Open for (neuro)science tutorials: New from the Allen Cell Types Database

Thanks for attending the webinar or viewing the video of the Open for (neuro)science tutorial: New from the Allen Cell Types Database, with @Luke_E, Stephanie Seeman, and @gouwens! This webinar demonstrated how to use the patch-seq and synaptic physiology datasets.

Please feel free to chime in here with additional questions or follow-up questions to those asked at the live webinar. You can find the recording of this webinar and information about other webinars in this series here .

Question: Is there a link for the jupyter notebook?

Kaitlyn and Nathan: We’ll post a link as a reply to this thread.

Question: I work as a computational modeller. I was curious to know if there was a way by which one could map transcriptomic data, such as (relative) gene expressions to their equivalent (relative) conductance in the cell?

Nathan: That’s a really interesting but complex question. The measured expression of a particular gene in a single cell is fairly noisy, so you’d have to account for that when relating it to other properties. Also, we don’t know the specifics of how transcriptomic levels correspond to levels of, say, an ion channel in the cell membrane. But those are questions we (and others) are very interested in pursuing.

Question: How do you recognize axon? Are you always sure that this is an axon and not the dendrite? Particularly, in inhibitory cells. Was it easy?

Nathan: We have a team that does the morphology reconstructions, and they identify whether the process is an axon or dendrite based on the original 63x image stack. They are fairly distinguishable based on things like diameter, the presence of boutons, etc.

Follow-up question: Thank you! Is axon’s diameter smaller than dendrite’s?

Nathan: Yes, that is correct.

Question: What brain regions does your cell type dataset include?

Kaitlyn: Different datasets within the Allen Cell Types Database include different brain regions. Detailed information is available in the metadata for each dataset. Patch-seq currently includes mouse primary visual cortex and human middle temporal gyrus.

Question: will it work also if you don’t get a nucleated patch? I am asking because I was curious to know if this technique could be also reliably used in in-vivo experiment or if it’s currently limited to slice electrophysiology

Nathan: You can see data on the difference in yield between nucleated patch vs not in our biorxiv preprint (see, e.g., Figure 4)

Luke: Yes, but the technique is most robust, especially the transcriptomics data, when you get a nucleated patch.

Question: Have you been able to inquire about chromatin modifications (ATAC-seq / Cut&Tag) through patchseq once you have isolated the nuclei?

Luke: We have not done that, although one would expect the isolated nuclei to be amenable to these techniques.

Question: How big the slices are usually to get a reliable 3d reconstruction. And is there a protocol available?

Kaitlyn: The detailed protocol is available here:

Nathan: The slices for Patch-seq are typically 350 microns thick, so there are some inevitable cutting artifacts. We also mostly use morphological features in the x- and y-dimensions for analysis, rather than those perpendicular to the cutting plane.

Steph: We use the same slice thickness and all the same techniques for synaptic physiology as patch-seq, to try to get full axons. In our dataset is also information about the depth of the cells and how long the axons were before they were truncated, if they were, which helps calculate connection probability.

Question: What patch solution do you use to record facilitation and depression simultaneously ? Do you use Cesium methanesulfonate ?

Kaitlyn: The detailed protocol is available here: Synaptic Physiology Experimental Methods -

Steph (live): I would say go look at the protocol directly. We don’t use a cesium-based solution, all our solutions are potassium-based.

Question: What was the prevalence of autapses in your data set?

Steph (live): We did not look at autapses, self-synapses. They’re a bit challenging to find since they’re very close to the presynaptic spike, so we haven’t looked at that yet. All our data is available through the download so if you want to go dig into that yourself, we encourage you to do so!

Question: Does the nuclear extraction take a place soon after stimulation? I am wondering whether immediate early gene activation is captured with this protocol

Nathan (live): Basically right after we finish the stimulation, we start the extraction. The extraction is fairly slow, to make sure we get the nucleated patch, but it’s only a couple minutes at most. Probably some immediate gene activation going. We think we see that more in this dataset than in the dissociated cell dataset. It doesn’t affect the mapping – it’s not typically the genes used to define the cell types.

Question: How many nuclei can you extract per slice?

Nathan (live): It depends on the slice itself and what regions we’re targeting. For humans, we try to get as much data as possible – a couple dozen cells, so they’re not overlapping and they’re distinguishable. In mouse, it’s more constrained because the region isn’t as large. We’re interested in the visual cortex in mouse for this, so we’ll get a little more cells per slice. One thing to clarify, we’re pulling a nucleated patch, we’re not just getting the nucleus, we’re getting cytoplasm too, it just helps us to get the whole nucleus and all its transcriptional machinery.

Question: How often do you get all eight cells in synaptic physiology experiments, the maximum for simultaneous recordings?

Steph (live): We pretty routinely get 4 cells, that’s our median, but we’ve gotten about 50 octopatches, or all 8 cells. It’s rare, but we’re getting better at it.

1 Like

I tried very hard to convert my ABF files to NWB and use the feature extraction both from ABFconverter imported from ipfx and from pynwb. It seems there is no detailed script I can follow to convert my files and I always get errors like my “outputFeedbackChannel” is not correct. I would appreciate if someone can share a py or jupyter script showing how to exactly convert ABF to NWB for further analysis,

We don’t have a detailed script for converting from ABF files, since that’s not something we do at the Allen Institute (we output our files to NWB using the MIES software that runs in Igor Pro). If your goal is to create NWB files, you may have better luck contacting the NWB developers - there is a Slack community for NWB that is linked on their homepage where they give help and advice.

However, if you want to use IPFX for feature analysis, you don’t necessarily have to convert your files into NWB first. There are parts of IPFX that just use numpy vectors for time, voltage, etc. as inputs, so you may be able to use those directly if you can get your ABF files into Python in some form. For example, you could do something like this to analyze spikes:

from ipfx.feature_extractor import SpikeFeatureExtractor

# Assumes that you have variables t (time stamps), v (membrane voltage) and i (stimulus current waveform) loaded from your ABF files

# Extract information about the spikes
ext = SpikeFeatureExtractor()
results = ext.process(t=t, v=v, i=i)

# Plot the results, showing two features of the detected spikes
plt.plot(t, v)
plt.plot(results["peak_t"], results["peak_v"], 'r.')
plt.plot(results["threshold_t"], results["threshold_v"], 'k.')

The example Jupyter notebook is now available. We have also put a link to the notebook (and the video) on the Multimodal Characterization in Mouse Visual Cortex for future reference.