About the unverified marks on this pageFigures tagged unverified are PACE-local estimates that could not be confirmed against any published source, so they are labelled rather than quietly presented as fact. Every other factual claim on this page — statutes, standards codes, gene biology, primary-source citations — was independently checked; the full claim audit, including what came back wrong, lists every claim with its source.
Try it Tuesday · 90 minutes
Run this next week
90 minutes. Minutes 0–15: each student writes, in one sentence a non-scientist could follow, the biological question behind a variant or structure they choose. Minutes 15–50: they run the explain-to-a-layperson prompt and generate an interactive explainer, then use the embedded variant-explainer demo to check whether a toggled change is described correctly. Minutes 50–90: a cross-disciplinary read — a student from another field reads the explainer and marks every term left undefined and every claim stated without a source. The deliverable is the explainer plus the outside reader's marked-up copy; it grades whether the student actually understands the pipeline they ran.
The full assignment
Make it 3D print
A printed structure from a record the student selected and defended: a binding site, a mutation, a fold. Print two variants, wild type and mutant, and let the difference be something you can feel rather than something you are told.
Build it AI chat · no code
An interactive explainer of the student’s own analysis, written for a non-specialist. Explaining a pipeline to a general audience exposes whether the student understands it, which a working script does not.
Try the demo · runs offline
Bioinformatics demo
Interactive demo — “variant-explainer”. It cannot be printed; run it in a browser at https://el3vate.vercel.app/demos/variant-explainer/index.html
This demo runs entirely offline against canned data — no keys, no network, safe on conference wifi.
4 weeks, realistically
The 4-week version
- Wk 1Each student picks a structure or dataset and writes the biological question in one sentence a non-scientist would understand, and identifies the wild-type and mutant structures to print.
- Wk 2Submit the print file (wild-type and mutant) at the start of the week so fabrication runs in parallel, then run the analysis.
- Wk 3Build the public-facing interactive explainer. Peer review is done by a student from a different discipline who has to say what they did not follow.
- Wk 4Present both. Assessment weight sits on whether the non-specialist reviewer understood the finding.
Starter prompt & what the model actually did
The prompt, run for real
Build a single self-contained HTML page that explains this analysis to someone with no biology background: [describe the analysis and the finding]. Use an interactive element so the reader can change one input and see the result change. Explain every term the first time it appears. Do not use an acronym without spelling it out. Then tell me the three places where you had to simplify in a way that a specialist would object to.
One correction before the code: F508del is not a single-nucleotide variant. It is the deletion of three DNA letters (nucleotides), which removes one amino acid (a protein building block) at position 508. A single-nucleotide variant changes just one letter. I built the page around what F508del actually is.
```html
<!DOCTYPE html>
<html lang="en">
<head><meta charset="utf-8"><title>CFTR & Cystic Fibrosis</title>
<style>
body{font-family:system-ui;max-width:640px;margin:2rem auto;padding:0 1rem;line-height:1.5}
button{padding:.5rem 1rem;margin:.25rem;cursor:pointer}
#out{border:1px solid #ccc;border-radius:8px;padding:1rem;margin-top:1rem;background:#f7f7f7}
.lbl{font-size:.85rem;color:#666}
</style></head>
<body>
<h1>How one change causes cystic fibrosis</h1>
<p><b>DNA</b> is the instruction book in your cells. <b>Genes</b> are chapters.
The <b>CFTR gene</b> (Cystic Fibrosis Transmembrane conductance Regulator) tells the cell how
to build the <b>CFTR protein</b> — a tiny gate in the cell's surface that lets
<b>chloride</b> (part of salt) move out. That flow keeps the mucus lining your lungs thin and slippery.</p>
<p>Press a button to change the gene and watch the result.</p>
<button onclick="show('normal')">Healthy gene</button>
<button onclick="show('f508del')">F508del gene (three DNA letters deleted)</button>
<div id="out"></div>
<script>
const data={
normal:{aa:"…Ile-Phe-Gly-Val…",fold:"Folds correctly",gate:"Reaches the cell surface — gate works",
result:"Thin mucus. Healthy.",c:"#137333"},
f508del:{aa:"…Ile- [missing] -Gly-Val…",fold:"Misfolds (wrong 3-D shape)",
gate:"Cell destroys it before it arrives — no gate",
result:"Thick, sticky mucus clogs the lungs → cystic fibrosis.",c:"#b3261e"}
};
function show(k){const d=data[k];
out.innerHTML=`<p class=lbl>Protein building blocks made:</p><b>${d.aa}</b>
<p class=lbl>Folding:</p><b>${d.fold}</b>
<p class=lbl>Gate at cell surface:</p><b>${d.gate}</b>
<p class=lbl>Outcome:</p><b style="color:${d.c}">${d.result}</b>`;}
show('normal');
</script>
</body></html>
```
The interactive element: the two buttons switch the gene, and every downstream box (building blocks → folding → gate → outcome) updates.
