Automated hierarchical density shaving: a robust automated clustering and visualization framework for large...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene function prediction is crucial in systems biology.
• Clustering biological data (e.g., microarrays) often yields sparse functionally related gene groups.
• Existing methods may struggle with varying cluster densities and require manual parameter tuning.

Purpose of the Study:

• To develop a robust framework for identifying functionally related genes from high-dimensional biological data.
• To address the challenge of detecting clusters with varying densities and sizes.
• To provide an unsupervised method for cluster selection and ranking.

Main Methods:

• Introduced Automated Hierarchical Density Shaving (Auto-HDS), a novel clustering framework.
• Implemented a fast hierarchical density-based clustering algorithm.
• Developed an unsupervised model selection strategy with a stability criterion for cluster ranking.
• Created a 2D visualization for interactive exploration of cluster hierarchies.

Main Results:

• Auto-HDS effectively identifies functionally related gene clusters from microarray data (Gasch and Lee datasets).
• The framework automatically selects clusters of different densities and presents them hierarchically.
• Cluster ranking based on stability provides a reliable measure of biological relevance.
• The 2D visualization aids in intuitive understanding and exploration of clustering results.

Conclusions:

• Auto-HDS offers an effective and automated solution for discovering functionally related genes.
• The framework's ability to handle varying cluster densities and provide stability rankings enhances biological data analysis.
• Auto-HDS represents a significant advancement in unsupervised clustering for biological applications.