Note

This page corresponds to a Jupyter notebook you can try out yourself.
(The original version is here.)

---

Simulation Tutorial

This notebook shows how to estimate the delay and amplitude response between primary brain regions for a given neuron. In general, the delay within a compartment tends to be isopotential meaning a region to region response matrix is a reasonable characterization of the output response of a neuron.

The electrical simulation uses a simple linear passive model using a network of resistors and capacitors based on a skeleton representation of the neuron. Users can specify specific membrane resistance and capacitance values and cytoplasm resitance.

Note: You must install some additional dependencies before using simulation functions:

conda install -c conda-forge ngspice umap-learn scikit-learn matplotlib

Setup neuron model

• create neuprint client using the public hemibrain dataset

• build a neuron spice model for a given body id

from neuprint import Client

TOKEN = "" # <--- Paste your token here
           # (or define NEUPRINT_APPLICATION CREDENTIALS in your environment)

client = Client("neuprint.janelia.org", "hemibrain:v1.2.1", TOKEN)

# simulate a PEG neuron (same type as in hemibrain paper)
from neuprint import NeuronModel
nm = NeuronModel(449438847)

Estimate neuron timing

Run simulation for the given neuron

Specify the number of inputs per brain region (randomly selected) which will be tested. A current is injected at each input site one at a time and delay and amplitude response is measured.

res = nm.simulate(2)

Examine the response results

compute_region_delay_matrix reports the delay between regions where the row index is the input brain region and the column names represent the output region. A delay (ms) and amplitude (mV) are returned. They are averages over all the response for a given region input-output pair.

Note: the results to synapses that are not in brain regions (“none”) are shown for completeness. But those output responses are not necessarily co-located and cannot be treated as an ROI.

delay, amp = res.compute_region_delay_matrix()

# delay matrix in ms
delay

	EB	LAL(R)	PB	none
CRE(R)	2.074233	0.427321	3.208169	2.539196
EB	1.258467	2.456816	3.166434	2.483917
FB	1.657618	2.120017	1.880404	1.058834
LAL(R)	2.356702	0.329966	3.474082	2.809558
PB	3.143595	3.550282	0.437841	0.713781
none	2.590528	3.010865	0.676969	0.154951

# amplitude in mV
amp

	EB	LAL(R)	PB	none
CRE(R)	0.153407	0.489385	0.122230	0.134334
EB	0.181624	0.143877	0.121702	0.134000
FB	0.160437	0.149854	0.149690	0.167537
LAL(R)	0.146963	0.749892	0.117411	0.128987
PB	0.123022	0.116236	0.416325	0.261497
none	0.134309	0.126698	0.283109	0.311731

Plot response from one brain region to others

Each dot represents the response from each input from the specified input region (LAL(R) here) to every output in the neuron (each output brain region is indicated).

res.plot_response_from_region("LAL(R)")

Decompose neuron connections by electrical domain

Use simple linear passive delay progragation through the neuron to determine distinct domains for representing the neuron. The decomposition can be used to derive ROIs using electrical delay.

Simulate the delay ‘distances’ between parts of the neuron

This function checks propagation delays throughout the neuron based on the number of specified points. The points are randomly chosen even between inputs and outputs. In principle, any points could be chosen.

res = nm.estimate_intra_neuron_delay(num_points=100)

Show delay propogation closeness between simulated points

This plot should show whether there are distinct electrical domains for the neuron. The primary ROI labels are provided to show how well the points correspond to the annotated regions.

res.plot_neuron_domains()

Partition the neuron into different domains based on simulated points

Cluster the simulated points and map a partition id to each synapse in the neuron. The function returns input and output connections broken down by these domain ids.

summary, detailed_io, plot = res.estimate_neuron_domains(3, plot=True)

# input / output table broken down by domain id from partitioning
summary

	io	partner	weight	domain_id	rois
0	output	696682163	147	0	[EB]
1	output	974300015	97	1	[PB, none]
2	output	858587718	88	0	[EB]
3	output	858587718	72	2	[LAL(R)]
4	output	1228692168	66	0	[EB]
...	...	...	...	...	...
720	input	5813057963	1	0	[EB]
721	input	5813062805	1	0	[CRE(R)]
722	input	5813061383	1	1	[PB]
723	input	5813070465	1	0	[EB]
724	input	5813080269	1	0	[EB]

725 rows × 5 columns

# domain id for each x,y,z location
detailed_io

	type	x	y	z	roi	partner	domain_id
0	post	21510	26069	20648	EB	387364605	0
1	post	23679	24857	20538	EB	387364605	0
2	post	22135	26177	20494	EB	387364605	0
3	post	21613	25974	20679	EB	387364605	0
4	post	23870	23196	20455	EB	387364605	0
...	...	...	...	...	...	...	...
7092	pre	23030	24895	20167	EB	1002507131	0
7093	pre	23621	24908	20429	EB	1002507131	0
7094	pre	22615	24507	19708	EB	1002507131	0
7095	pre	22992	24888	20123	EB	1002507131	0
7096	pre	23278	24606	20178	EB	1002507131	0

7097 rows × 7 columns

plot