Self-evoluting framework of deep convolutional neural network for multilocus protein subcellular localization.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Proteomics

Background:

• Predicting protein subcellular localization is crucial for understanding protein function.
• Multilocus protein subcellular localization presents challenges due to feature correlation and classifier structure determination.
• Existing methods struggle with the complexity of multi-class classification for proteins with multiple locations.

Purpose of the Study:

• To propose a self-evoluting framework using deep convolutional neural networks (DCNNs) for multilocus protein subcellular localization.
• To address challenges in feature extraction and classifier design for complex protein localization.
• To enhance the accuracy of predicting multiple subcellular locations for proteins.

Main Methods:

• Integration of Ant Colony Optimization with DCNNs to create a self-evoluting algorithm.
• Random grouping of subcellular sites using a limited random k-labelsets (RAkEL) multi-label classification method.
• Random feature selection and feature fusion for robust protein sequence data preprocessing.

Main Results:

• The proposed self-evoluting DCNN framework achieved an overall correct rate of 67.17% on a human database.
• This accuracy surpasses established methods including stacked self-encoder (SAE), support vector machine (SVM), random forest classifier (RF), and single DCNN.
• The algorithm effectively handles complex multi-label classification through a divide-and-conquer approach and flexible expansion model.

Conclusions:

• The developed self-evoluting framework offers a significant advancement in multilocus protein subcellular localization.
• Combining ant colony optimization with DCNNs provides a powerful approach for optimizing complex biological predictions.
• The method demonstrates superior performance and robustness in predicting protein locations.