Abstract

Biclustering algorithms are effective techniques for uncovering previously unknown facts, hidden in gene expression data. The usefulness of clustering algorithms for analyzing gene expression data has been limited because they partition data into mutually exclusive groups rather than real, overlapping gene sets. These gene sets can reveal novel insights helpful for disease diagnosis, prognosis and drug development. This study presents a Combinatorial Firefly-Differential Evolution Algorithm (Combi FI-DE) which enhances the original Differential Evolution (DE) Optimization Algorithm using some concepts in set theory and improves the original Firefly algorithm by sorting the initial biclusters (replacing its if-condition). The efficiency of the algorithm was tested using real gene expression data. The proposed algorithm had favorable comparison with some selected biclustering techniques in terms of the average volume and gene count of the extracted biclusters. In conclusion, the result of the GO enrichment analysis conducted on the mined biclusters suggests that the Combi FI-DE algorithm discovered genes that are involved in protein binding, and other functions which are useful for disease diagnosis and drug development. Therefore, the algorithm is valuable for revealing meaningful insights in gene expression data.

Keywords: Biclustering, gene expression data, firefly algorithm, differential evolution.