Scalable clustering algorithms for continuous environmental flow cytometry.

Area of Science:

• Oceanography
• Environmental Science
• Data Science

Background:

• Technological advancements enable high-frequency flow cytometry in aquatic environments.
• The SeaFlow cytometer collects vast amounts of data on microbial phytoplankton.
• Traditional methods are insufficient for analyzing large-scale, high-frequency flow cytometry datasets.

Purpose of the Study:

• To evaluate existing algorithms for classifying large-scale environmental flow cytometry data.
• To develop and propose a scalable clustering approach for high-frequency flow cytometry data.

Main Methods:

• Applied large-scale Gaussian mixture models to massive datasets using Hadoop.
• Explored partitioning of data into homogeneous sections for enhanced classification.
• Implemented the approach in Java for use with Hadoop.

Main Results:

• Gaussian mixture models with partitioning outperform current state-of-the-art cytometry classification algorithms in accuracy.
• The proposed approach effectively classifies microbial phytoplankton populations from large-scale datasets.

Conclusions:

• Gaussian mixture models with partitioning provide a scalable and accurate solution for analyzing high-frequency flow cytometry data.
• This method addresses the challenges posed by large-scale environmental data streams.