Module 22: Scientific Writing and Presentation

Deliver clear scientific talks and written summaries for technical and mixed audiences without oversimplifying connectomics evidence.

Stylized vector art: speech arcs widening from a speaker to connected listeners.

Lesson Flow

Learn

Goals and Concepts

Start with the capability target and concept set for this module.

Practice

Studio Activity

Apply the ideas in a guided activity tied to realistic outputs.

Check

Assessment Rubric

Use the rubric to verify competency and identify improvement targets.

Interactive Lab

Practice in short loops: checkpoint quiz, microtask decision, and competency progress tracking.

Talk and Q&A Checkpoint

Q1. Your result survived one null model and no sensitivity analysis. Which rung of the uncertainty ladder is honest on the slide?

The ladder runs from "we measured" through "consistent with" and "suggests" down to "we cannot distinguish" and "we did not test". One null and no robustness check is the "suggests" rung: claiming a measurement overstates what a single null buys, and claiming indistinguishability understates it. Whichever rung you choose has to be the same in the slide text, the spoken claim and the prepared answer.

Q2. You are presenting to a physiologist who has never opened an EM volume. What may change?

The invariant set is what was measured, on which data, at which version, and what the result does not establish; everything else is presentation. The third option swaps an anatomical count for a physiological claim the data cannot support, which is distortion rather than simplification. A useful test is to write the caveat twice, expert and non-expert, and check the two are logically identical.

Q3. A questioner proposes that spatial proximity explains your enrichment, and your null did not control for distance. What is the right answer shape?

An alternative-explanation question is asking one thing: does your null already rule this out? When it does not, conceding costs nothing and defending costs credibility. Ending on the specific next test is what turns "I don't know" from a full stop into a research plan, and this question was predictable enough that the answer should have been rehearsed.

Time Budget Microtask

Rehearsal of your 10-minute talk runs to 13 minutes. What is the fix?

Progress Tracker

State is saved locally in your browser for this module.

0% complete

Capability target

Deliver a 10-minute connectomics talk with evidence-linked claims, explicit uncertainty, and audience-appropriate language, then respond to questions without overclaiming. Operationally: every slide carries one claim and names the dataset version behind it, you can write down in advance the two questions you are most likely to be asked, and you have a rehearsed answer to each that ends in a next test rather than a defense.

Why this module matters

Many strong analyses fail to influence practice because communication is either too vague or too overloaded. The specific hazard in connectomics is that the data are visually spectacular and inferentially fragile at once. A rotating 3D reconstruction will hold a room; the same room will not notice that the connection counts behind it came from an unpinned segmentation that has since changed. Audiences reward the render and rarely audit the provenance, so the discipline has to come from you.

Presentation norms are also the clearest case of the hidden curriculum: strict, consequential, and almost never written down. Learners who grew up around working scientists absorbed by observation which minute may be spent on background, whether “I don’t know” reads as failure or as calibration, and what a senior person’s question is actually asking for. Learners who did not usually read their own confusion as lack of ability rather than lack of information. Stating the rules out loud costs one slide of session time.

Concept set

1) Evidence-first narrative

2) Audience adaptation without distortion

3) Q&A as scientific reasoning

4) Connectomics-specific presentation challenges

Presenting connectomics research poses unique difficulties that require deliberate design choices. Explaining electron microscopy to non-expert audiences demands analogies and visual scaffolding: show the scale progression from brain region to neuropil to individual synapses. Visualizing inherently 3D data on 2D slides requires showing both the raw EM cross-section and the 3D reconstruction of the same structure side by side so viewers can connect what is imaged to what is reconstructed. Every microscopy image should include scale bars and arrows pointing to key features, since EM images are visually unfamiliar to most audiences.

Apply the “so what?” test to every slide: if a viewer cannot articulate why a particular image, graph, or diagram matters to the argument after 15 seconds, the slide needs revision. Pair morphological images with quantitative summaries rather than relying on visual impression alone. When showing network diagrams, indicate what nodes and edges represent, how many are shown versus exist in the full dataset, and what thresholds or filters produced the visualization.

5) The time budget is fixed, so the cut list is the design

6) Uncertainty language is a graded scale

7) Question types and their answer shapes

Hidden curriculum scaffold

Give these to trainees in writing before the first practice talk.

Core workflow: technical talk preparation

  1. Build the claim tree on paper: question at the root, two or three claims, one evidence item and one caveat under each, deleting any claim you cannot attach evidence to.
  2. Write the time budget for your slot, then select the minimal slide set that preserves the inferential logic; every surviving slide must answer “which node of the claim tree is this?”
  3. Draft the one-line provenance statement for the data slide: species and region, imaging modality and resolution, segmentation pipeline, proofreading or materialization version, and any exclusion criterion.
  4. Rehearse against a timer with no audience and cut to the budget, then rehearse again with transitions spoken aloud, because the sentence carrying slide 4 into slide 5 is the one people improvise badly.
  5. Run peer critique with one narrow brief: mark every sentence where the spoken claim is stronger than the slide’s evidence.
  6. Write the two most likely questions and a three-sentence answer to each, choosing the answer shape from the question-type taxonomy.
  7. Revise with explicit uncertainty statements, checking the rung is identical in slide, speech, and prepared answer.

Choosing the depth: talk-format decision table

The same result becomes six different talks. Choose the row before drafting, because the cut list follows from it.

