Link Prediction Based on Extended Local Path Gain in Protein-Protein Interaction Network

Abstract

Protein–protein interaction (PPI) plays key role in each cellular process of any living cell, however, almost all organisms’ PPIs are still incomplete. In this study, we firstly proposed a computational method Extended Local Path (ELP), which estimated links’ existence likelihood by integrating all their neighbours’ local paths in the network. In addition, on this basis, we extended it to Extended Local Path Gain (ELPG), which estimated gain effect when adding or deleting one potential link to the network. Applying both ELPG and ELP methods and other four recognized outstanding methods on four public PPI data of Yeast, E. coli, Fruit fly and Mouse, we demonstrated that ELPG and ELP obtained better performance under two standard measures: area under curve (AUC) and Precision. Besides, ELP and ELPG were identified as the best features for classifying existing and unknown links by using support vector machine-recursive feature elimination (SVM-RFE) for feature selection.

Keywords

• Extended Local Path (ELP)
• Extended Local Path Gain (ELPG)
• Link prediction
• Protein–Protein Interaction (PPI)