Biclustering Gene Expression Data with Subspace Evolution

Biclustering is crucial for gene expression data analysis, but evolutionary algorithm-based methods often suffer from high computational costs. To address this, we propose a novel subspace evolution-based biclustering method that significantly reduces runtime by constraining the search space and enhancing population diversity. Our approach first partitions the original expression matrix into submatrices, within which bicluster seeds are efficiently identified using a newly designed selection strategy. These seeds are then expanded and merged to form final biclusters. Experimental evaluation on both synthetic and real-world gene expression datasets demonstrates that our method outperforms existing typical biclustering algorithms in bicluster quality, achieving an average improvement of 57.31% in mean recovery. Moreover, it reduces runtime by approximately 63.42% compared to state-of-the-art evolutionary biclustering methods.