Nanoscale connectomics

Slide 1 of 12

H01 electron microscopy with object segmentation and original 2 µm scale bar

NeuroTrailblazers · Deck template

Membranes, neurites, and the connections between them.

Presenter name · Affiliation · Date

Human cortex · H01
Object segmentation over electron microscopy

H01 release · Lichtman Lab / Harvard & Connectomics at Google · CC BY 4.0
Shapson-Coe et al. (2024) · doi:10.1126/science.adk4858

Slide 2 of 12

How to use this deck

Start here

Copy neurotrailblazers-template.marp.md and rename it.
Keep theme: neurotrailblazers in the front matter.
Set footer: to the unit or module name.
Render with ./scripts/render_marp.sh, which registers the theme.

One idea per slide. If you need a second heading, you need a second slide.

Slide classes available

Class	Use for
`title`	Opener, once
`section`	Divider between parts
(none)	Standard content
`mist`	Agenda, summary, quiet slides
`figure`	Image gets the room
`stat`	One number that matters
`dark`	Emphasis content slide
`closing`	Last slide

Slide 3 of 12

Part 1

Section divider

One sentence saying what this part settles, in the reader's terms.

Slide 4 of 12

Standard content slide

A short heading identifies the subject or states a supported finding.

Lead with the mechanism. Say what happens, then why it matters for the learner's next decision.
Numbers over adjectives. "About 2 PB per mm³" beats "very large".
One callout at most. A slide with two boxes has two slides in it.
Third-level detail belongs in the speaker notes, not on the slide.

Note callout. Use for the caveat that a careful practitioner would raise unprompted: an assumption, a version pin, a boundary of the claim.

Source: unit page or DOI-pinned citation. Every figure or number slide carries one of these.

Figure description and teaching notes

Speaker notes go here. Marp keeps HTML comments as presenter notes.

Slide 5 of 12

Two columns: text against evidence

What the learner should be able to do

Decide whether a merge error or a split error is the more expensive fix for a given task.
State the proofreading budget that makes a connectivity claim defensible.
Name the null model before the metric.

What this slide does not claim

That proofreading rates transfer across datasets. They do not; calibrate per volume.

H01 neuronal soma with original scale bar

A compartment label needs evidence from the tissue.

H01 release · site render · CC BY 4.0 · Shapson-Coe et al. (2024), doi:10.1126/science.adk4858

Slide 6 of 12

Membranes and model labels

Raw EM alongside subcompartment classification, with original scale bars

The same field, before and after classification. A model label is a hypothesis to check.

H01 release · site render · CC BY 4.0 · Shapson-Coe et al. (2024), doi:10.1126/science.adk4858

Slide 7 of 12

2,000 GB

Per cubic millimetre of imaged cortex, before any reconstruction. The infrastructure unit exists because of this number.

Source: Unit 04, Volume reconstruction infrastructure; MICrONS data release.

Slide 8 of 12

Three numbers, when one is not enough

technical units, each ending in a graded lab

session kits with worksheets, rubrics and decks

127

dictionary terms with a typical value and a common confusion

Slide 9 of 12

Tables and rubrics

Level	Segmentation call	Evidence required
Not yet	Trusts the automated merge	No inspection recorded
Proficient	Inspects at branch points	Screenshot and coordinate per decision
Strong	Reports its own error rate	Blind re-check on a held-out sample

Good callout. The behaviour that separates a practitioner from someone who followed the steps.

Warn callout. The failure that looks like success: the query runs, the number is plausible, the version was never pinned.

Slide 10 of 12

Dark content slide for the turn in the argument

Use sparingly, for the one slide where the room should sit up.

Key point. Structure is evidence of organisation and constraint, not direct proof of dynamics. The connectome bounds what the circuit can do; it does not say what it did.

# Code stays short and pinned
client = CAVEclient("minnie65_public")
syn = client.materialize.synapse_query(post_ids=[root_id],
                                       materialization_version=1300)

Slide 11 of 12

Summary slide, quiet background

Settled in this session

The claim, in one sentence.
The metric, and the null it beat.
The boundary: what this does not show.

Next

Lab: produce the artifact and rubric-score it.
Read: the unit's "Go deeper" links.
Bring: one question the figure did not answer.

Slide 12 of 12

Mapping connections. Making connections.

Questions, corrections and pull requests: neurotrailblazers.org · github.com/wrgr/neurotrailblazers

Supported by the NIH BRAIN CONNECTS program. Content CC BY 4.0 unless noted.