Format Who is in the room Methods depth to show What you cut What it costs you
Lab meeting, 30-60 min People who know the dataset better than you Everything, including QC plots and the analysis that failed Nothing; the mess is the point here One prep session, plus exposure to people who can actually check you
Departmental seminar, 45 min Neuroscientists, mostly not connectomics One slide on reconstruction and proofreading state Parameter sweeps and alternative nulls, moved to backup Several rehearsals and a real reframing, not a trim of the specialist version
Contributed talk, 10-12 min Specialists who challenge the null model first Null model and its constraints, on the result slide All background beyond 90 seconds Highest prep cost per delivered minute, against a hard stop
Poster pitch, 90 s Whoever stops walking Dataset name and version only Everything except claim, evidence, limitation Low prep, high repetition: dozens of deliveries that must each sound like the first
Public or outreach talk Non-scientists None; scale analogies instead of protocols All numbers except one memorable anchor Largest rewrite cost of any row; reuse from the research talk is near zero

60-minute tutorial run-of-show

  1. **00:00-08:00 Framing and exemplar**
    • Instructor demonstrates one evidence-linked opening slide.
    • Show two versions of the same opener, one starting with field history and one with the question. Script line: “You have sixty seconds before the audience decides how hard to listen.”
  2. **08:00-18:00 Claim tree workshop**
    • Learners draft question-claim-evidence-caveat map, capped at three claims.
    • Circulate asking one question only: “what is the evidence node under this claim, and which dataset version?”
  3. **18:00-30:00 Slide drafting sprint**
    • Build 4-slide mini-talk (problem, method, result, limitation).
    • Require the provenance line and a named uncertainty rung. Early finishers draft backup slides, not more main slides.
  4. **30:00-42:00 Peer critique round**
    • Review for clarity, caveat visibility, and claim discipline, under one narrow brief: mark every place the spoken claim outruns the slide’s evidence.
  5. **42:00-54:00 Q&A simulation**
    • Each learner answers two critique questions.
    • Assign types so everyone gets one methods challenge and one alternative-explanation question. Name the type before judging the answer.
  6. **54:00-60:00 Debrief and competency check**
    • Submit revised claim language and one uncertainty statement, then name the question you most fear. That list is next session’s material.

Worked example: repairing an opener, then surviving the question

Maya has a reciprocity result from a cortical dataset and a 10-minute slot. Her first opener:

“Connectomics is revolutionizing our understanding of the brain. Today I’ll be talking about my work on network motifs in cortical circuits.”

Move 1, replace the field claim with a question. It is about the field, not her work, it is unfalsifiable, and every other speaker will use it. In its place: “Do excitatory and inhibitory neurons in layer 2/3 connect back to each other more often than chance?”

Move 2, add the stake. A question without a stake makes the audience wonder why they should care. “Reciprocal excitatory-inhibitory pairs are the substrate most circuit models assume, and almost nobody has counted them at synapse resolution.”

Move 3, state the news. “My work on network motifs” names a topic, not a finding, and the audience calibrates how hard to listen on whether there is news to hear. “We counted them, and we find about twice as many as a degree-preserving null model predicts.”

Move 4, pre-empt the first challenge inside the opener. Anyone working on reconstruction will wonder whether false merges manufactured those pairs. Naming it first converts a challenge into evidence of competence: “The number moves with proofreading state, so I will show the enrichment on the proofread subset as well as the full volume.”

The repaired opener runs about 35 seconds. The effect is still described as enrichment relative to a named null model, not as the circuit “preferring” reciprocity.

Then the question. A senior person asks: “Isn’t your enrichment just a segmentation artifact? Merge errors create spurious reciprocal pairs.”

Defensive: “The segmentation is state of the art and error rates are low.” This asserts authority instead of evidence, names no mechanism, and invites a follow-up she cannot answer because she quoted no number.

Capitulating: “That’s a really good point, it might well be an artifact.” This abandons a result she has evidence for, and it teaches the room to discount everything else she said. Over-conceding is as much a calibration failure as overclaiming.

The answer that works: “It could be, and the mechanism is real: a merge between an excitatory and an inhibitory arbor manufactures reciprocity in exactly the direction I’m reporting. Two things make me think it isn’t the whole story. The enrichment is present in the proofread subset, where those merges are corrected, though the interval is wider because n is smaller. And an artifact of that kind should also inflate reciprocity between cell-type pairs where I see none. What I have not done is a merge-injection simulation, adding merges at a known rate to measure how much reciprocity that buys. That is the test that would settle it, and it is next.”

That answer classifies the question as a methods challenge, concedes the mechanism, offers two independent pieces of counter-evidence, states what was not done, and ends on a specific test. It runs about forty seconds, and it leaves her better off than silence: the room now knows she understands her own failure mode.

Studio activity: mini-talk and critique loop

Scenario: You are presenting one connectomics result to mixed audience members (domain experts + trainees). Use your own result if you have one. Otherwise: in a cortical EM volume, layer 4 excitatory neurons form 3.2x more synapses onto PV+ interneurons than a degree-preserving null model predicts, from 847 connections at a specific materialization version, with roughly a third of the relevant arbors proofread. Two people in the room work on segmentation; one is a physiologist who has never opened an EM volume.

Tasks

  1. Create a 4-slide mini-talk with a provenance line on the method slide and a named uncertainty rung on the result slide.
  2. Deliver it in 3 minutes against a timer you can see.
  3. Answer two audience questions, naming the question type aloud before answering each.
  4. Revise one slide and one spoken claim from the feedback, recording what changed and why.

Expected outputs

Assessment rubric

Common errors and how to recover

What this module does not cover

Content library references

Teaching resources

Quick practice prompt

Write your 60-second talk opener with the core question, one evidence-backed finding, one explicit caveat, and one sentence pre-empting the challenge you most expect. Read it aloud against a timer; if it runs past 60 seconds, cut the background sentence first.

Teaching Materials

Activity Worksheet

Learner worksheet aligned to the studio activity and rubric.

Open worksheet

Slide Source

Marp source file for editing and rendering.

course/decks/marp/modules/module22.marp.md

Related Content