flowClust

flowClust performs model-based clustering of flow cytometry data to identify cellular populations using t-mixture models, Box-Cox transformations, and an Expectation-Maximization algorithm.

Key Features:

• T-Mixture Models: Generalizes Gaussian mixture models to handle outliers and support nonelliptical cluster shapes.
• Box-Cox Transformation: Applies Box-Cox transformations for data normalization to improve clustering adaptability and accuracy.
• Expectation-Maximization Algorithm: Uses EM to estimate model parameters and select transformations simultaneously.
• Outlier Robustness: Accommodates outliers to increase robustness against model misspecification.
• Nonelliptical Cluster Support: Detects nonelliptical cluster shapes beyond Gaussian assumptions.
• Accurate Cluster Number Estimation: Improves estimation of the number of clusters compared to popular mixture models.
• High-throughput Quantification: Enables quantification of multidimensional characteristics across millions of cells.
• Automation of Gating: Provides an automated alternative to manual gating and can mimic manual gating results to reduce subjectivity.
• Reproducible Results: Produces reproducible clustering outputs suitable for research requiring consistency and precision.

Scientific Applications:

• Automated Flow Cytometry Analysis: Automates gating and identification of cellular populations from flow cytometry datasets.
• Health Research and Clinical Studies: Supports analyses in health research, medical diagnosis, and treatment studies.
• High-Dimensional Cellular Phenotyping: Quantifies multidimensional cellular characteristics across large cell counts for phenotyping.
• Method Comparison via Simulation: Facilitates comparative evaluation against popular mixture models through simulation studies.

Methodology:

Implements t-mixture models combined with Box-Cox transformations and applies an Expectation-Maximization algorithm for simultaneous parameter estimation and transformation selection; performance has been assessed via simulation studies showing robustness to model misspecification and improved cluster-number estimation.