Prediction of protein structural classes by recurrence quantification analysis based on chaos game representation

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Predicting protein structural classes is crucial for understanding protein function.
• Existing methods face challenges with low-homology datasets.
• Accurate classification requires robust feature representation.

Purpose of the Study:

• To develop a novel method for predicting protein structural classes.
• To apply recurrence quantification analysis (RQA) to protein sequence time series.
• To evaluate the method's performance on low-homology datasets.

Main Methods:

• Decomposition of protein chaos game representation into time series.
• Application of recurrence quantification analysis (RQA) for feature extraction (16 parameters).
• Classification using Fisher's linear discriminant algorithm.

Main Results:

• Achieved 65.8% and 64.2% accuracy on 1189 and 25PDB datasets, respectively.
• Outperformed existing methods by 6.3% and 4.1% in accuracy.
• Utilized only 16 characteristic parameters, fewer than other methods.

Conclusions:

• The RQA-based method is effective for predicting protein structural classes, especially for low-homology data.
• The method offers a complementary approach to existing techniques.
• The reduced feature set makes it computationally efficient and promising for future applications.