Elsevier

Genomics

Volume 102, Issue 4, October 2013, Pages 237-242
Genomics

PPIevo: Protein–protein interaction prediction from PSSM based evolutionary information

https://doi.org/10.1016/j.ygeno.2013.05.006Get rights and content
Under an Elsevier user license
open archive

Highlights

  • We introduce a novel evolutionary feature extraction algorithm for PPI prediction.

  • Algorithm doesn't need protein annotation and is used for genome-wide PPI prediction.

  • PPIevo shows better performance and robustness compared with other algorithms.

  • Using negative dataset selection could lead to acute performance overestimation.

  • PPIevo is a standalone Java program which can be run on various operating systems.

Abstract

Protein–protein interactions regulate a variety of cellular processes. There is a great need for computational methods as a complement to experimental methods with which to predict protein interactions due to the existence of many limitations involved in experimental techniques. Here, we introduce a novel evolutionary based feature extraction algorithm for protein–protein interaction (PPI) prediction. The algorithm is called PPIevo and extracts the evolutionary feature from Position-Specific Scoring Matrix (PSSM) of protein with known sequence. The algorithm does not depend on the protein annotations, and the features are based on the evolutionary history of the proteins. This enables the algorithm to have more power for predicting protein–protein interaction than many sequence based algorithms. Results on the HPRD database show better performance and robustness of the proposed method. They also reveal that the negative dataset selection could lead to an acute performance overestimation which is the principal drawback of the available methods.

Keywords

Protein–protein interaction map
Protein interaction networks
Computational intelligence
Machine learning
Position-specific scoring matrix

Cited by (0)