Agent-Based Models and Cell Assemblies

[whock]

During Manny's, Erika's, and my presentation we talked about further graph theoretic techniques that could be applied to a dynamic graph of neural cell assemblies. One avenue I found is something called agent-based models (ABMs). In an ABM, as I understand them, nodes are 'agents' corresponding to some entity in the world, for example a neuron or neural assembly. These agents exist in some state space and are controlled by a set of rules that determine how their states evolve over time as well as how they interact with one another. These time-varying interactions can be thought of as the edges and are governed by the rule set as well as, I believe, some learning rules. Here's the first chapter of a textbook on ABMs: http://press.princeton.edu/chapters/s9639.pdf [1] So far ABMs have been used to study biological phenomena from insect population dynamics [2] to bacterial colony aggregation[3]. By establishing rules that determine how the agents interact over time, emergent behaviors can arise and you can gain insight into the population as a whole. I think this would be a cool method to apply to cell assemblies where each assembly is an agent and there are rules for how assemblies interact with one another based on empirical neural recordings. This might allow us to better understand how assembly activity evolves over time, and above all, towards what goal i.e. what are the emergent computational features of the rule-based agent interactions.

Are people familiar with the pros and cons of this method? Seems like its been used more for social sciences so far but I don't see why it wouldn't be helpful for understanding cell assemblies.

[1] Agent-Based and Individual-Based Modeling:A Practical Introduction, Steven F. Railsback & Volker Grimm, Princeton University Press

[2] Evans, A., Morgan, D, & Parry, H. (2004). Aphid Population Dynamics in Agricultural Landscapes: An Agent-based Simulation Model. International Environmental Modeling and Software Society (iEMSs) 2004 International Congress on Environmental Modeling and Software, Osnabruck, Germany.

[3] Lardon LA, Merkey BV, Martins S, Dötsch A, Picioreanu C, Kreft JU, Smets BF (2011). iDynoMiCS: next-generation individual-based modelling of biofilms. Environmental Microbiology 13: 2416-2434

[dlee138]

Is size/number of agents perhaps an issue for using ABM for cell assembly? The raw number of neurons involved in a network in biological applications may make it too computational difficult, so perhaps it is used more in social sciences (I have no idea if this is true, just giving a suggestion).

Wouldn't an issue with ABM be the fact that it is only as reliable as its "agents" are? If people in a social environment or neurons in the brain behave irrationally, wouldn't the model be thrown off?

[whock]

These are interesting points. For the first one, I think scale and computational load is frequently a problem in neural modeling. So there are three solutions:

1) Scale down the net. Pros: still dealing with neurons Cons: lose n, may lose emergent features [1] 2) Model the agent as a cluster of neurons. This is consistent with, as we mentioned in the presentation, "assembly" often referring to multiple neurons: Pros: can still model whole net, consistent with literature. Cons: Not dealing with neurons explicitly. Consistent with literature. 3) Set of all options I have not mentioned.

[1] Anderson, Philip W. "More is different." Science 177.4047 (1972): 393-396.