T3The Loop
00Prototype_001 · THING3
You have 288 tests. Predict which approach will do best, then see how three ways of choosing compare. About six minutes.
Who this is for
For investors, partners and buyers who need to understand the approach without a background in enzyme science.
Best on a laptop or tablet. A simplified illustration: .
01The real-world version
They speed up chemical reactions. This one, from a yeast, breaks down fats, and factories use it to make medicines, food ingredients and materials.
Real structure: lipase B from Candida antarctica (Protein Data Bank 1TCA). Folding motion simplified.
0 possible versions, from changing just three of its 317 parts. Most work worse; a rare few work better.
Each experiment costs time and money. Here, your budget is three trays of 96 tests.
A simplified picture, not a physical place: versions one swap apart sit side by side, so a tested neighbour is a useful clue, though one swap can still change a lot. You learn the map one test at a time.
02Make a call
How sure are you?
GuessingVery sure
03Run them
Three copies of the map from before, one for each approach. The best version is hidden somewhere on it.
Now compare the predictions with actual performance. The grey landscape is revealed for comparison: the search methods cannot see it.
No predictions at all: the baseline.
One ranking (blue) from a model trained on other enzymes. Lab results don’t update it.
Predictions made for this enzyme (green), updated with each tray’s lab results.
04One more question
Imagine you’re running the third approach. The second tray of lab results doesn’t match what the computer model predicted.
How sure are you?
GuessingVery sure
05In your words
06What happened
In this example, feeding results back into the model helped it choose better candidates.
An illustration on an invented landscape, not real results or a benchmark. It doesn’t show that feedback always wins, and the second and third approaches also started from different predictions.
For the studio
Did the explanation work? Here’s how we could check.