A general framework for biclustering gene expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Data Mining

Background:

• Numerous biclustering methods exist for gene expression data analysis.
• Current methods are limited in discovering all pattern types and often require algorithm redesign for novel patterns.

Purpose of the Study:

• To propose a universal biclustering approach capable of discovering all computable patterns in gene expression data.
• To leverage Kolmogorov complexity for a more comprehensive pattern detection strategy.

Main Methods:

• Utilizing Kolmogorov complexity to quantify submatrix randomness as a measure of bicluster quality.
• Developing a Markov Chain Monte Carlo algorithm based on algorithmic probability for bicluster searching.
• Extending the method for conventional and checkerboard biclustering tasks.

Main Results:

• Demonstrated the ability to discover a wide range of patterns, unlike existing methods.
• Showcased the versatility and promise of the Kolmogorov complexity-based approach.
• Preliminary experiments on simulated and real data yielded positive outcomes.

Conclusions:

• The proposed method offers a versatile and promising approach for comprehensive bicluster discovery in gene expression data.
• Kolmogorov complexity provides a robust theoretical foundation for identifying diverse biological patterns.
• The method's adaptability extends its utility to related clustering problems.