Brief communication: Effect of example weights on prediction of protein-protein interactions

  • Authors:

  • Ming-Hui Li;Xiao-Long Wang;Lei Lin;Tao Liu

  • Affiliations:

  • Bioinformatics Research Group, ITNLP Lab, Department of Computer Science and Technology, Harbin Institute of Technology, 92 West Da Zhi Street, Harbin 150001, China;Bioinformatics Research Group, ITNLP Lab, Department of Computer Science and Technology, Harbin Institute of Technology, 92 West Da Zhi Street, Harbin 150001, China;Bioinformatics Research Group, ITNLP Lab, Department of Computer Science and Technology, Harbin Institute of Technology, 92 West Da Zhi Street, Harbin 150001, China;Bioinformatics Research Group, ITNLP Lab, Department of Computer Science and Technology, Harbin Institute of Technology, 92 West Da Zhi Street, Harbin 150001, China

  • Venue:
  • Computational Biology and Chemistry
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

Protein-protein interactions (PPIs) prediction is an important issue in biology. Recently many computational methods have been proposed to determine PPIs. However, there is no golden standard dataset for these methods now. Furthermore, there exists different quality among training examples and the quality is always ignored by the current methods. In the condition of low-quality examples, the system should tolerate the data noise. Example weighting strategy is used in this paper to build a robust system and solve the problem of data noise. Training examples are investigated and a new example selecting/using strategy is proposed. Training example weighting method based on confidence is proposed. Different weight setting strategies are discussed and the corresponding results are given in the experiment. A new model integrating example weighting strategy, attraction-repulsion (AR) weight model, is proposed. Experimental results on Saccharomyces cerevisiae demonstrate that the new model outperforms the original AR model in the ROC score measure by over 8%. Furthermore, the example weighting strategy is applied to another domain-based PPIs prediction method, maximum likelihood estimation (MLE) method, and the modified MLE method obtains better performance than the original MLE method. At same time, our examples weighting strategy can be applied to any other training example based PPIs prediction methods.