|
Abstract:
|
We developed an efficient Java based software tool that specializes in building and validating sets of molecular markers from whole genome array data. Comparing to other popular tools, our tool has two major advantages. First, instead of selecting markers using an arbitrary P value or FDR cutoff, our method finds the optimal set of markers based on cross validated model performance. Second, unlike other marker ranking mechanisms, which are always fixed tradeoffs between signal strength and signal consistency, our method can adapt to each individual dataset by finding an optimal tradeoff guided by cross validated model performance. Our tool has proven to be particularly effective for identifying amplified markers that are associated with a subset of cancers in a responding population. This technique is being effectively applied to most preclinical oncology assets at GSK.
|
