|
Abstract:
|
Although it is well known that bulk tumor gene expression is a mixture profile of tumor and other cell types, until recently, bulk data is still widely used as representation of tumor gene expression. Recent advances in deconvolution methods have emphasized the importance of cell composition in bulk gene expression. However, two issues remain to be addressed: First, most deconvolution methods are unable to simultaneously estimate proportions of tumor and other cell types; Secondly, all deconvolution models generically assume that cell fraction is the only contributor to differential bulk gene expression, while in reality both cell-specific differential expression and proportional change can be the reason. In this talk, we present our work on addressing the above two problems. For the first issue, we developed a nonnegative matrix factorization(NMF)-based method that could simultaneously estimate tumor and other cell types’ proportions from bulk data. Secondly, we proposed a statistical framework to quantify contributions from differential expression and cell proportion change, respectively. We will demonstrate the capacities of our methods by simulations and results from real data.
|
