A prediction-based resampling method for estimating the number of clusters in a dataset

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray technology is vital in biological and medical research, particularly for tumor classification.
• Identifying new tumor classes from gene-expression profiles presents a significant statistical challenge.
• Estimating the number of clusters and assigning samples are key aspects of this problem.

Purpose of the Study:

• To develop and evaluate a novel method for estimating the number of clusters in datasets.
• To address the challenge of determining the optimal number of tumor classes in gene-expression data.
• To improve the accuracy and robustness of cluster number estimation in cancer research.

Main Methods:

• Development of a prediction-based resampling method named Clest.
• Comparison of Clest against six existing methods using simulated and real-world gene-expression data.
• Evaluation of method performance on four published cancer microarray studies.

Main Results:

• Clest demonstrated superior accuracy and robustness compared to existing methods.
• The new method effectively estimates the number of clusters in gene-expression datasets.
• Performance was validated across diverse cancer microarray datasets.

Conclusions:

• Combining prediction accuracy with resampling yields reliable estimates for the number of clusters.
• Clest offers an accurate and robust solution for a critical problem in cancer genomics.
• The findings support the use of Clest for analyzing gene-expression data in tumor classification.