A real decision, worked the way ShapeForge teaches: read a prediction for what it is — confidence is not correctness, and a structure is a hypothesis, not proof. You decide; the consequence is realistic; then reflect.
A real case · you decide
Help Rafi read a predicted structure honestly
Rafi — a 17-year-old using a structure-prediction tool for a research project
The idea in play: a predicted structure is a hypothesis — confidence is not correctness, and prediction is not proof.
Rafi runs a protein through a structure-prediction tool. Most of the model comes back with high confidence, but one flexible loop is flagged LOW confidence. A teammate is excited to write up a bold claim about that loop.
The tool colours each region by a confidence score. The core is high-confidence; the loop is low.
Rafi remembers that these tools predict SHAPE — and even a great shape prediction is a starting point, not an experiment.
The teammate wants to publish a strong claim resting on the LOW-confidence loop. How should Rafi treat it?
A confidence score is not a correctness score — it tells you where to trust the model and where to test it.
Rafi’s model predicts a mutation adds a hydrophobic patch to the surface. Does that PROVE the mutation causes disease?
Prediction is not proof. A structural insight is a hypothesis about function — powerful, but it still has to be tested.
Rafi presents the high-confidence core as a strong result, flags the low-confidence loop as open, and frames the mutation as a hypothesis to test. The project is more honest — and more useful — because it reads a prediction for what it is.
