CNS*2020 F2: Using evolutionary algorithms to explore single-cell heterogeneity and microcircuit operation in the hippocampus

Please use this topic to discuss the second featured oral session.

F2: could you confirm that you ‘imposed’ a theta rhythm when looking at deep and superficial differences? (CCK and PV changes)

…as opposed to the functional pyramidal heterogeneity part, where theta was there autonomously in model, right?

Was there any heterogeneity in the types of neurons in terms of their firing properties? In other words, were they all Type 1 as shown or was there a mix of Type 1 and Type 2?

Also, another small question - when you did ‘no inhibition’ in removing PV and CCK cells (with presumably the imposed theta - see previous question), there would still be inhibition from the other inhibitory cell types, right?

More questions here from the presentation:

  • how do the “genes” correspond to different intrinsic properties of the pyramidal cells? I don’t understand this mapping exactly, thanks!

  • The choice of the objective (cost) function to select and generate the individuals at each step is critical in genetic algorithm, especially when there are several target outputs to satisfy. What is the mathematical expression of your objective function ? How much fine tuning did it require ?

Is there any specific function for NMDA vs AMPA currents in the excitatory inputs? Vice versa, GABA-B vs GABA-A?

  • A very basic question: The 2nd slide shows an image where Theta-modulated neurons are represented. The colour codes the types of neurons; what do the width of each ring indicate?

  • could you confirm that you ‘imposed’ a theta rhythm when looking at deep and superficial differences?

  • Is there any difference in the correlation between morphology and phase with apical vs basal tree complexity (number of branches) or other morphological features eg length?

  • was there any heterogeneity in the FI Curves of the neurons. In other words, were they all T1 as shown, or was their a mix of Type 1 and 2?

“Could you confirm that you ‘imposed’ a theta rhythm when looking at deep and superficial differences? (CCK and PV changes) as opposed to the functional pyramidal heterogeneity part, where theta was there autonomously in model, right?”

Hmm… No, I would say that theta is always imposed because the input rhythmicity makes the pyramidal cell behave in that oscillatory regime. The paradigm was always the same in both cases, looking for the impact of the factors in the basal condition (all synaptic factors as in the Genetic Algorithm output), and the CCK/PV exploration (where the number of synapses from each population was lowered): theta was achieved by making the imputs have their corresponding theta-modulated temporal distributions


A very basic question: The 2nd slide shows an image where Theta-modulated neurons are represented. The colour codes the types of neurons; what do the width of each ring indicate?

The width changes so if there is an overlap you can see that there are many inputs at a time :slight_smile:


The choice of the objective (cost) function to select and generate the individuals at each step is critical in genetic algorithm, especially when there are several target outputs to satisfy. What is the mathematical expression of your objective function ? How much fine tuning did it require ?

Here it is how I computed it:

MSE = 0
# Evaluating the I/O curve
for i in xrange(len(amps)):
    if ( (freqs[i]<targetMin(amps[i])) or (freqs[i]>targetMax(amps[i])) ):
        MSE += min([ (targetMin(amps[i]) - freqs[i] )**2, 
                     (targetMax(amps[i]) - freqs[i] )**2 ] ) 
    else:
        MSE += 0
    print MSE

where:

  • amps: is the input of the input/output curve
  • freqs: is the output of the input/output curve
  • targetMin and targetMax: define the target curve
  • MSE: how did I compute the error

so it was minimum squared error compared with the size of the target area closer to the output frequency

1 Like

“Was there any heterogeneity in the types of neurons in terms of their firing properties? In other words, were they all Type 1 as shown or was there a mix of Type 1 and Type 2?”

I’m not sure what are you referring to, what are the differences between type 1 and type 2? In that regard I can say that I saw similar spiking patterns, there was no bursting, maybe something similar to a complex spike I recall seeing some times…


“Also, another small question - when you did ‘no inhibition’ in removing PV and CCK cells (with presumably the imposed theta - see previous question), there would still be inhibition from the other inhibitory cell types, right?”

No, in that plot there was not inhibition at all. It’s a good idea though to see what are the differences between not inhibition at all, and all inhibition except CCK and PVbc! But I didn’t run that :sweat_smile: I suppose as they are perisomatic they have more weight in deciding spike timings


“how do the “genes” correspond to different intrinsic properties of the pyramidal cells? I don’t understand this mapping exactly, thanks!”

So here quoting “genes” is actually a very good thing! The term “genes” comes from the algorithm, that is based evolutionary algorithms. In this algorithm, each gene represents a certain parameter on the model, so for example if “gene 1” was representing the gA parameter, in the HH equations the value of the ion channel density of A potassium channels would be the value of that “gene”. Basically each individual was a list of parameters of the model that I wanted to be fitted by the algorithm


“Is there any difference in the correlation between morphology and phase with apical vs basal tree complexity (number of branches) or other morphological features eg length?”

OH! This is nice! I didn’t test that, but actually it’s a cool idea. We did performed some analysis with the morphological profile (like number of branches withing some moving interval across all the apical axis) from deep and superficial reconstructed morphologies and we didn’t find statistical differences, so we didn’t continue on that direction

1 Like

ok - thanks for the clarification

MANAGED BY INCF