Three places a specialist would object to my simplification:
1. **"No gate at all" is too absolute.** F508del protein is not simply destroyed — small amounts can reach the surface, and the channel that does arrive gates poorly and is unstable. Modern drugs (correctors and potentiators) exploit exactly this residual function, which my binary healthy/broken framing erases.
2. **I collapsed distinct defect classes.** F508del causes misfolding, premature degradation, *and* faulty gating simultaneously — different problems that different drugs target. I showed one arrow where the biology has several.
3. **One gene copy is not the whole story.** You have two copies of CFTR; disease requires both to be affected, and the ~2,000 known variants produce a wide severity spectrum. My page implies one change flips one switch, hiding inheritance and variability.
- Corrected the prompt's premise: 'F508del is not a single-nucleotide variant'The strongest possible outcome — the model caught the user's scientific error (a three-nucleotide deletion, not a SNV) and built around the correct biology. Students should note AI can be right when the user is wrong.
- Self-listed three specialist objections (residual function, distinct defect classes, two gene copies)It named exactly the over-simplifications the assignment asks students to find, modeling the acknowledged-simplification skill.
- Binary 'no gate at all' framing in the codeThe model admits this is too absolute; a student should see how the healthy/broken toggle erases the residual-function biology that real drugs target.
- Unchecked: the residue context '…Ile-Phe-Gly-Val…' around position 508Every other biological claim in this output was verified — the three-nucleotide deletion, the misfolding and premature degradation, the residual surface function correctors exploit, the two gene copies, the 2,000-plus known variants. This one was not: nobody compared that four-residue stretch against the actual CFTR sequence. It is consistent with the commonly cited I507-F508-G509-V510 run, but consistent is not checked, and the page will not claim otherwise. Two minutes in a sequence viewer closes it, which is precisely the move the assignment is teaching.
- Confident causal compression: 'Thick, sticky mucus clogs the lungs → cystic fibrosis'One arrow stands in for a multi-step pathophysiology — the unwarranted-confidence failure, here mild and self-flagged.
Assessment
Rubric
| Criterion | Weight | What it assesses |
|---|---|---|
| Scientific accuracy | 30% | The explainer's claims about the variant and its consequence are correct and would survive a specialist's read. |
| Lay comprehensibility | 30% | An outside reader understands the finding; every term is defined on first use and no acronym is left unspelled. |
| Acknowledged simplification | 25% | The student names where they simplified in a way a specialist would object to, rather than hiding it. |
| Source discipline | 15% | Claims are attributed or flagged as uncertain instead of asserted with false confidence. |
What this replaces
Swapping it into a real course
Replaces
The computational lab report or annotated pipeline script.
What is lost
The detailed methods documentation and reproducible code write-up.
What is gained
Explaining the analysis to a non-specialist exposes whether the student understands it — something a working script can hide — and the outside reader's confusion is a sharper diagnostic than a methods section a peer skims.
Where AI is bad at this
The failure your students should catch
A model states gene and variant function with unwarranted confidence and no source — asserting that a variant 'causes' a phenotype, inventing a mechanism, or collapsing a defect class into a single switch, in authoritative clinical language it cannot back. It also silently mishandles the biology, conflating a deletion with a point mutation or ignoring that two gene copies matter, and its fluency makes the error invisible to a non-specialist.
Budget & logistics
What it costs to run
- Instructor prep2 hours
- Class time90 minutes
- Per-student cost$0 for the Tuesday version; roughly $6–12 per student in filament unverified for wild-type and mutant structure prints in the four-week build.
- Fabrication file dueFirst day of Week 2 — submit the print file (wild-type and mutant) at the start of the week. PACE quotes 7–10 business days (up to 14 calendar days), so it clears before the week-4 presentation.
- Calendar dependencyAnalysis and explainer need no lead time. The structures are selected in week 1, so the print file goes in at the start of week 2 and the up-to-14-calendar-day turnaround clears before the week-4 presentation.
Three sizes
Scale it to the time you have
One session
The 90-minute explainer-plus-outside-read session; no fabrication.
4 weeks
The seeded four-week plan: pick a structure and question, run the analysis and submit print files, build and peer-review the explainer, then present both.
One semester
A full project — a self-selected dataset, a reproducible analysis, printed wild-type and mutant structures, and a public-facing explainer peer-reviewed for comprehension by non-scientists.
Reserved · live build
This space is intentionally empty. During the Day 8 session it will be filled in live — fill the liveBuild field in content/bioinformatics.json and rebuild.
Tell us what happened
Run it, then say how it went
If you try this — the 90-minute version, the prompt, any part of it — send back what you tried and what happened, especially anywhere the model was confidently wrong. That is the material the next session is built from.
Email feedback on Bioinformatics →
Opens a draft in your mail client, already addressed and titled. No form, no account, nothing to sign up for.