Cluster ensemble based on Random Forests for genetic data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genetics

Background:

• Clustering is vital in bioinformatics for pattern detection in genetic data, particularly for population structure analysis using genetic variations like single nucleotide polymorphisms.
• High-throughput DNA sequencing generates massive genetic datasets, necessitating efficient analytical methods.
• Existing methods struggle with the scale and complexity of modern genetic data.

Purpose of the Study:

• To propose RFcluE, a cluster ensemble approach for uncovering the underlying structure of genetic data.
• To leverage the Random Forests (RFs) proximity measure within an unsupervised learning framework.
• To evaluate the effectiveness of RFcluE on high-dimensional, real genetic datasets.

Main Methods:

• Developed RFcluE, a cluster ensemble framework that aggregates results from multiple RF clustering executions.
• Utilized the proximity measure generated by Random Forests (RFs) for unsupervised analysis.
• Conducted experiments on large-scale genetic datasets to assess performance and parameter impact.

Main Results:

• RFcluE demonstrates efficiency in handling high-dimensional genetic data.
• The cluster ensemble approach, combining multiple RF clusterings, yields robust and high-quality results.
• Parameter variations and ensemble diversity significantly influence RFcluE performance.

Conclusions:

• RFcluE effectively addresses population structure analysis in genetic data.
• Combining multiple RF clusterings via an ensemble method enhances robustness and accuracy.
• The RF algorithm's inherent features (bagging, random subspace) provide diverse data views crucial for ensemble effectiveness.