Behnam Reihani Matanagh - Department of Computer Scinces, Faculty of Mathematics, Shahid Beheshti University, Tehran, Iran
Soheil Jahangiri Tazehkana - Princess Margaret Cancer Research Center, UHN, University of Toronto, Toronto, Canada
Changiz Eslahchi - School of Biological Sciences, Institute for Research in Fundamental Sciences (IPM), Tehran, Iran

In the past decade and a half, many proteomics data statistical analysis approaches have been proposed to identify proteins whose expression profile is useful for the diagnosis, prognosis-treatment, and deciphering cause of a disease [1]. Unfortunately, the problems with using these approaches are a large number of false positives, lacking reproducibility and stability, irrelevance selected features and batch effects [2]. Recently, there has been a paradigm shift from looking at individual proteins to preselected protein complexes. Protein complex-based feature selection (PCBFS) methods divide into two classes. The first -class contains methods that use primary protein expression data to select features .The second -class contains methods that work on normalized expression data [3,4]. We present powerful method, method, CSNet ,that uses a combination of both expression data and normalized data to calculate the complexes score .CSNet is particularly resistant against batch effects and selects features strongly correlated with class but not batch.