RFAmyloid: A Web Server for Predicting Amyloid Proteins

Mengting Niu, Yanjuan Li, Chunyu Wang, Ke HanInternational Journal of Molecular Sciences 2018 · DOI

TL;DR

This paper introduces RFAmyloid, a novel web server and predictor for identifying amyloid proteins. By combining advanced feature extraction techniques with a random forest classifier, RFAmyloid achieves high accuracy in distinguishing amyloid proteins, which is crucial for understanding and diagnosing amyloid-related diseases like Alzheimer's.

What Problem Does This Paper Address?

Amyloid proteins are implicated in a range of debilitating diseases, including Alzheimer's and Creutzfeldt-Jakob's disease, due to their tendency to mis-aggregate into insoluble fibrils. Early and accurate identification of these proteins is essential for disease research, diagnosis, and the development of therapeutic strategies. However, precisely identifying amyloidogenic regions within protein sequences remains a significant challenge. Existing methods often rely on specific models or limited feature sets, necessitating the development of more robust and comprehensive prediction tools.

Methods

The core of RFAmyloid lies in its sophisticated feature extraction and classification approach. The authors employed a multi-feature representation method, combining two key techniques:

  • SVMProt 188-D feature extraction: This method leverages protein composition and physicochemical properties. It extracts a total of 188 dimensions by considering various aspects like amino acid composition, hydrophobicity, polarity, and solvent accessibility, among others.
  • Pse-in-one feature extraction: This method incorporates a broader range of features, including k-mers, distance-based residue information, autocorrelation features (AC, CC, ACC, PDT), pseudo amino acid composition (PC-PseAAC, SC-PseAAC), profile-based features, and predicted structure features.

These extracted features are then fed into a Random Forest (RF) classifier. Random forests are an ensemble learning method that builds multiple decision trees and uses a voting mechanism to improve prediction accuracy and robustness. The authors specifically utilized a ten-fold cross-validation strategy to evaluate the model's performance.

Datasets

The study utilized a self-built dataset named Amy. The construction of this dataset involved:

  1. Data Acquisition: Protein sequences were sourced from the Universal Protein (UniProt) and AmyPro databases.
  2. Filtering: Sequences shorter than 50 amino acids were removed.
  3. De-redundancy: The CD-HIT program was used to cluster proteins and eliminate redundancy and homology biases, ensuring a more reliable dataset.

The final Amy dataset comprised 165 amyloid proteins and 382 non-amyloid proteins. This dataset is available for download from the RFAmyloid server.

Key Insights and Ideas

  • Feature Fusion is Key: The paper strongly suggests that combining diverse feature representations (like the 188-D and Pse-in-one methods) significantly enhances prediction accuracy compared to using individual feature sets. This highlights the complexity of amyloid protein identification, requiring a multi-faceted approach.
  • Random Forest's Efficacy: The study reaffirms the power of Random Forests as a robust classifier for biological sequence prediction tasks, demonstrating its ability to handle complex patterns and achieve high performance.
  • The Importance of a Dedicated Server: The development of the RFAmyloid web server makes this powerful prediction tool accessible to the broader research community, facilitating further investigation into amyloid proteins and related diseases.
  • Balanced vs. Unbalanced Datasets: While not the primary focus, the paper touches upon the impact of dataset balance, noting that while an unbalanced dataset yielded slightly higher overall accuracy, a balanced dataset improved sensitivity, underscoring the need for careful dataset preparation.

Key Results

The RFAmyloid predictor demonstrated strong performance across various evaluation metrics:

  • Ten-fold Cross-Validation:
    • Accuracy (ACC): 89.19%
    • F-measure: 0.891
    • MCC: 0.739, SE: 0.781, SP: 0.927
  • Independent Test Set (External Validation):
    • Accuracy (ACC): 89.71%
    • F-measure: 0.897
    • MCC: 0.757, SE: 0.818, SP: 0.932

These results were consistently superior to other feature representation methods and classifiers tested, including Naive Bayes, SGD, and various ensemble methods. The RFAmyloid web server is available at: http://server.malab.cn/RFAmyloid/.

Limitations and Future Work

The authors acknowledge that while their model shows high accuracy, there is always room for improvement. Future work suggested includes:

  • Optimizing Feature Extraction: Exploring more advanced feature extraction algorithms to further enhance prediction performance.
  • Improving Classification Algorithms: Investigating the use of integrated classifiers, which combine multiple classifiers, to potentially boost classification accuracy.
  • Expanding Datasets: While the Amy dataset is self-built, incorporating larger and more diverse datasets could further validate and refine the model.

Summary updated

Abstract

Amyloid is an insoluble fibrous protein and its mis-aggregation can lead to some diseases, such as Alzheimer's disease and Creutzfeldt⁻Jakob's disease. Therefore, the identification of amyloid is essential for the discovery and understanding of disease. We established a novel predictor called RFAmy based on random forest to identify amyloid, and it employed SVMProt 188-D feature extraction method based on protein composition and physicochemical properties and pse-in-one feature extraction method based on amino acid composition, autocorrelation pseudo acid composition, profile-based features and predicted structures features. In the ten-fold cross-validation test, RFAmy's overall accuracy was 89.19% and F-measure was 0.891. Results were obtained by comparison experiments with other feature, classifiers, and existing methods. This shows the effectiveness of RFAmy in predicting amyloid protein. The RFAmy proposed in this paper can be accessed through the URL http://server.malab.cn/RFAmyloid/.
BibTeX
@article{niu2018rfamyloid,
  author = {Mengting Niu and Yanjuan Li and Chunyu Wang and Ke Han},
  title = {{RFAmyloid: A Web Server for Predicting Amyloid Proteins}},
  journal = {International Journal of Molecular Sciences},
  year = {2018},
  doi = {10.3390/ijms19072071},
}
Figure