Research Abstracts Online
2008 - March 2009
University of Minnesota Twin Cities
School of Public Health
Division of Biostatistics
PI: Wei Pan
Network-based Support Vector Machine for Classification and Solution Path Algorithm
The proliferation of microarray data and the extensive research on extracting genetic pathways make it possible to develop new methods that attempt to achieve higher accuracy and improve variable selection by integrating biological information into model fitting. By extending the concept of grouped variable selection in the context of microarrays, these researchers have investigated the neighboring-gene support vector machine (NG-SVM) for binary classification, and extended this to include an attempt to capture ultimate disease-causing genes carrying weak effects and to recover the regulatory pathways.
This project investigates a disease-gene-centric SVM that encourages recovering the regulatory pathways and identifying the center gene of the network. The tuning parameter must be carefully selected because it determines the performance of regularization methods on future data. To facilitate the selection of the tuning parameter, the researchers are exploring the possibility of applying the existing solution path algorithms in the cases of SVM with the proposed penalty terms.
Benhuai Xie, Graduate Student
Yanni Zhu, Graduate Student