KMD clustering: robust general-purpose clustering of biological data

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Biological data is often noisy and high-dimensional, necessitating specialized clustering algorithms.
• Current methods exhibit variable performance and require difficult hyperparameter tuning.
• There is a need for robust, general-purpose clustering solutions for large biological datasets.

Purpose of the Study:

• To introduce k minimal distance (KMD) clustering, a novel, general-purpose clustering algorithm.
• To address the challenges of hyperparameter tuning and scalability in biological data analysis.
• To provide a robust clustering method applicable to diverse biological datasets.

Main Methods:

• KMD clustering is a generalization of single and average linkage hierarchical clustering.
• A generalized silhouette-like function is introduced to eliminate the need for the hyperparameter k.
• Sampling techniques are employed to enable analysis of million-object datasets.

Main Results:

• KMD clustering shows consistent high performance across simulated, mass cytometry, and single-cell RNA sequencing (scRNA-seq) datasets.
• The method effectively handles noisy and high-dimensional biological data.
• KMD clustering outperforms existing general and specialized clustering approaches.

Conclusions:

• KMD clustering offers a robust and general-purpose solution for biological data analysis.
• The method overcomes key limitations of existing clustering techniques, including hyperparameter sensitivity and scalability.
• KMD clustering provides a reliable tool for researchers working with large and complex biological datasets.