A fully automated clustering method for flow cytometry based on density and nonparametric clustering

Abstract

The invention discloses a flow cytometry automatic clustering method based on density and non-parameter clustering, which belongs to the technical field of medical data processing and flow cytometry data analysis, and takes into account the characteristics of different algorithms and flow cytometry data analysis. The process does not require the user to specify the number of cell groups in the whole process, which is conducive to the automatic process and the discovery and mining of new cell groups; the dimensionality reduction speed is fast, and the UMAP dimensionality reduction speed is 2-10 times faster than the t-SNE dimensionality reduction, which greatly saves the time for automatic clustering ; DBSCAN combined with FlowPeaks algorithm can accurately distinguish cell populations of any shape, and can effectively eliminate noise interference and non-specific signals.

Description

technical field [0001] The invention relates to the technical field of medical data processing and flow cytometry data analysis, in particular to a flow cytometry automatic clustering method based on density and nonparametric clustering. Background technique [0002] Existing cell clustering methods are based on principal component analysis (PCA) dimensionality reduction and K-means clustering. In this algorithm, PCA has a fast dimensionality reduction speed, which is suitable for the case where the positive and negative groups are clearly grouped. Once the positive and negative groups are not sufficiently separated, the groups after dimensionality reduction will overlap, resulting in difficulty in clustering; K-means clustering The class accuracy is poor, and it is only suitable for accurate clustering to distinguish the cell groups with a circular distribution after dimensionality reduction. The clustering ability of the irregularly distributed cell groups after dimensiona...

AI Technical Summary

Problems solved by technology

In this algorithm, the PCA dimensionality reduction speed is fast, and it is suitable for the situation where the positive group and the negative group are very clearly grouped. Once the positive and negative groups are not sufficiently separated, the groups after dimensionality reduction will overlap, resulting in difficulty in clustering; K-means clustering The class accuracy is poor, and it is only suitable for accurate clustering to distinguish cell populations that are distributed in a circular manner after dimensionality reduction, and the cell populations that are irregularly distributed after dimensionality reduction have extremely poor clustering ability, and the K-means algorithm needs to specify cell populations in advance Quantity, which is an obstacle to automatic clustering, will introduce artificial clustering errors, which is not conducive to the automatic process and the discovery and mining of new cell populations

There are also methods that do not involve dimensionality reduction algorithms, and directly use the neural network model to perform grouping and cell property judgment in multidimensional space, which will cause a large loss in accuracy and performance.

Images