Enhanced Protein Structural Class Prediction Using Effective Feature Modeling and Ensemble of Classifiers.

Area of Science:

• Computational biology
• Bioinformatics
• Structural bioinformatics

Background:

• Protein Secondary Structural Class (PSSC) is crucial for understanding protein structure, function, and drug discovery.
• Experimental PSSC identification is laborious and expensive.
• Existing computational models for PSSC prediction often lack generalization.

Purpose of the Study:

• To develop an effective, novel, and generalized computational model for predicting PSSC from protein sequences.
• To improve the accuracy and robustness of PSSC prediction compared to state-of-the-art methods.

Main Methods:

• A hybrid feature engineering approach combining sequence embedding, SkipXGram bi-grams, and General Statistical (GS) features.
• An ensemble classification strategy integrating Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting Machines (GBM).
• Model validation on five diverse benchmark datasets (z277, z498, 25PDB, 1189, FC699) and a large-scale low-similarity dataset.

Main Results:

• The proposed model achieved high accuracies across benchmark datasets: 93.55% (z277), 97.58% (z498), 81.82% (25PDB), 81.11% (1189), and 93.93% (FC699).
• Achieved 81.11% accuracy on a large-scale updated low-similarity dataset (≤ 25%).
• Demonstrated consistent superior performance over existing state-of-the-art models.

Conclusions:

• The developed model offers an effective and generalized solution for PSSC prediction from protein sequences.
• The hybrid feature extraction and ensemble classification approach enhance prediction accuracy and robustness.
• This work contributes to advancing computational approaches for protein structure prediction and facilitates drug discovery efforts.