Biclustering of gene expression data using reactive greedy randomized adaptive search procedure

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Biclustering algorithms simultaneously cluster rows and columns of gene expression data.
• The mean squared residue score is a popular metric for evaluating bicluster quality.
• Existing methods like GRASP and Cheng and Church's approach are foundational in bicluster analysis.

Purpose of the Study:

• To introduce a novel biclustering algorithm, Reactive GRASP, for analyzing large microarray datasets.
• To improve upon existing biclustering techniques by incorporating self-adjusting parameters.
• To enhance the detection of significant biclusters in gene expression data.

Main Methods:

• The proposed Reactive GRASP method utilizes k-means clustering to generate high-quality bicluster seeds.
• Seeds are then grown using Reactive GRASP, which self-adjusts a key parameter based on solution quality.
• This approach involves a construction and local search phases, characteristic of GRASP metaheuristics.

Main Results:

• The Reactive GRASP approach demonstrated superior performance compared to basic GRASP and the Cheng and Church method.
• Statistical and biological validations confirmed the quality of the biclusters identified by Reactive GRASP.
• Experimental results indicate significant improvements in bicluster detection accuracy and efficiency.

Conclusions:

• Reactive GRASP provides a robust method for detecting significant biclusters.
• The algorithm eliminates the need for extensive calibration efforts.
• This approach offers a more efficient and effective tool for analyzing complex gene expression data.