[MRG] Implement beta event modulated ERP model #348

ntolley · 2021-06-06T18:39:11Z

This PR addresses one of the bullet points of #338. The goal is to create a function that instantiates the network from Law et al. 2021 (which was recently accepted!).

The ultimate goal is to load the network via

net = beta_erp_network() # or beta_erp_model() ?

Setting add_drives_from_params=True will simulate the main result of the paper, a prestimulus beta event inhibiting an ERP.

codecov-commenter · 2021-06-06T18:57:27Z

Codecov Report

Merging #348 (708a67d) into master (3cffe1d) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #348      +/-   ##
==========================================
+ Coverage   89.91%   89.93%   +0.01%     
==========================================
  Files          15       16       +1     
  Lines        2895     2930      +35     
==========================================
+ Hits         2603     2635      +32     
- Misses        292      295       +3

Impacted Files	Coverage Δ
hnn_core/__init__.py	`100.00% <100.00%> (ø)`
hnn_core/network.py	`91.91% <100.00%> (-1.47%)`	⬇️
hnn_core/network_models.py	`100.00% <100.00%> (ø)`
hnn_core/parallel_backends.py	`82.85% <0.00%> (-0.86%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3cffe1d...708a67d. Read the comment docs.

hnn_core/network_models.py

examples/plot_simulate_beta.py

hnn_core/network_models.py

hnn_core/network.py

ntolley · 2021-06-16T23:20:25Z

The tyranny of the incredibly awkward all_to_all_connect function is finally over. Now the network and drives are all defined through net.add_connection.

Now to the original purpose of this PR. Really all that's left for this PR is adding ERP's to the example to show how beta events impact their waveform.

hnn_core/network.py

hnn_core/network_models.py

hnn_core/network.py

ntolley · 2021-06-19T22:28:09Z

@jasmainak once I can check the example on circleCI, I'm going to push the updated tests. Besides that I am pretty happy with this PR and am fairly confident that the results from the paper are being reproduced. We can iterate on the exact wording in some followup PR's, as I imagine we'll be cleaning everything up before the 0.2 release.

ntolley · 2021-06-19T22:28:23Z

I think a very necessary discussion will be which examples we are ok with using a longer dt. Especially since this PR is adding two more simulations to the pile. At the moment it is set to 0.025 ms or 40kHz which is right in line with state-of-the-art recording setups. While a larger dt may result in less accurate simulations, most (if any) of the examples will see no appreciable change. Especially for subthreshold simulations where we're only interested in 1-100 Hz oscillations (slight inaccuracies in the ODE solver may be more impactful for spiking dynamics).

hnn_core/network_models.py

examples/plot_simulate_beta.py

jasmainak · 2021-06-20T03:31:35Z

examples/plot_simulate_beta.py

+# but modified to reflect that parameters used in Law et al. 2021.
+# Specifically, we are considering the case where a tactile stimulus is
+# delivered at 150 ms. 25 ms later, the first input to sensory cortex arrives
+# as a proximal drive, followed by one distal and a final late proximal drive.


I am reading the paper as I review the code. A couple of points:

it would be helpful to list the drives in temporal order

also perhaps explain what each drive represents. E.g., first proximal drive represents the thalamic input, then before adding the next drive add a text block explaining that the second drive represents feedback from SII region, and finally the third proximal is due to thalamocortical loop.

I know we haven't done this in the past, but we should start to think of having some examples that are "tutorial style" and others which are more "example style". I am guessing that the ones reproducing a paper will be more tutorial style

it might also be helpful to mention why the parameters are different. It's because they have been tuned to match the two conditions in the data

also perhaps explain what each drive represents. E.g., first proximal drive represents the thalamic input, then before adding the next drive add a text block explaining that the second drive represents feedback from SII region, and finally the third proximal is due to thalamocortical loop.

Along with our discussion to flush out the examples, I think this would belong better in the plot_simulate_evoked.py example when we add more thorough dialogue. I can make a brief reference to it here, but explaining the origin of the ERP is not the main purpose of this paper/tutorial.

examples/plot_simulate_beta.py

doc/api.rst

doc/whats_new.rst

jasmainak · 2021-06-20T03:45:11Z

hnn_core/network_models.py

+    return net
+
+
+def law_model(params=None, add_drives_from_params=False):


is this function covered by any tests?

