Precise calcium-to-spike inference using biophysical generative models.

Area of Science:

• Neuroscience
• Biophysics
• Computational Biology

Background:

• Intramolecular dynamics of fluorescent calcium indicators complicate the relationship between calcium signals and neural action potentials (spikes).
• This distortion hinders accurate spike inference from calcium imaging data, a crucial technique in neuroscience.

Purpose of the Study:

• To characterize the calcium response kinetics of widely used indicators (GCaMP6f, jGCaMP7f, jGCaMP8f).
• To develop improved methods for inferring neural spikes from calcium imaging data by addressing indicator-induced distortions.

Main Methods:

• Utilized in vitro stopped-flow measurements and brain slice recordings to characterize indicator kinetics.
• Developed a multistate biophysical model of GCaMP.
• Created Bayesian Sequential Monte Carlo and machine learning inference models trained on synthetic data derived from the biophysical model.

Main Results:

• Identified novel kinetic features, such as use-dependent slowing of fluorescence decay, causing errors in linear inference models.
• The developed biophysically-inspired models (C-SPIKES) significantly outperformed existing methods in spike timing accuracy and correlation.
• Inference models trained on synthetic data from the biophysical model surpassed those trained on extensive experimental data.

Conclusions:

• The C-SPIKES framework offers a generalizable strategy for accurate spike inference from calcium imaging.
• Separating indicator characterization from inference provides a robust approach for current and future calcium indicators.
• Accurate spike inference is critical for understanding neural circuit dynamics and function.