A self-training subspace clustering algorithm based on adaptive confidence for gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Gene clustering is crucial for identifying co-expressed genes and understanding biological processes.
• Semi-supervised learning, particularly self-training, shows promise for gene clustering but is susceptible to performance degradation due to mislabeling.
• Existing methods struggle with the accumulation of errors in self-training for gene expression data.

Purpose of the Study:

• To propose a novel self-training subspace clustering algorithm with adaptive confidence (SSCAC) for gene expression data.
• To address the challenge of mislabeling inherent in self-training methods.
• To improve the accuracy and robustness of gene clustering by leveraging low-rank representation and adaptive label confidence.

Main Methods:

• Utilizing low-rank representation with distance penalty to uncover the subspace structure of gene expression data.
• Developing a semi-supervised clustering objective function incorporating label confidence within a self-training subspace clustering framework.
• Implementing an adaptive adjustment strategy for label confidence, guided by the gravitational search algorithm, to minimize the impact of mislabeled data.

Main Results:

• The proposed SSCAC algorithm effectively mines the subspace structure of gene expression data.
• The adaptive label confidence adjustment successfully mitigates the negative effects of mislabeled data.
• Extensive experiments on benchmark datasets demonstrate the superiority of SSCAC over state-of-the-art algorithms.

Conclusions:

• SSCAC offers a robust and accurate approach to gene clustering using semi-supervised learning.
• The algorithm's adaptive confidence mechanism enhances reliability in the presence of noisy labels.
• SSCAC represents a significant advancement in analyzing gene expression data for functional genomics research.