Integrative Approach to Gene Expression Data Analysis: Combining Biclustering Techniques with Gene Ontology

This study refined biclustering methods for gene expression analysis, introducing quality criteria based on mutual information for defining bicluster structures. A hybrid biclustering model utilizing ensemble algorithms and Bayesian optimization was developed to optimize these criteria effectively. Tested on cancer gene expression data, the model used objective functions based on mean squared residue (MSR) and mutual information. Results showed the mutual information criterion to be superior, leading to fewer, more informative biclusters, enhancing gene subset identification for diagnostic purposes. Additionally, gene ontology analysis was integrated into the bicluster quality evaluation, facilitating significant gene subset formation. The findings confirmed that biclustering based on mutual information is more effective than the MSR metric for classifying samples with significant gene subsets, demonstrating the model’s utility in identifying relevant genetic markers for disease diagnosis.