An empirical study of different approaches for protein classification

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Automatic protein classification is crucial for various scientific domains.
• Current feature extraction methods lack generalizability across different protein datasets.
• There is a need for robust and versatile protein representation techniques.

Purpose of the Study:

• To evaluate and compare multiple protein feature extraction approaches.
• To assess the generalizability of different protein descriptors.
• To develop a fused descriptor system for improved protein classification.

Main Methods:

• Utilized protein representations derived from Position-Specific Scoring Matrices (PSSM), amino-acid sequences, and 3D tertiary structures.
• Developed and tested novel protein descriptor variants.
• Employed Support Vector Machines (SVMs) for classification, combining results using a sum rule fusion strategy.

Main Results:

• Individual descriptors showed variable performance across different datasets.
• The fused descriptor system demonstrated robust and high performance across all tested datasets.
• The combined approach achieved superior results compared to several state-of-the-art methods.

Conclusions:

• Feature fusion is an effective strategy for enhancing the generalizability and performance of automatic protein classification systems.
• The developed system offers a reliable and efficient solution for protein classification across diverse biological data.
• This work advances the field of bioinformatics by providing a more versatile approach to protein representation.