Bayesian Sampling Method For Calcium Imaging Data

Calcium imaging is an increasingly popular technique for large scale data acquisition in neuroscience.This method detects underlying,single neuron activity indirectly through observations of fluorescent indicators for calcium concentration.At the same time,with the formation and development of engineering of genetically encoded calcium indicators,people can reliably detect action potentials in vivo.From a statistical perspective,these developments pose significant challenges,which can be summarized to be three major problems.One of them,which is studied in this paper,is the inference of exact spike times from the noisy calcium signal.This paper takes calcium imaging data as the subject of study to solve the problem of inferring exact spike times from the noisy calcium signal.We compare the Markov Chain Monte Carlo(MCMC)algorithm and the adaptive Metropolis algorithm in this paper.In view of the shortcomings of the MCMC algorithm and for deconvolving the spiking activity of each neuron from the noisy signal,the paper uses the adaptive Metropolis to sample spike times.The mian idea of adaptive Metropolis algorithm is that the proposal distribution function is updated adaptively in the iteration process other than determined beforehand,thus avoiding the selection problem of the proposal distribution and improving the convergence speed of the algorithm.In order to verify the validity and practicability of the algorithm,this paper applies this adaptive Metropolis algorithm to calcium imaging data from spinal cord neurons in-vitro,collected using the calcium indicator GCaMP6 s.The Root Mean Squared Error and R-square are compared with the existing MCMC algorithm.Meanwhile,this paper also designes a calcium imaging data analysis system for testing.The results show that adaptive Metropolis algorithm is effective in obtaining better estimations of the true spike times.