Ensemble of Multiple Classifiers for Multilabel Classification of Plant Protein Subcellular Localization

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate prediction of protein subcellular localization is crucial for functional genome annotation.
• Plant protein localization prediction is complex due to multiclass, multilabel, and imbalanced data challenges.
• Existing methods may struggle with the diverse range of plant cellular compartments.

Purpose of the Study:

• To develop an improved and reliable strategy for predicting plant protein subcellular localization.
• To address the challenges of multiclass, multilabel, and imbalanced data in plant protein localization.
• To enhance the accuracy and reliability of functional genome annotation in plants.

Main Methods:

• Extraction of 479 diverse feature spaces for protein localization.
• Application of feature selection to reduce dimensionality and identify informative subsets.
• Development of an ensemble machine learning model using average voting among three heterogeneous classifiers.

Main Results:

• The ensemble method successfully predicted subcellular localizations for single and multilabel locations.
• Achieved an overall classification accuracy of 84.58% for 11 plant cellular compartments.
• Demonstrated improved prediction accuracy and reliability compared to individual models.

Conclusions:

• The proposed ensemble machine learning method offers a robust approach for plant protein subcellular localization prediction.
• This strategy effectively handles multiclass, multilabel, and imbalanced datasets.
• The findings contribute to more accurate functional genome annotation in plants.