A New Sequential Forward Feature Selection (SFFS) Algorithm for Mining Best Topological and Biological Features to...

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein-protein interaction networks (PPINs) are crucial for understanding cellular processes.
• Predicting protein complexes from PPINs is a complex challenge.
• Existing methods often overlook biological information within protein sequences.

Purpose of the Study:

• To develop an improved method for protein complex prediction.
• To integrate topological and sequence-based biological features.
• To enhance the accuracy of complex identification in PPINs.

Main Methods:

• Computed diverse topological and sequence-derived features (e.g., bag of words, physicochemical, spectral).
• Implemented a novel Sequential Forward Feature Selection (SFFS) using random forest-based Boruta.
• Employed decision tree, linear discriminant analysis, and gradient boosting classifiers.

Main Results:

• The proposed SFFS method combined with classifiers achieved high performance across multiple datasets (yeast, human, mouse).
• Achieved precision, recall, and F-measure rates up to 98.87% on the CORUM human dataset.
• Outperformed existing state-of-the-art algorithms in protein complex prediction.

Conclusions:

• The integration of topological and sequence-based features offers a powerful approach for complex prediction.
• The novel SFFS feature selection method effectively identifies relevant features for improved accuracy.
• This work advances the field of protein complex prediction, aiding biological process understanding.