Fold Commons

TolerancePaint

It is easy to think a protein is a house of cards — that any letter you change breaks it. AlphaMissense says otherwise: it scores every possible single-letter change in every human protein, and across all of them more than half come out likely harmless. But proteins differ: the three here are touchier than most, and the forgiving parts are the stretches to find. Paint which stretches you think tolerate change, then reveal the real heatmap and see how much of the protein is actually forgiving. Everything runs locally; nothing about you is collected.

This is the tool. Your painting is scored against the live AlphaMissense predictions for each protein, fetched in your browser from the AlphaFold Database. The proteins are listed below (works without JavaScript).

Your painting — N-terminus (left) to C-terminus (right); solid = tolerant, stripes = mixed, cross-hatch = sensitive

Loading the game… If it does not start, the full protein guide below still works.

How to play. Press Start. Select a region (slider bar: click, or ←/→) and mark it 1 Tolerant, 2 Mixed, or 3 Sensitive; A paints the whole chain tolerant. Press R to reveal the real AlphaMissense heatmap and score. Press Enter or Next to continue.

Why the sensitive stretches matter

The stretches that read sensitive are mostly where the chain is buried or holds something: p53's DNA-binding domain, where hotspots such as R175, R248 and R273 sit, and the interior of haemoglobin's fold around the haem. AlphaMissense scores are predictions from sequence and structure, not a diagnosis; clinicians combine them with other evidence.

What AlphaMissense says

AlphaMissense (Cheng et al., Science 2023) predicts, for every one of the ~19 possible substitutions at every position, whether it is likely benign (pathogenicity ≤ 0.34), likely pathogenic (≥ 0.564), or ambiguous (in between). Laid out as a grid — 20 rows of amino acids by the length of the protein. Across all human proteins, more than half of the possible substitutions are predicted benign (about 57%, against about 32% likely pathogenic), but one protein can be far redder: in the three here, between about a fifth and under half are predicted benign. The damaging red stripes fall on the buried core, the active site, and key interface residues. That is the point of this game: a protein is not a house of cards, and it is not a brick either; some parts are touchy and some forgiving.

The proteins

Answer key — open it after you play
ProteinGeneAccessionWhat the heatmap shows
Haemoglobin, beta chain HBB P68871 Mostly sensitive, a small tightly packed protein; the sickle-cell change is a single surface substitution (Glu→Val at UniProt position 7, the classic “β6”) that AlphaMissense rates likely benign.
PTEN (tumour suppressor) PTEN P60484 A large sensitive fraction — the active site and folded core disable easily; surface loops tolerate change.
p53 (tumour suppressor) TP53 P04637 Disordered ends tolerate almost anything; the DNA-binding core (hotspots ~175/248/273/282) does not.

Methods & about

The heatmap is the live AlphaMissense prediction for each protein, fetched in your browser from the AlphaFold Database (…-aa-substitutions.csv) and drawn as a grid; your painting is scored region-by-region against the mean predicted pathogenicity. AlphaMissense is a prediction, not a clinical verdict — it is a research tool, and a “benign” score is not medical advice. It predicts a variant’s likely effect on the protein; that is not the same as clinical pathogenicity (a damaging change can still be tolerated in the body — e.g. through gene redundancy). The class thresholds (≤ 0.34 benign, ≥ 0.564 pathogenic) are AlphaMissense’s own. Every protein note is written by hand, not generated, and science-checked. No accounts, no storage, no servers of ours (your browser talks straight to the AlphaFold Database).

Want to explore variants on a structure? Open the free AlphaFold viewer. Everything we make is free and non-profit; learn more on the home page.