A flexible Bayesian framework for unbiased estimation of timescales

Area of Science:

• Dynamical systems analysis
• Statistical modeling
• Neuroscience

Background:

• Timescales are crucial for understanding natural processes.
• Current methods for estimating timescales often suffer from statistical bias due to finite sample sizes.
• This bias can lead to inaccurate characterizations of dynamic processes.

Purpose of the Study:

• To develop a novel method for accurate timescale estimation.
• To address the limitations of traditional autocorrelation fitting methods.
• To provide a robust framework for analyzing dynamic processes in various scientific fields.

Main Methods:

• Utilized a generative model based on a mixture of Ornstein-Uhlenbeck (OU) processes.
• Employed adaptive approximate Bayesian computations (aABC) for timescale estimation.
• Developed a method that accounts for finite sample size and data noise.

Main Results:

• The proposed method accurately recovers timescales from synthetic data.
• Demonstrated the method's effectiveness on real-world data from primate cortex recordings.
• The approach provides a posterior distribution of timescales, quantifying estimation uncertainty.

Conclusions:

• The new Bayesian approach offers a more accurate and reliable way to estimate timescales compared to standard methods.
• The framework is flexible and applicable to diverse datasets and dynamic processes.
• A customizable Python package is available for implementing the proposed methodology.