High-frequency spike inference with particle Gibbs sampling.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biophysics

Background:

• Calcium-sensitive fluorescent indicators are crucial for monitoring neuronal activity in vivo.
• Existing spike-time inference methods struggle with high firing rates (>20 Hz) and quantifying estimation uncertainty.

Purpose of the Study:

• To develop a robust statistical method for accurate spike-time inference and uncertainty quantification.
• To address limitations in current algorithms for high-frequency neuronal firing and baseline fluorescence modulation.

Main Methods:

• Introduced a novel statistical model incorporating bursting activity and baseline fluorescence.
• Applied a Monte Carlo strategy (particle Gibbs with ancestor sampling) for joint posterior distribution estimation.
• Validated the method on CASCADE benchmark datasets and with GCaMP8f indicator data.

Main Results:

• The developed Bayesian inference method achieves competitive performance against state-of-the-art algorithms.
• Successfully resolved interspike intervals as short as 5 ms, demonstrating high temporal resolution.
• Provided unbiased estimates of spike times and model parameters, enabling robust uncertainty quantification.

Conclusions:

• The study presents a flexible Bayesian framework for neuronal spike detection and uncertainty quantification.
• The particle Gibbs sampler offers a powerful tool for analyzing calcium imaging data with high precision.
• This method enhances the reliability of interpreting neuronal population activity from fluorescence recordings.