Metric for measuring the effectiveness of clustering of DNA microarray expression

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray technology enables simultaneous expression analysis of thousands of genes.
• Gene expression data is often clustered using algorithms like hierarchical and k-means clustering with metrics such as Euclidean distance or Pearson correlation.
• Current validation indices for gene clustering lack direct correlation with biological features, confusing practitioners.

Purpose of the Study:

• To propose a novel metric for evaluating gene clustering algorithm effectiveness.
• To assess clustering based on biological features like biological processes and molecular functions.
• To provide a quantitative measure for comparing different clustering algorithms.

Main Methods:

• Developed a metric quantifying inter-cluster cohesiveness and intra-cluster separation.
• Applied the metric to gene expression data from a study on retinoids and cancer suppression.
• Evaluated hierarchical and k-means clustering with Euclidean and Pearson correlation distances.

Main Results:

• Genes with similar expression profiles are more closely related to biological processes than molecular functions.
• The proposed metric effectively measures both inter- and intra-cluster cohesiveness.
• The metric provides a single quantitative value (0 indicates maximum cohesiveness and separation) for easier algorithm comparison.

Conclusions:

• The best gene clustering algorithms should exhibit biological feature-based cohesiveness within clusters and maximal separation between clusters.
• The proposed metric is novel, easy to compute, and offers a unified measure of clustering effectiveness.
• The metric can be extended to other gene features (e.g., DNA binding sites, protein interactions) and applied to other domains.