Classification of gene expression data: A hubness-aware semi-supervised approach.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Gene expression data classification is crucial for biomedical recognition tasks.
• Obtaining labeled data for large samples is often challenging, necessitating semi-supervised methods.
• High-dimensional gene expression data presents challenges like the curse of dimensionality and hubness.

Purpose of the Study:

• To propose a semi-supervised extension of the Naive Hubness-Bayesian k-Nearest Neighbor (NHBNN) classifier.
• To introduce a novel hubness-aware certainty score for the self-training schema.
• To improve classification accuracy and reduce computational costs in gene expression analysis.

Main Methods:

• Developed a semi-supervised extension of NHBNN using a self-training schema.
• Proposed a new hubness-aware certainty score to enhance the self-training process.
• Conducted experiments on publicly available gene expression datasets.

Main Results:

• The proposed semi-supervised NHBNN classifier demonstrated superior performance compared to existing methods.
• Each component of the proposed approach (classification algorithm, semi-supervised technique, hubness-aware certainty score) was found to be relevant to its performance.
• Experiments confirmed the effectiveness of the hubness-aware certainty score in improving classification.

Conclusions:

• The developed approach can increase classification accuracy and reduce computational costs for gene expression data.
• Hubness-aware techniques show promise for broader applications in biomedical machine learning.
• An open-source implementation (PyHubs) is provided to facilitate the adoption of these techniques.