research Program. His research is focused on developing statistical methods for understanding the genomic and epigenetic determinants of cancer progression. He has developed a set of methodologies called Bayesian ANOVA for uncovering signatures comprised of markers found individually as well as invisible fence methods for identifying sets of markers potentially acting along the same biological pathways. More recently, Dr. Rao has been interested in methods for the detection and characterization of extreme subgroups - those subgroups of individuals who might tend to do best on a given treatment. The technique is known as bump hunting. Colon cancer has been the main area of focus for these areas of research - in particular with Dr. Sanford Markowitz at Case Western Reserve University. A new collaboration in lung cancer has started up with the European Cancer Consortium and in particular Dr. Yudi Pawitan at the Karolinksa Institute in Stockholm.
Other important collaborations include with Dr. Michael Taylor at Sick Kids Hospital in Toronto where the genetic landscape of pediatric medulloblastoma was laid out, development of a computational algorithm for predicting cancer transcription factor binding sites (TFBS) with Dr. Carlos Moreno at Emory University, and development of the ColoSure diagnostic test (with Dr. Sanford Markowitz) for early detection of colon cancer by picking up in the stool, methylation of the vimentin gene which is typically not expressed in normal colon epithelial tissue. This test is now FDA approved. Most recently, Dr. Rao has focused his attention on developing statistical methods for dealing with high dimensional genomic studies that also contain missing data. The focus here is to be able to pick up true biological signal from noise in either GWAS studies or large scale expression studies. The technique is known as the E-MS algorithm, and can be shown to find the true set of informative markers with large enough samples sizes in the presence of missing data. Another area of focus has been in the development of robust prediction estimators for statistical models that are misspecified. This misspecification can be due to incomplete information or some other inherent bias in a study.
Dr. Rao developed the best predictive estimator methodology with Dr. Jiming Jiang at UC-Davis, and this is now becoming more widely studied in many areas of statistics. Statistical methods for generating optimal personalized predictions is a very hot topic. Ideally with so much genomic and other "omic" data available, the thought is that a truly optimal, personalized prediction should be available for each individual which could in theory, give higher probability for a more positive treatment outcome. How exactly to generate these predictions has been a mystery. Dr. Rao recently developed the idea of classified mixed prediction for tackling such a problem and its many derivatives. Finally, unravelling some of the determinants of cancer health disparities is something that careful statistical development can help shed some light on. Working together with Dr. Erin Kobetz of the University of Miami, Dr. Rao has developed a prism regression model which can show how individual level variables and community level variables can interact in complex ways to inform disparity in cancer outcomes between groups of interest. From this, new ideas of intervention aimed at eliminating disparities directly, can emerge.