Joint Modeling and Registration of Cell Populations in Cohorts of High-Dimensional Flow Cytometric Data

Abstract

In biomedical applications, an experimenter encounters different potential sources of variation in data such as individual samples, multiple experimental conditions, and multivariate responses of a panel of markers such as from a signaling network. In multiparametric cytometry, which is often used for analyzing patient samples, such issues are critical. While computational methods can identify cell populations in individual samples, without the ability to automatically match them across samples, it is difficult to compare and characterize the populations in typical experiments, such as those responding to various stimulations or distinctive of particular patients or time-points, especially when there are many samples. Joint Clustering and Matching (JCM) is a multi-level framework for simultaneous modeling and registration of populations across a cohort. JCM models every population with a robust multivariate probability distribution. Simultaneously, JCM fits a random-effects model to construct an overall batch template – used for registering populations across samples, and classifying new samples. By tackling systems-level variation, JCM supports practical biomedical applications involving large cohorts. Software for fitting the JCM models have been implemented in an R package EMMIX-JCM, available from http://www.maths.uq.edu.au/~gjm/mix_soft/EMMIX-JCM/.

Mapping cell populations in flow cytometry data for cross‐sample comparison using the Friedman–Rafsky test statistic as a distance measure

“…In recent years, computational gating methods have made significant advances in identifying cell populations at the individual sample level 5. Model‐based computational gating approaches, such as Gaussian and multivariate skew‐ t mixture model fitting 6, 7, 8, 9, employ statistical assumptions on the shape and location of cell population distributions. Non‐model‐based methods, such as grid‐based density clustering 10 and spectral clustering 11 algorithms, group cells into homogeneous populations based on unsupervised data clustering.…”

“…Small intersample variations in cell population locations are associated with low population misclassification rates. Other existing approaches, including FLAME 7, HDPGMM 6, JCM 9, and flowMatch 14, bundle the cell population identification method and cross‐sample mapping function together, with the mapping component operating under the principle of global template finding. In FLAME 7, mapping cell populations across samples is the last step of their computational gating method.…”

“…The sample cell populations are then matched to the global populations based on the respective mode locations. In both HDPGMM 6 and JCM 9, a multilevel modeling approach is applied to simultaneously identify cell populations and map populations across samples. A global template is generated based on shared location and shape characteristics among cell populations across samples in the same cohort.…”

Modeling of inter‐sample variation in flow cytometric data with the joint clustering and matching procedure

“…Among these methods, mixture models have been widely employed as the underlying mechanism for characterizing the heterogeneous cell populations (4)(5)(6)(7)(8)(9)(10)(11), taking advantage of the convenient and formal framework offered by a modelbased approach to modeling these complex and multimodal datasets. Using this approach, the FCM data can be conceptualized as a mixture of populations each of which consists of cells with similar expressions, the distribution of which can be characterized by a parametric density.…”

“…It is well known that data generated from flow cytometric studies are often asymmetrically distributed, multimodal, as well as having longer and/or heavier tails than normal. To accommodate this, several methods use a mixture of mixtures approach (7,12) where a final cluster may consist of more than one mixture component, while some others adopt mixture models with skew distributions as components (4,5,7) to enable a single component distribution to correspond to a cluster.…”