Also: law_2021_model to conform with convention

Added some tests but I'm not sure how extensive they should be. For example, is it necessary to have a separate dpl.txt for each model we create? Or just a few checks like I have added to make sure things aren't breaking.

examples/plot_simulate_beta.py

Co-authored-by: Christopher J. Bailey <bailey.cj@gmail.com>

Co-authored-by: Christopher J. Bailey <bailey.cj@gmail.com> Co-authored-by: Mainak Jas <jasmainak@users.noreply.github.com>

ntolley · 2021-06-24T00:30:17Z

@cjayb @jasmainak I'm ready for the next round of reviews on this PR. I have added more details and references to the examples.

If there are glaring typos and errors then let's definitely fix them here. But for more general improvements to the commentary in the example I would prefer iterate on that in a separate PR. This is mainly so I can focus my efforts on the calcium model for the next few weeks.

jasmainak

Mostly some minor comments. I'll let @cjayb merge if he's happy

jasmainak · 2021-06-24T02:10:49Z

examples/workflows/plot_simulate_beta.py

+                         constrained_layout=True)
+net_beta.cell_response.plot_spikes_hist(ax=axes[0], show=False)
+axes[0].set_title('Beta Event Generation')
+plot_dipole(dpls_beta, ax=axes[1], layer='agg', tmin=1.0, show=False)


why not smooth here as well? For ease of comparison with the other graphs and real data?

Eh I think smoothing here really obstructs your ability to see the unique features of the beta event (sharp distal trough, long asymmetric tail). Also this better reflects the paper where no smoothing was used to plot the beta events (and a mixture of smoothed/unsmoothed for the ERP's)

jasmainak · 2021-06-24T02:28:09Z

examples/workflows/plot_simulate_beta.py

+    syn_delays_d1 = {'L2_basket': 0.1, 'L2_pyramidal': 0.1,
+                     'L5_pyramidal': 0.1}
+    net.add_evoked_drive(
+        'evdist1', mu=70.0 + stimulus_start, sigma=0.0, numspikes=1,


this is just for my knowledge. Why is numspikes=1 ?

This is how it was simulated in the current study and the original 2009 paper https://journals.physiology.org/doi/full/10.1152/jn.00535.2009

EVOKED RESPONSE INPUTS.
These were simulated as in Jones et al. (2007), with a sequence of FF input, followed by FB, followed by a reemergent late feedforward (LFF) input (schematically drawn in Fig. 9A). The timing of the input sequence was fixed as in Table 2. Each driving spike train consisted of a single presynaptic spike on each trial and the weights were distributed uniformly across the SI networks as in Table 2.

As for theoretically why it's a single spike, it's a good question. Presumably there is some stochasticity in arrival time at the cortex, and plot_simulate_evoked.py is likely the more realistic simulation of an evoked input.

I see ... I have to say I am super confused what's the interaction between sigma=0, sync_within_trials=True, numspikes=1 and how it affects the histogram plot. I'll raise a separate issue though

jasmainak · 2021-06-24T02:36:48Z

examples/workflows/plot_simulate_beta.py

+# :ref:`evoked example <sphx_glr_auto_examples_plot_simulate_evoked.py>`,
+# but modified to reflect the parameters used in Law et al. 2021.
+# Specifically, we are considering the case where a tactile stimulus is
+# delivered at 150 ms. 25 ms later, the first input to sensory cortex arrives


I think it would be nice to change the time axis so that 150 ms = 0 when plotting the dipole. ERP researchers normally name these waveforms ... N20, P50 etc. based on the latencies. I was trying to figure out which one you're trying to simulate but I'm not sure ... perhaps @cjayb has more opinions

I can do this in a hacky matplotlib hacky way for this example, but maybe this is an API issue? Like in #315 setting a tstart that is negative should allow for this sort of plotting.

okay yes, let's avoid hacky solutions. Could you raise an issue and I'll see if it can be addressed as part of #315 ?

jasmainak · 2021-06-24T02:38:19Z

hnn_core/tests/test_network.py

+    # Test succeful creation of a new cell type
+    net_new = net.copy()
+    n_cells_orig = net_new.n_cells
+    new_cell_type = {'new_type': net_new.cell_types['L2_basket']}
+    net_new.cell_types.update(new_cell_type)
+    net_new._add_cell_type('new_type', pos=[(0, 0, 0)])
+    net_new._update_cells()
+    assert net_new.n_cells == n_cells_orig + 1


rebasing issue?

