CytOpT: Optimal transport with domain adaptation for interpreting flow cytometry data

The automated analysis of flow cytometry measurements is an active research field. We introduce a new algorithm, referred to as CytOpT, using regularized optimal transport to directly estimate the different cell population proportions from a biological sample characterized with flow cytometry measurements. We rely on the regularized Wasserstein metric to compare cytometry measurements from different samples, thus accounting for possible misalignment of a given cell population across samples (due to technical variability from the technology of measurements). In this work we rely on a supervised learning technique, based on the Wasserstein metric, that is used to estimate an optimal reweighting of class proportions in a mixture model from a source distribution (with known segmentation into cell sub-populations) to fit a target distribution with unknown segmentation. Due to the high dimensionality of flow cytometry data, we use stochastic algorithms to approximate the regularized Wasserstein metric to solve the optimization problem involved in the estimation of optimal weights representing the cell population proportions in the target distribution. Several flow cytometry data sets are used to illustrate the performances of CytOpT that are also compared to those of existing algorithms for automatic gating based on supervised learning.