Towards multipurpose bio-realistic models of cortical circuits - Anton Arkhipov

Some questions from the presentation:

- how do your simulation times compare to other simulators?
- Modeling LGN neurons as filters does not take into account thalamic dynamics as bursts, oscillations and so on. How does this impact the dynamics you observe?
- What is the simplest model (number of cells, etc) that led to a physiologically realistic values? On the other hand, how can this model be refined?

**1. how do your simulation times compare to other simulators?**

It is similar. Simulation times are affected by the specific hardware you run the simulation on, compilers, etc. On specific hardware and with the same settings, the times we get with our tool (BMTK) are similar to what one gets with NEURON or NEST (which are the underlying simulation engines used by BMTK).

**2. Modeling LGN neurons as filters does not take into account thalamic dynamics as bursts, oscillations and so on. How does this impact the dynamics you observe?**

Good question. We haven’t looked into this in detail, but that is certainly important. This is something to consider for the future generation of the model, and addressing this well also requires introducing cortico-thalamic feedback, which is something we’d like to do.

**3. What is the simplest model (number of cells, etc) that led to a physiologically realistic values? On the other hand, how can this model be refined?**

We didn’t try to reduce the number of cells. We tried two models – the biophysical and point-neuron – and they performed similarly on the metrics we tested. In previous work (Arkhipov et al., PLOS Comp. Bio., 2018) we saw that point-neuron model didn’t capture the magnitude of the amplification of thalamic excitatory current by cortical recurrent connections… There are probably other issues as well. In terms of refinement, the next thing we’d like to do is to improve the representation of connectivity in these models. Allen Institute is now producing Electron Microscopy-based reconstruction of connectivity in 1 mm^3 of V1, and we are very interested in creating a model using connectivity from that dataset.