Cluster Based Association Measures with Applications

Area of Science:

• Biostatistics
• Bioinformatics
• Data Science

Background:

• Variable relationships are often non-linear and datasets may contain hidden subgroups.
• Standard correlation measures like Pearson or Spearman can be misleading in such complex scenarios.
• High-dimensional data with substructures are increasingly common in biomedical research.

Purpose of the Study:

• To develop a novel association procedure that accounts for hidden data clusters.
• To quantify associations between univariate and multivariate variables, irrespective of their relationship form.
• To provide a robust measure for heterogeneous data common in biomedical research.

Main Methods:

• Developed CLuster based Association Measures (CLAM), a novel procedure.
• Integrated clustering algorithms to detect hidden subgroups.
• Utilized association measures suitable for arbitrary relationships within detected clusters.

Main Results:

• CLAM accurately quantifies associations in data with hidden clusters.
• The method is versatile, applicable to both univariate and multivariate variables.
• Demonstrated performance on synthetic and diverse real-world datasets including cell-cycle genes, microbiome data, and imaging datasets.

Conclusions:

• CLAM offers a robust solution for association analysis in complex, heterogeneous datasets.
• The method addresses the limitations of traditional correlation measures in the presence of hidden substructures.
• CLAM is well-suited for biomedical research and other fields generating high-dimensional data.