Your class runs a complete peptide discovery workflow — entirely in the browser.
Synthesize. Run the assay. Wait for the result. Real data — but weeks of setup, thousands of dollars in reagents, and equipment most programs never touch.
Students learn the vocabulary of discovery without running its steps. The evidence ladder is something they read about, not something they climb.
Free, browser-accessible tools — sequence retrieval, structure prediction, function annotation, experimental cross-reference — run with the same rigor, same documentation, same honesty requirements.
Public reference standard. UniProt P01275 (proglucagon). GLP-1(7–36) amide — the active 30-residue fragment. No IP constraints. Decades of published structure data — a teaching specimen with a known answer to calibrate against.
30 residues. Compact enough for AlphaFold DB lookup in seconds. Highlighted: N-terminal activation residues (Phe12/Thr13/Asp15) + C-terminal ECD-contact helix (Glu27–Leu32).
Receptor-binding residues are marked in the published literature. Students predict them computationally — then check against the experimental record. A real structure-function calibration exercise.
Pull the FASTA from UniProt (P01275). Identify the active fragment — residues 98–127. Note what the annotation flags as functionally significant before you predict anything.
Look up P01275 at the AlphaFold Protein Structure Database (alphafold.ebi.ac.uk). Download the PDB and examine the pLDDT score per residue. Which regions does the model trust — and which doesn't it?
Run the sequence through ProteInfer. Read the GO-term predictions and their confidence scores. This is function prediction from sequence alone — not function confirmation.
Pull the RCSB PDB deposited structure. Compare to your AlphaFold prediction. Where do they agree? Where do they diverge? Document the divergence — that's the finding, not the failure.
Model is confident in the local backbone geometry. Well-structured region. Good starting point for receptor-binding hypotheses. Still computational — pending wet-lab corroboration.
Generally correct backbone, some loop flexibility. Useful for structure-function reasoning — with explicit hedging. Annotate as "predicted, moderate confidence."
Disordered or flexible region. May be functionally significant — intrinsically disordered regions often mediate binding. Annotate, don't ignore.
Don't build function hypotheses here. Flag as unresolved. The correct move is to document the uncertainty, not to paper over it.
Average across the active fragment must clear 70. Below this, the prediction is too uncertain to anchor function hypotheses.
ProteInfer returns at least one Gene Ontology term with confidence above 0.5 — a predicted function, not just sequence homology.
Tool, version, date, parameters. A prediction without provenance is opinion. Provenance makes it reproducible — and that distinction is the lesson.
FASTA retrieval, structure prediction, pLDDT interpretation, experimental cross-reference — the same workflow a working discovery pipeline runs, executed by the student.
A provenanced computational characterization of a real peptide — authored by the student, confidence-labeled per the OmniRayn Foundry tier system.
The ability to distinguish what a computation says from what it means — and to label the gap between them honestly. A skill that transfers to every paper they'll ever read.
"Computational predictions, pending wet-lab; structure-function framing only; not experimental evidence and not a disease, treatment, or efficacy claim."
Free for accredited institutions, non-commercial use. Commercial use, substantiation, or spin-out escalates to the commercial Data License.