Field Notes
How it works
A beginner-friendly guide to AlphaFold confidence, the game’s scoring system, and what the predictions actually tell us.
01
What is Fold or Flop?
Fold or Flop is a guessing game about protein structure prediction. For each round, you see a 3D protein predicted by AlphaFold — and your job is to paint each region by how confident you think the AI is in its own prediction. Then you reveal the real confidence map and score how close your intuition was.
Paint
Brush each region with your guess — High, Medium, or Low confidence.
Reveal
See the real AlphaFold confidence map overlaid on your painting.
Score
Earn full credit for exact matches, partial for adjacent guesses.
02
What is AlphaFold confidence?
AlphaFold2 is a deep-learning model from DeepMind that predicts the 3D shape of a protein from its amino-acid sequence alone. Along with each prediction it outputs a per-residue confidence score called pLDDT — short for predicted local distance difference test.
pLDDT runs from 0 to 100. A high score means AlphaFold thinks it placed that amino acid accurately; a low score means it was uncertain. Crucially, pLDDT measures the model’s own confidence — it is a useful signal, not a guarantee of experimental accuracy.
Worth knowing
Predicted structures are useful hypotheses with confidence estimates — not guaranteed experimental truth. A high-pLDDT region should still be validated experimentally before drawing clinical or functional conclusions.
03
What do High, Medium, and Low mean?
Fold or Flop maps pLDDT into three categories. These thresholds follow the same conventions AlphaFold uses in its own colour scheme.
High
AlphaFold places this region with high confidence. The backbone coordinates are likely accurate.
Medium
Moderate confidence. The rough fold is probably right, but exact atomic positions are less certain.
Low
Low confidence. AlphaFold found it hard to place this region — often because it is flexible or disordered.
04
Why can low-confidence regions still matter?
A low pLDDT score does not mean the prediction is wrong — it often means the protein itself does not have a fixed shape in that region. These are called intrinsically disordered regions, and they are extremely common in biology.
Disordered regions can serve as flexible linkers, binding sites that fold when they meet a partner, or regulatory switches. Some of the most biologically important parts of a protein have low pLDDT. Uncertainty is not the same as unimportance.
05
How scoring works
Each residue earns points based on how close your painted category is to the true AlphaFold category. Your final score is total points earned ÷ total residues × 100.
Exact match
You guessed the same category
One level off
e.g. High ↔ Medium, or Medium ↔ Low
Opposite ends
High vs Low (two levels apart)
Left unpainted
Residue was never painted
Score labels
85–100
Fold Master
65–84
Mostly Folded
40–64
Needs More Evidence
0–39
Total Flop
06
Keep in mind
Fold or Flop is a game first. Here’s what it doesn’t do:
The game collapses the full 0–100 pLDDT scale into just three categories — nuance within a band is lost.
Proteins are filtered so each game has a meaningful mix of confidence levels. Not all AlphaFold entries are playable.
Fold or Flop does not perform molecular docking, binding-site prediction, or any clinical or experimental validation.
Low confidence may reflect genuine flexibility or intrinsic disorder, not a failed prediction — these regions are biologically real, just structurally uncertain.
Ready to play?
Give it a try.
Start with a random protein and see how your intuition measures up.