Hierarchical clustering of high-throughput expression data based on general dependences.

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• High-throughput technologies generate complex datasets with thousands of features (genes, metabolites).
• These datasets contain both linear and nonlinear relationships between features.
• Traditional clustering methods, based on linear associations, fail to capture critical nonlinear regulatory patterns.

Purpose of the Study:

• To develop a novel clustering method capable of identifying general dependences, including nonlinear relations, in high-dimensional biological data.
• To overcome limitations of existing methods in handling high dimensionality and noise.
• To reveal biologically relevant patterns missed by conventional approaches.

Main Methods:

• Developed a sensitive nonparametric measure for general dependence between random variables in high dimensions.
• Integrated this measure into a hierarchical clustering algorithm.
• Evaluated the method using simulation studies and a real-world microarray dataset.

Main Results:

• The proposed method demonstrated superior performance compared to correlation- and mutual information-based clustering in identifying nonlinear feature dependences.
• Application to cell-cycle time-series gene expression data yielded biologically relevant clustering results.
• The method effectively handles high dimensionality and noise in biological data.

Conclusions:

• The developed hierarchical clustering method based on general dependence is effective for analyzing complex biological data.
• It offers a significant advancement over traditional methods for uncovering nonlinear regulatory mechanisms.
• The approach provides a valuable tool for biological discovery using high-throughput expression data.