Bayesian Trees for Automated Cytometry Data Analysis

Abstract

Cytometry is an important single cell analysis technology in furthering our understanding of cellular biological processes and in supporting clinical diagnoses across a variety hematological and immunological conditions. Current data analysis workflows for cytometry data rely on a manual process called gating to classify cells into canonical types. This dependence on human annotation significantly limits the rate, reproducibility, and scope of cytometry's use in both biological research and clinical practice. We develop a novel Bayesian approach for automated gating that classifies cells into different types by combining cell-level marker measurements with an informative prior. The Bayesian approach allows for the incorporation of biologically-meaningful prior information that captures the domain expertise of human experts. The inference algorithm results in a hierarchically-structured classification of individual cells in a manner that mimics the tree-structured recursive process of manual gating, making the results readily interpretable. The approach can be extended in a natural fashion to handle data from multiple different samples by the incorporation of random effects in the Bayesian model. The proposed approach is evaluated using mass cytometry data, on the problems of unsupervised cell classification and supervised clinical diagnosis, illustrating the benefits of both incorporating prior knowledge and sharing information across multiple samples.

Cytomulate: accurate and efficient simulation of CyTOF data

“…According to (35) and (36), the k-means algorithm and GMM-based model (SWIFT) are often not the optimal choice for clustering CyTOF data; 2. There are already plenty of well-established cell identification algorithms and automatic gating techniques such as FlowSOM (10), Bayesian Tree (12), ACDC (37), SCINA (38), etc, all of which have similar goals but slightly different design considerations. For example, using FlowSOM requires minimal prior knowledge on the cell population other than some crude idea on the number of cell types.…”

“…As a result, it requires not only the expression matrix of one particular CyTOF dataset to mimic but also the cell type label associated with each cell as inputs. Although Cytomulate can potentially perform cell type identification or clustering using common algorithms of the likes of K-means, we leave the choice of such algorithm for cell typing to our users as a design choice for the following two reasons: (i) According to (47) and (48), the k-means algorithm and GMM-based model (SWIFT) are often not the optimal choice for clustering CyTOF data; (ii) There are already plenty of well-established cell identification algorithms and automatic gating techniques such as FlowSOM (11), Bayesian Tree (13), ACDC (49), SCINA (50), etc., all of which have similar goals but slightly different design considerations. For example, using FlowSOM requires minimal prior knowledge on the cell population other than some crude idea on the number of cell types.…”

“…Deep learning algorithms such as residual networks (ResNet) (8) and generative adversarial networks (GAN) (9,10) were also adopted to resolve the same problem. While some software such as FlowSOM (11), LAMBDA (12), and Bayesian trees (13) focus on automatic cell type identification, others like CytoTree (14) attempt to explore the trajectories along which cells differentiate into other cells. Various data visualization and dimension reduction algorithms have also been applied to CyTOF (15).…”

Machine Learning of Discriminative Gate Locations for Clinical Diagnosis

“…However, it may be possible to reduce this complexity in practice, for example by leveraging prior knowledge available about markers and cell‐types to provide hints to the search algorithm, or by using an initial manual gating as a starting point and then having the algorithm make local changes to the structure to try to improve it in terms of its predictive accuracy. This initialization could also be done in a semisupervised manner based on given phenotype of cell populations of interest, without requiring prior knowledge for the locations of the gates (e.g., using a Bayesian tree method (Ji et al )).…”

Recent Advances in Computer-Assisted Algorithms for Cell Subtype Identification of Cytometry Data

“…Some of these tools use computer learning algorithms such as linear discriminant analysis (LDA) (Abdelaal et al, 2018) or neural networks to apply the patterns extracted from the training sets to annotate cells from a new dataset. 2The semi-supervised clustering algorithms incorporate user provided marker matrix of known marker associations with particular cell types (Lee et al, 2017;Ji et al, 2018) to guide cellular clustering and identification. These marker matrices are composed of marker expression patterns in various cell types that serve as a cluster dictionaries indicating whether the markers are negative, positive or ignorable for each cell types.…”