Detection of Interactions between Proteins through Rotation Forest and Local Phase Quantization Descriptors

Leon Wong, Zhu‐Hong You, Zhong Ming, Jianqiang Li, Xing Chen, Yu‐An HuangInternational Journal of Molecular Sciences 2015 · DOI

TL;DR

This paper introduces a novel computational method for predicting protein-protein interactions (PPIs) using only protein sequences. By combining a Physicochemical Property Response Matrix with Local Phase Quantization for feature extraction and a Rotation Forest classifier, the method achieves significantly higher accuracy than previous approaches, offering a more efficient and cost-effective alternative to experimental methods.

What Problem Does This Paper Address?

Understanding how proteins interact (Protein-Protein Interactions or PPIs) is crucial for deciphering complex cellular processes, disease mechanisms, and developing new drugs. While high-throughput experimental methods exist for identifying PPIs, they are often expensive, time-consuming, and can yield false positives or negatives. This creates a need for accurate and efficient computational methods that can predict these interactions directly from protein sequence data, without requiring prior experimental knowledge.

Methods

The core of this work lies in a novel sequence-based approach for PPI prediction. The authors first represent each protein sequence as a Physicochemical Property Response Matrix (PR Matrix). This matrix captures the physicochemical properties of amino acids within the sequence. To extract meaningful features from this matrix, they employ the Local Phase Quantization (LPQ) method, a technique borrowed from image processing known for its robustness to blur. LPQ analyzes the local phase information in the Fourier transform of the PR Matrix. Finally, these extracted features are fed into a Rotation Forest (RF) classifier, an ensemble learning method known for its high classification performance. The RF classifier is trained to distinguish between interacting and non-interacting protein pairs.

Datasets

The proposed method was evaluated on three publicly available PPI datasets:

  • Saccharomyces cerevisiae (yeast)
  • Homo sapiens (human)
  • Helicobacter pylori (bacteria)

These datasets were preprocessed to remove redundant protein pairs and those with high sequence identity. The authors also compared their method against existing approaches using these established benchmarks.

Key Insights and Ideas

  • Leveraging Image Processing for Biological Data: The successful application of LPQ, a technique from image texture analysis, to biological sequence data highlights the potential for cross-disciplinary innovation in bioinformatics.
  • Sequence-Based Prediction is Viable and Powerful: The study demonstrates that high accuracy PPI prediction can be achieved using only protein sequence information, reducing reliance on complex experimental data or prior knowledge.
  • Ensemble Methods Excel: The use of Rotation Forest, an ensemble classifier, significantly boosts prediction accuracy compared to single classifiers, suggesting that combining multiple decision trees is beneficial for this task.
  • PR Matrix Captures Crucial Information: The PR Matrix representation, by encoding physicochemical properties, appears to capture essential information for distinguishing interacting proteins that might be missed by simpler sequence representations.

Key Results

The proposed method demonstrated strong performance across all three datasets:

  • Saccharomyces cerevisiae: Achieved an average accuracy of 93.8%.
  • Homo sapiens: Achieved an impressive average accuracy of 97.96%.
  • Helicobacter pylori: Achieved an average accuracy of 89.47%.

These accuracies were significantly higher than those reported by previous state-of-the-art methods, including a comparison with the Support Vector Machine (SVM) on the human dataset, where the proposed method showed substantial improvements in accuracy, precision, sensitivity, and MCC.

Limitations and Future Work

  • Sequence Length Constraint: The paper mentions that sequences exceeding a certain length (larger than 250x250 for the PR matrix) might be ignored due to computational limitations. This could be a bottleneck for predicting interactions involving very long proteins.
  • Complementary Role: The authors suggest that their method can serve as a complementary tool to experimental proteomics research, implying it's not a complete replacement but a valuable aid.

Future work could focus on addressing the sequence length limitation and exploring further optimizations of the feature extraction and classification stages.

Summary updated

Abstract

Protein-Protein Interactions (PPIs) play a vital role in most cellular processes. Although many efforts have been devoted to detecting protein interactions by high-throughput experiments, these methods are obviously expensive and tedious. Targeting these inevitable disadvantages, this study develops a novel computational method to predict PPIs using information on protein sequences, which is highly efficient and accurate. The improvement mainly comes from the use of the Rotation Forest (RF) classifier and the Local Phase Quantization (LPQ) descriptor from the Physicochemical Property Response (PR) Matrix of protein amino acids. When performed on three PPI datasets including Saccharomyces cerevisiae, Homo sapiens, and Helicobacter pylori, we obtained good results of average accuracies of 93.8%, 97.96%, and 89.47%, which are much better than in previous studies. Extensive validations have also been explored to evaluate the performance of the Rotation Forest ensemble classifier with the state-of-the-art Support Vector Machine classifier. These promising results indicate that the proposed method might play a complementary role for future proteomics research.
BibTeX
@article{wong2015detection,
  author = {Leon Wong and Zhu‐Hong You and Zhong Ming and Jianqiang Li and Xing Chen and Yu‐An Huang},
  title = {{Detection of Interactions between Proteins through Rotation Forest and Local Phase Quantization Descriptors}},
  journal = {International Journal of Molecular Sciences},
  year = {2015},
  doi = {10.3390/ijms17010021},
}
Figure