Yeah the test below caused a merge conflict

I think this needs to be fixed ... this test has moved elsewhere now

hnn-core/hnn_core/tests/test_network.py

Line 253 in 3cffe1d

def test_add_cell_type():

jasmainak · 2021-06-24T02:39:27Z

hnn_core/tests/test_network.py

+        target_type = conn['target_type']
+        if src_type not in net.cell_types.keys():
+            assert set(conn['gid_pairs'].keys()) == set(
+                net.external_drives[src_type]['conn'][target_type]['src_gids'])


we should unify the two at some point ...

cjayb · 2021-06-24T13:45:16Z

Fantastic contribution @ntolley ! I'm very impressed by the tutorial, which almost doubled out CircleCI build time ;)

cjayb · 2021-06-24T13:45:50Z

@jasmainak will you pick up on your final review comments as new issues as you see fit?

jasmainak · 2021-06-24T13:56:19Z

yeah, sounds good. Final example did turn out quite nice. Great work @ntolley !

jasmainak reviewed Jun 7, 2021

View reviewed changes

hnn_core/network_models.py Outdated Show resolved Hide resolved

jasmainak reviewed Jun 7, 2021

View reviewed changes

examples/plot_simulate_beta.py Outdated Show resolved Hide resolved

ntolley mentioned this pull request Jun 9, 2021

[MRG] Fix net.add_bursty_input() weight updating #350

Merged

ntolley force-pushed the beta_erp branch from f137deb to 10e4ac7 Compare June 13, 2021 16:37

jasmainak reviewed Jun 14, 2021

View reviewed changes

hnn_core/network_models.py Outdated Show resolved Hide resolved

ntolley force-pushed the beta_erp branch from 10e4ac7 to da35bb8 Compare June 15, 2021 22:21

jasmainak reviewed Jun 16, 2021

View reviewed changes

hnn_core/network_models.py Outdated Show resolved Hide resolved

jasmainak reviewed Jun 16, 2021

View reviewed changes

hnn_core/network.py Show resolved Hide resolved

jasmainak reviewed Jun 16, 2021

View reviewed changes

hnn_core/network.py Show resolved Hide resolved

jasmainak reviewed Jun 17, 2021

View reviewed changes

hnn_core/network_models.py Outdated Show resolved Hide resolved

ntolley mentioned this pull request Jun 17, 2021

release 0.2 #365

Closed

15 tasks

ntolley force-pushed the beta_erp branch from e19c2af to e03529f Compare June 18, 2021 02:24

jasmainak mentioned this pull request Jun 19, 2021

[MRG+1] unify cell connectivity and drive connectivity #369

Merged

jasmainak reviewed Jun 19, 2021

View reviewed changes

hnn_core/network.py Outdated Show resolved Hide resolved

ntolley changed the title ~~WIP: Implement beta event modulated ERP model~~ [MRG] Implement beta event modulated ERP model Jun 19, 2021

jasmainak requested a review from cjayb June 19, 2021 22:39

jasmainak reviewed Jun 19, 2021

View reviewed changes

hnn_core/network_models.py Outdated Show resolved Hide resolved

jasmainak reviewed Jun 19, 2021

View reviewed changes

examples/plot_simulate_beta.py Outdated Show resolved Hide resolved

ntolley force-pushed the beta_erp branch from 8bcf871 to dacbaf5 Compare June 19, 2021 22:46

