Combining Neural Networks, UMAP and DBM Clustering to Identify Cell Populations Accurately, Quickly and Easily in Mass and Fluorescence Cytometry

Abstract Identifying cell populations in flow cytometry data is mostly done via a “manual gating” method that often lacks verifiability and reproducibility, even in the hands of experienced investigators. In 2018, flowLearn outperformed the previous best automatic gating methods by demonstrating near perfect identification of known populations on two fluorescence cytometry datasets. Then LDA exemplified similar accuracy on five mass cytometry datasets, but with much easier dataset training. However, for optimal results, LDA requires extensive manual gating on the samples it classifies with that training. Here, we introduce an easily trainable multilayer perceptron (MLP) neural network for automatic gating. Compared to LDA, FlowSOM and PhenoGraph on three mass and six fluorescence cytometry datasets, MLP is most accurate by a significant margin, tied with LDA as the fastest and uniquely able to replace manual gating except for training purposes. Furthermore, combining MLP with UMAP’s guided dimensionality reduction feature and DBM’s clustering effectively detects new populations.