Variable selection from a feature representing protein sequences: a case of classification on bacterial type IV...

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Protein classification is vital for understanding biological functions.
• Encoding protein sequences is key for feature extraction in computational methods.
• Supervised learning for protein prediction is limited by the lack of experimentally validated labeled datasets, necessitating unsupervised approaches.

Purpose of the Study:

• To propose a novel method for variable selection from encoded protein sequences.
• To evaluate the effectiveness of different encoding approaches and variable selection techniques in protein classification.
• To address the challenge of identifying discriminative features for protein function prediction.

Main Methods:

• Developed a new variable selection technique for encoded protein sequences.
• Utilized a benchmark dataset of 1947 protein sequences, including 399 bacterial type IV secreted effectors (T4SE) and 1548 non-T4SE.
• Employed position-specific scoring matrices (PSSM) as a component of the encoded feature.

Main Results:

• Demonstrated that specific components of the encoded feature, such as PSSM, are effective for discrimination.
• Achieved comparable and quantifiable results using selected variables, highlighting the method's efficacy.
• Showcased the successful identification of bacterial type IV secreted effectors (T4SE) using the proposed approach.

Conclusions:

• Selected variables, independent of their original encoded feature, significantly contribute to protein type discrimination.
• Ensemble classifiers with automatic base classifier assignment enhance classification performance.
• The study underscores the importance of feature selection in improving protein classification accuracy.