For those seeking a review of connectomics and cell types, Neuron recently had a special issue with many great reviews and perspectives (and no, we're not saying this just because they cite us!)
Our model is based on Charles Kemp's Infinite Relational Model, a Bayesian nonparametric extension of the stochastic block model. Stochastic block models assume that each cell (or entity) has a single type, and those types heavily influence the resulting connectivty patterns.
"Discovering Latent Classes in Relational Data" is an excellent tech report by Kemp, Griffiths, and Tenenbaum which motivates the analysis and latent modeling of relational data.
For an introduction to the Dirichlet Process prior that we are using, section 2 of Tom Griffith's introduction to the Indian Buffet process is concise and accessable A more detailed introduction can be found in Kevin Murphy's excellent "Machine Learning: A Probabilistic Perspective" book. You should read this book anyway, it will make you a better person.
Inference via Markov Chain Monte Carlo is a massive field unto itself, but both Murphy's book as well as Chris Bishop's book have good introductions.