PT  - JOURNAL ARTICLE
AU  - Sebastien Kirchherr
AU  - Sebastian Mildiner Moraga
AU  - Gino Coudé
AU  - Marco Bimbi
AU  - Pier F Ferrari
AU  - Emmeke Aarts
AU  - James J Bonaiuto
TI  - Bayesian multilevel hidden Markov models identify stable state dynamics in longitudinal recordings from macaque primary motor cortex
AID  - 10.1101/2022.10.17.512024
DP  - 2022 Jan 01
TA  - bioRxiv
PG  - 2022.10.17.512024
4099  - http://biorxiv.org/content/early/2022/10/21/2022.10.17.512024.short
4100  - http://biorxiv.org/content/early/2022/10/21/2022.10.17.512024.full
AB  - Neural populations, rather than single neurons, may be the fundamental unit of cortical computation. Analyzing chronically recorded neural population activity is challenging not only because of the high dimensionality of activity in many neurons, but also because of changes in the recorded signal that may or may not be due to neural plasticity. Hidden Markov models (HMMs) are a promising technique for analyzing such data in terms of discrete, latent states, but previous approaches have either not considered the statistical properties of neural spiking data, have not been adaptable to longitudinal data, or have not modeled condition specific differences. We present a multilevel Bayesian HMM which addresses these shortcomings by incorporating multivariate Poisson log-normal emission probability distributions, multilevel parameter estimation, and trial-specific condition covariates. We applied this framework to multi-unit neural spiking data recorded using chronically implanted multi-electrode arrays from macaque primary motor cortex during a cued reaching, grasping, and placing task. We show that the model identifies latent neural population states which are tightly linked to behavioral events, despite the model being trained without any information about event timing. We show that these events represent specific spatiotemporal patterns of neural population activity and that their relationship to behavior is consistent over days of recording. The utility and stability of this approach is demonstrated using a previously learned task, but this multilevel Bayesian HMM framework would be especially suited for future studies of long-term plasticity in neural populations.Competing Interest StatementThe authors have declared no competing interest.