jasmainak reviewed Jun 20, 2021

View reviewed changes

examples/plot_simulate_beta.py Outdated Show resolved Hide resolved

jasmainak reviewed Jun 20, 2021

View reviewed changes

examples/plot_simulate_beta.py Outdated Show resolved Hide resolved

jasmainak reviewed Jun 20, 2021

View reviewed changes

doc/api.rst Outdated Show resolved Hide resolved

jasmainak reviewed Jun 20, 2021

View reviewed changes

doc/whats_new.rst Outdated Show resolved Hide resolved

jasmainak reviewed Jun 20, 2021

View reviewed changes

examples/plot_simulate_beta.py Outdated Show resolved Hide resolved

ntolley and others added 16 commits June 23, 2021 17:26

Add evoked drives

2d35409

Clean up example

e4bcce9

Parameter value updates

cfc30e9

Better seeding

821ff5f

Add amplitude comparison to example

70578ac

Better plotting

f08c8b3

Fix src_gid assignment

39bbe5c

Add src_gid test for drives

720dcd5

Doc updates

49d5ca3

Apply suggestions from code review

3e5453e

Co-authored-by: Christopher J. Bailey <bailey.cj@gmail.com>

Apply suggestions from code review

f0d2824

Co-authored-by: Christopher J. Bailey <bailey.cj@gmail.com> Co-authored-by: Mainak Jas <jasmainak@users.noreply.github.com>

Add reference to example

1134b5d

Rename to law_2021_model

41a0e3c

use functions for drives in example

278f0f7

better plotting

da6078f

Start tests for network models

9c12205

ntolley force-pushed the beta_erp branch from 81cf7ba to 9c12205 Compare June 23, 2021 21:27

ntolley added 6 commits June 23, 2021 17:35

Move example into workflows folder

350e4a1

WIP: Improvements to example commentary

642c4ee

Add spectrogram

1693af8

Only GABAa connection on distal dendrites, remove parameter arguments

0e537b5

Add published models section

c9d6f81

More detailed commentary in tutorial

708a67d

jasmainak approved these changes Jun 24, 2021

View reviewed changes

ntolley mentioned this pull request Jun 24, 2021

User defined simulation tstart #379

Open

cjayb merged commit 572b3fe into jonescompneurolab:master Jun 24, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MRG] Implement beta event modulated ERP model #348

[MRG] Implement beta event modulated ERP model #348

ntolley commented Jun 6, 2021 •

edited

codecov-commenter commented Jun 6, 2021 •

edited

ntolley commented Jun 16, 2021

ntolley commented Jun 19, 2021

ntolley commented Jun 19, 2021

jasmainak Jun 20, 2021

ntolley Jun 23, 2021

jasmainak Jun 20, 2021

cjayb Jun 20, 2021

ntolley Jun 23, 2021 •

edited

ntolley commented Jun 24, 2021

jasmainak left a comment

jasmainak Jun 24, 2021

ntolley Jun 24, 2021 •

edited

jasmainak Jun 24, 2021

ntolley Jun 24, 2021

ntolley Jun 24, 2021

jasmainak Jun 24, 2021

jasmainak Jun 24, 2021

ntolley Jun 24, 2021

jasmainak Jun 24, 2021

jasmainak Jun 24, 2021

ntolley Jun 24, 2021

jasmainak Jun 24, 2021

jasmainak Jun 24, 2021 •

edited

cjayb commented Jun 24, 2021

cjayb commented Jun 24, 2021

jasmainak commented Jun 24, 2021

		return net


		def law_model(params=None, add_drives_from_params=False):

[MRG] Implement beta event modulated ERP model #348

[MRG] Implement beta event modulated ERP model #348

Conversation

ntolley commented Jun 6, 2021 • edited

codecov-commenter commented Jun 6, 2021 • edited

Codecov Report

ntolley commented Jun 16, 2021

ntolley commented Jun 19, 2021

ntolley commented Jun 19, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ntolley Jun 23, 2021 • edited

Choose a reason for hiding this comment

ntolley commented Jun 24, 2021

jasmainak left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ntolley Jun 24, 2021 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jasmainak Jun 24, 2021 • edited

Choose a reason for hiding this comment

cjayb commented Jun 24, 2021

cjayb commented Jun 24, 2021

jasmainak commented Jun 24, 2021

ntolley commented Jun 6, 2021 •

edited

codecov-commenter commented Jun 6, 2021 •

edited

ntolley Jun 23, 2021 •

edited

ntolley Jun 24, 2021 •

edited

jasmainak Jun 24, 2021 •

edited