GateNet

Flow cytometry is a widely utilized technique for identifying cell populations in patient-derived fluids, such as peripheral blood (PB) or cerebrospinal fluid (CSF). Despite its prevalence in research and clinical practice, flow cytometry involves a manual gating process for cell type identification, which is labor-intensive and prone to errors. The need for automated solutions to streamline this process has led to the development of GateNet, a novel neural network architecture designed to enable end-to-end automated gating without the necessity for correcting batch effects.

Research Project

GateNet is trained using an extensive dataset comprising over 8,000,000 events from N=127 PB and CSF samples, manually labeled independently by four experts. The neural network achieves human-level performance on novel, unseen samples, with an F1 score ranging from 0.910 to 0.997. The robustness of GateNet is further demonstrated by its application to a publicly available dataset, where it exhibits a F1 score of 0.936, confirming its generalization capabilities.

A notable feature of GateNet is its efficient processing, requiring only 15 microseconds per event, thanks to implementation on graphics processing units (GPU). Moreover, the study highlights that GateNet attains human-level performance with as few as ~10 samples, making it widely applicable across various domains of flow cytometry.

In summary, GateNet represents a significant advancement in the field of flow cytometry by providing a fully automated gating solution. Its ability to achieve human-level performance, efficient processing, and generalization across datasets positions it as a valuable tool for improving the accuracy and efficiency of cell type identification in diverse applications of flow cytometry.