Selecting Representative Samples From Complex Biological Datasets Using K-Medoids Clustering.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell sequencing generates massive datasets, posing challenges for selecting representative samples.
• Conventional sampling methods struggle with diverse cell properties and complex datasets.

Purpose of the Study:

• To develop an efficient toolkit, Cookie, for selecting representative samples from massive single-cell populations.
• To address the need for robust and compact sampling that balances diverse cell properties.

Main Methods:

• Cookie quantifies sample relationships using Manhattan distances.
• It determines optimal sample size by evaluating key property coverage.
• K-medoids clustering identifies representative samples as cluster centers.

Main Results:

• Cookie demonstrated high efficacy, efficiency, and flexibility in selecting representative samples.
• Comparisons were made using single-cell atlas, epidemiology, and simulated datasets.
• The toolkit successfully identified representative subsets from complex cell populations.

Conclusions:

• Cookie provides an effective solution for representative sampling in large-scale single-cell studies.
• The toolkit enhances experimental characterization by ensuring data generalizability.
• Cookie is freely available as an R package.