Cookie

Cookie selects representative samples from large single-cell sequencing datasets to enable experimental characterization of diverse cellular populations.

Key Features:

• Relationship quantification: Vectorizes sample properties and computes Manhattan distances to quantify relationships and similarities among samples.
• Sample size determination: Evaluates coverage of key properties across candidate sample sizes to determine an appropriate subset size that balances diverse properties and priority levels.
• K-medoids clustering: Applies k-medoids clustering and selects cluster medoids as representative samples, leveraging k-medoids' robustness to noise and outliers.
• Benchmarking across datasets: Demonstrates efficacy, efficiency, and flexibility via comparisons with conventional sampling methods on a single-cell atlas dataset, epidemiology surveillance data, and simulated datasets.
• Implementation: Implemented in R.

Scientific Applications:

• Single-cell sequencing sample selection: Selects representative subsets from high-dimensional single-cell sequencing data to support downstream experimental characterization.
• Epidemiological surveillance sampling: Identifies representative samples from epidemiology surveillance datasets for monitoring and analysis.
• Method benchmarking and simulation studies: Uses simulated datasets to evaluate sampling performance and compare against conventional methods.

Methodology:

Vectorize sample properties, compute Manhattan distances to quantify similarities, evaluate coverage across candidate sample sizes to choose an appropriate size, and perform k-medoids clustering to select cluster medoids as representative samples.