SNFM: A semi-supervised NMF algorithm for detecting biological functional modules

Math Biosci Eng. 2019 Mar 7;16(4):1933-1948. doi: 10.3934/mbe.2019094.

Abstract

Unraveling protein functional modules from protein-protein interaction networks is a crucial step to better understand cellular mechanisms. In the past decades, numerous algorithms have been proposed to identify potential protein functional modules or complexes from protein-protein interaction (PPI) networks. Unfortunately, the number of PPIs is rather limited, and some interactions are false positive. Therefore, the algorithms that only utilize PPI networks may not obtain the expected results related to functional modules. In this study, we propose a novel semi-supervised functional module detection method based on non-negative matrix factorization(NMF)(SNFM), which incorporate high-quality supervised PPI links from complexes as prior information.Our method outperforms all the other competitors with improvements on performance by around 15.4% in Precision, 28.9% in Recall, 27.1% in F-score (on DIP data set) by using PCDq as gold standards.

Keywords: DIP; NMF; PPI; functional modules; semi-supervised.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Algorithms
  • Animals
  • Bayes Theorem
  • Biological Phenomena
  • Cluster Analysis
  • Computational Biology / methods*
  • Humans
  • Protein Interaction Maps*
  • Proteins / chemistry*

Substances

  • Proteins