Biological network clustering by robust NMF
- Additional Document Info
- View All
Copyright © 2014 ACM. We propose a Robust Non-negative Matrix Factorization (RNMF) formulation by introducing L1-norm regularization terms for decomposed factors to cluster noisy biological net- works for identification of functional modules. To solve ro- bust NMF, we develop an accelerated alternative proximal method, which takes advantages of a fast iterative shrinkage- Thresholding strategy to update each factorized component at each step. We compare the performance of this acceler- Ated proximal method with a multiplicative algorithm and a general proximal method for the same RNMF formulation. Experiments on synthetic networks and Protein-Protein In- Teraction (PPI) networks demonstrate that the accelerated proximal method is superior to the other algorithms in terms of efficiency and effectiveness for functional module identification.
author list (cited authors)