A semi-supervised ensemble clustering algorithm for discovering relationships between different diseases by...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Hereditary diseases, including cancer, are often linked to genetic and environmental factors.
• Gene expression data offers valuable insights into hereditary diseases and their interrelationships.
• Analyzing gene expression can reveal connections between different diseases, such as shared cellular damage in breast and blood cancers.

Purpose of the Study:

• To discover cell-to-cell biological communications within gene expression data.
• To identify relationships between various diseases by extracting intercellular communication patterns.
• To develop a novel computational approach for analyzing complex gene expression datasets.

Main Methods:

• Development of a semi-supervised ensemble clustering algorithm.
• Implementation of a stratified feature sampling mechanism for high-dimensional data.
• Introduction of a novel similarity metric to enhance the diversity of data partitions.

Main Results:

• The algorithm was validated on UCI machine learning repository datasets.
• Application to the FANTOM5 dataset revealed significant cell-to-cell biological communications.
• Highest communication was observed between basophils and ciliary epithelial cells (62,809 promoters).

Conclusions:

• Cell-to-cell biological similarity, particularly within clusters, can effectively detect relationships between diseases.
• The proposed method provides a robust framework for uncovering disease connections through gene expression analysis.
• This approach has implications for understanding complex diseases like cancer.