Precise calcium-to-spike inference using biophysical generative models

The nonlinear intramolecular dynamics of fluorescent indicators of neural activity can distort the accurate estimate of action potential (“spike”) times. In order to develop a more accurate spike inference algorithm we characterized the kinetic responses to different calcium concentrations of three popular calcium indicators (GCaMP6f, jGCaMP7f, and jGCaMP8f) using in vitro stopped-flow and brain slice recordings. jGCaMP8f showed a use-dependent slowing of fluorescence responses that caused existing inference methods to generate numerous false positives. From these data we developed multistate models of GcaMPs and used them to create a Bayesian Sequential Monte Carlo (Biophys _SMC ) and machine learning (Biophys _ML ) inference methods that largely reduced false positives and improved the spike time accuracy from publicly available ground-truth data. The Biophys _SMC method detected individual spikes with an average uncertainty of 6 milliseconds, a 6-fold improvement over previous methods. Our framework highlights the advantages of physical model-based approaches over model-free algorithms.