BL-FlowSOM: Consistent and Highly Accelerated FlowSOM Based on Parallelized Batch Learning.

Area of Science:

• Computational biology
• Bioinformatics
• Data science in life sciences

Background:

• High-dimensional cytometry data analysis requires efficient computational methods.
• FlowSOM is a high-performing clustering algorithm but faces challenges in consistency and speed.
• Existing methods may exhibit variability and lack optimization for large datasets.

Purpose of the Study:

• To introduce Batch Learning FlowSOM (BL-FlowSOM), an improved clustering algorithm.
• To enhance the consistency and accelerate the computational speed of FlowSOM.
• To provide a robust and efficient tool for high-dimensional cytometry data analysis.

Main Methods:

• Implemented batch learning instead of online learning for the FlowSOM algorithm.
• Utilized principal component analysis (PCA) for initialization.
• Enabled parallelization of the batch learning process.
• Evaluated clustering quality and computational performance.

Main Results:

• BL-FlowSOM demonstrates improved consistency by eliminating randomness in clustering.
• The parallelized batch learning significantly accelerates the clustering process.
• Clustering quality achieved by BL-FlowSOM is equivalent to the original FlowSOM.
• BL-FlowSOM offers a more reliable and faster analysis for cytometry data.

Conclusions:

• BL-FlowSOM provides a consistent and accelerated approach to cytometry data clustering.
• The method maintains high clustering quality while improving computational efficiency.
• BL-FlowSOM represents a significant advancement for analyzing complex, high-dimensional cytometry datasets.
• The algorithm is accessible via Sony's Spectral Flow Analysis (SFA)-Life sciences Cloud Platform.