Augmented sequence features and subcellular localization for functional characterization of unknown protein sequences

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput techniques generate numerous unknown protein sequences (UPS) requiring functional characterization.
• Protein subcellular localization is crucial for understanding protein function, disease mechanisms, and drug repositioning.
• Existing feature extraction methods often yield suboptimal prediction performance for protein sequences.

Purpose of the Study:

• To develop and evaluate two novel feature augmentation strategies for protein sequences.
• To enhance the prediction accuracy of protein subcellular localization, particularly for unknown protein sequences.
• To compare the performance of individual and ensemble machine learning models using augmented features.

Main Methods:

• Two feature augmentation techniques were developed, integrating sequence-induced, physicochemical, and evolutionary information.
• Augmented features were designed to retain sequence-order information and protein-residue properties.
• Bacterial Gram-Positive (G+) and Gram-Negative (G-) datasets were preprocessed, and feature vectors were generated for model training and evaluation.

Main Results:

• The Decision Tree (C4.5) algorithm achieved the highest accuracy (99.57% on G+ and 97.47% on G-) for known protein sequences.
• For unknown protein sequences (UPS), Support Vector Machine (SVM) yielded 85.17% accuracy on G+ datasets, and Multi-Layer Perceptron (MLP) achieved 82.45% on G- datasets.
• Ensemble methods and individual classifiers demonstrated improved performance with the augmented feature sets.

Conclusions:

• The proposed feature augmentation methods effectively improve protein subcellular localization prediction accuracy.
• The integrated features provide a robust approach for characterizing both known and unknown protein sequences.
• This work offers valuable tools for advancing functional genomics and drug discovery through improved protein analysis.