Nunchaku: optimally partitioning data into piece-wise contiguous segments

Area of Science:

• Data analysis
• Statistical modeling
• Microbiology

Background:

• Analyzing 1D time series often involves identifying linear relationships between variables.
• Current methods for detecting these linear regions are typically ad hoc and subjective.

Purpose of the Study:

• To develop a statistically rigorous, Bayesian approach for partitioning 1D time series data into segments with linear relationships.
• To provide a general solution for identifying discontinuous change points in data.
• To apply the method to microbial growth analysis.

Main Methods:

• A Bayesian approach was developed to infer optimal partitioning of datasets.
• The method identifies contiguous piece-wise linear segments and segments described by linear combinations of arbitrary basis functions.
• The algorithm was applied to analyze microbial growth data, specifically optical density and cell counts.

Main Results:

• The algorithm successfully identified regions of linear proportionality between optical density and cell number for microbial growth.
• It automatically detected regions of exponential growth for Escherichia coli and Saccharomyces cerevisiae.
• The Monod constant for budding yeast growth on fructose was inferred.

Conclusions:

• The developed Bayesian algorithm offers a statistically rigorous and automated solution for identifying linear segments and change points in 1D time series data.
• This method enhances reproducibility and facilitates high-throughput data analysis in diverse scientific disciplines.
• The Nunchaku Python package is available for broader scientific use.