Testing for a difference in means of a single feature after clustering

Area of Science:

• Computational biology
• Statistical genetics
• Bioinformatics

Background:

• Interpreting and validating clustered observations is crucial for many applications.
• Classical hypothesis tests for comparing feature means between clusters inflate Type I error rates.
• Accurate cluster validation is essential for reliable data interpretation in fields like single-cell genomics.

Purpose of the Study:

• To propose a novel hypothesis test for comparing feature means between two clusters.
• To address the inflated Type I error rates associated with classical tests in cluster analysis.
• To provide an efficient and reliable method for validating clusters derived from hierarchical or k-means algorithms.

Main Methods:

• Development of a new statistical test for the difference in means between two clusters.
• The test is designed for use with hierarchical clustering and k-means clustering outputs.
• Evaluation of the test's performance through simulations and application to real-world single-cell RNA-sequencing data.

Main Results:

• The proposed test effectively controls the selective Type I error rate in finite samples.
• The method is computationally efficient.
• Simulations demonstrate the test's validity and statistical power.

Conclusions:

• The new test offers a statistically sound approach for validating cluster comparisons.
• It provides a reliable alternative to classical methods that suffer from inflated Type I errors.
• The test is applicable to diverse datasets, including single-cell RNA-sequencing, improving biological insights.