Spike trains, which are successive electrochemical signals generated by nerve cells, can facilitate inference about the brain’s state in a given environment. Inference about functional connectivity (FC), e.g. the statistical correlation between neurons based on spike trains, offers crucial insight on the interactions between different brain areas. The technological advancement in neural recording provides an abundance of data for statistical analyses. However, achieving both biological interpretability and computational scalability poses significant challenges in modelling FC. In this talk, we introduce a novel multi-neuron latent dynamics model based on the spike generation mechanism, coupled with an efficient approximate Bayesian inference procedure. To facilitate downstream analyses, we present a convenient test statistic for comparing inferred FCs. Application of our method to experimental data uncovers changes in FC in response to alcohol cues in the orbitofrontal cortex of rats.