[Automatic clustering method of flow cytometry data based on t-distributed stochastic neighbor embedding].

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Context:

• Traditional multi-parameter flow cytometry data analysis relies on manual gating, which is complex and requires specialized expertise.
• Existing methods struggle with complex cell populations, particularly those with asymmetric or trailing distributions.

Purpose:

• To develop and evaluate a novel clustering algorithm for multi-parameter flow cytometry data based on t-distributed stochastic neighbor embedding (t-SNE).
• To compare the performance of the t-SNE algorithm against Kernel Principal Component Analysis (KPCA) for dimensionality reduction and subsequent K-means clustering.

Summary:

• The proposed algorithm utilizes t-SNE to reduce high-dimensional flow cytometry data to a low-dimensional space by transforming Euclidean distances into conditional probabilities, representing data similarity.
• Human peripheral blood cells were analyzed using flow cytometry, and the resulting data were processed using the t-SNE algorithm and compared with KPCA.
• The t-SNE based approach demonstrated superior clustering performance, particularly for cell populations with challenging distributions, achieving a classification accuracy of 92.55%.

Impact:

• This t-SNE based clustering algorithm offers a promising solution for the automated analysis of complex multi-color, multi-parameter flow cytometry data.
• The improved accuracy and efficiency can aid researchers in more effectively identifying and quantifying cell populations.
• Potential to streamline experimental workflows and reduce the reliance on manual, subjective gating procedures in flow cytometry analysis.