Fold Commons

Class set · for teachers

The Wrong-Idea Index

Fifteen common wrong ideas about proteins and predicted structures. Each sits beside the right idea and the tool that makes a learner work with it, not only read about it. No new tools: this is a map of the ones we already have.

The card deck · Five readers, one fold

The index

How molecules move

  • Some people think: “Molecules move straight to the place they are needed.”

    Instead: Molecules are knocked about at random by the water around them; a meeting is a bump, and most bumps do nothing.

    Work with it: RatchetRun

  • Some people think: “A motor protein walks along its track on purpose, step after step.”

    Instead: A motor is knocked forward and back at random; the energy from ATP makes forward steps more likely, so only the average drifts one way.

    Work with it: RatchetRun

Shapes and binding

  • Some people think: “A medicine fits its pocket like a key in a lock that never moves.”

    Instead: Pockets flex. A pocket in a model is a shape in a prediction; whether and how tightly anything binds needs an experiment.

    Work with it: FlexiFit · PocketScout

Reading confidence

  • Some people think: “A predicted model is equally sure of every part of itself.”

    Instead: AlphaFold gives every residue its own confidence score (pLDDT); one model can be very sure in one place and unsure in another.

    Work with it: VanishPoint · PencilOrInk · NumberReader

  • Some people think: “If each part of a model is confident, the way the parts sit together must be right too.”

    Instead: Two domains can each be very confident while the model is unsure how they sit relative to each other. The PAE plot shows that.

    Work with it: HingeHunt · TwoIslands

  • Some people think: “A fuzzy, low-confidence part is a mistake, so it does not matter.”

    Instead: Very low confidence often marks a part that is flexible or disordered in the real protein, and that flexibility can be how the protein works.

    Work with it: DisorderDuty · PencilOrInk

What a model leaves out

  • Some people think: “A predicted model shows everything in the real structure, including its cofactors and partners.”

    Instead: An AlphaFold DB model is one chain on its own: haem, metals, other ligands and partner chains are not in it.

    Work with it: WhatsMissing · BuriedBet

Why predictions are trusted

  • Some people think: “AlphaFold is trusted because it is new and clever, not because it was tested.”

    Instead: Structure predictors are trusted as far as blind tests show: CASP scores predictions against experimental structures the predictors had not seen.

    Work with it: CASP, explained · FoldCompare

One letter, one change

  • Some people think: “AlphaMissense can tell you whether a person will get sick.”

    Instead: AlphaMissense is a prediction about a single letter change, made for research. It is not a diagnosis.

    Work with it: TolerancePaint · OneLetterLottery

Getting into a cell

  • Some people think: “Anything small can pass straight through the skin around a cell.”

    Instead: The membrane around a cell keeps most charged and water-loving things out; many pass only through channel proteins whose pore fits them.

    Work with it: PoreSort

One shape or many

  • Some people think: “A protein is rigid, like a key that never bends.”

    Instead: Real proteins move, and some switch between shapes. A model is one picture, and one gene can make more than one protein.

    Work with it: NotMarbles · SpliceSplit

Why a chain folds

  • Some people think: “All twenty amino acids behave about the same.”

    Instead: Amino acids differ: oily ones end up packed inside, away from water, and water-loving ones sit on the outside. That difference drives folding.

    Work with it: ProteinQuest · BeadFold

Inside a cell

  • Some people think: “A cell is mostly empty water, with one copy of each protein floating in it.”

    Instead: A cell is crowded: large molecules fill a large share of its volume, and each protein is present in its own number of copies, from a few to very many.

    No Fold Commons tool makes a learner work with this one yet.

Chains working together

  • Some people think: “Each chain in a complex works on its own; binding at one site changes nothing elsewhere.”

    Instead: In haemoglobin, oxygen binding at one site changes how the other sites bind. That is cooperativity.

    Work with it: GripChain

Words about molecules

  • Some people think: “A protein folds because it wants to do its job.”

    Instead: A chain folds because of physics: water, charge and packing. Purpose words describe what a fold does for the cell, not why the chain took it.

    Work with it: VerbSnap

Sources for the right ideas

The card deck

Fifteen cards, one per wrong idea, six to a printed page with cut lines. Print this section (⌘/Ctrl-P). Deal one card to each pair: they argue the wrong idea first, then turn to the tool.

How molecules move

“Molecules move straight to the place they are needed.”

Molecules are knocked about at random by the water around them; a meeting is a bump, and most bumps do nothing.

foldcommons.org/tools/ratchetrun

Shapes and binding

“A medicine fits its pocket like a key in a lock that never moves.”

Pockets flex. A pocket in a model is a shape in a prediction; whether and how tightly anything binds needs an experiment.

foldcommons.org/tools/flexifit · foldcommons.org/tools/pocketscout

Reading confidence

“A predicted model is equally sure of every part of itself.”

AlphaFold gives every residue its own confidence score (pLDDT); one model can be very sure in one place and unsure in another.

foldcommons.org/tools/vanishpoint · foldcommons.org/tools/pencilorink · foldcommons.org/tools/numberreader

Reading confidence

“If each part of a model is confident, the way the parts sit together must be right too.”

Two domains can each be very confident while the model is unsure how they sit relative to each other. The PAE plot shows that.

foldcommons.org/tools/hingehunt · foldcommons.org/tools/twoislands

Reading confidence

“A fuzzy, low-confidence part is a mistake, so it does not matter.”

Very low confidence often marks a part that is flexible or disordered in the real protein, and that flexibility can be how the protein works.

foldcommons.org/tools/disorderduty · foldcommons.org/tools/pencilorink

What a model leaves out

“A predicted model shows everything in the real structure, including its cofactors and partners.”

An AlphaFold DB model is one chain on its own: haem, metals, other ligands and partner chains are not in it.

foldcommons.org/tools/whatsmissing · foldcommons.org/tools/buriedbet

Why predictions are trusted

“AlphaFold is trusted because it is new and clever, not because it was tested.”

Structure predictors are trusted as far as blind tests show: CASP scores predictions against experimental structures the predictors had not seen.

foldcommons.org/tools/casptracker · foldcommons.org/tools/foldcompare

One letter, one change

“AlphaMissense can tell you whether a person will get sick.”

AlphaMissense is a prediction about a single letter change, made for research. It is not a diagnosis.

foldcommons.org/tools/tolerancepaint · foldcommons.org/tools/oneletterlottery

Getting into a cell

“Anything small can pass straight through the skin around a cell.”

The membrane around a cell keeps most charged and water-loving things out; many pass only through channel proteins whose pore fits them.

foldcommons.org/tools/poresort

One shape or many

“A protein is rigid, like a key that never bends.”

Real proteins move, and some switch between shapes. A model is one picture, and one gene can make more than one protein.

foldcommons.org/tools/notmarbles · foldcommons.org/tools/splicesplit

Why a chain folds

“All twenty amino acids behave about the same.”

Amino acids differ: oily ones end up packed inside, away from water, and water-loving ones sit on the outside. That difference drives folding.

foldcommons.org/tools/proteinquest · foldcommons.org/tools/proteinquest/beads

Inside a cell

“A cell is mostly empty water, with one copy of each protein floating in it.”

A cell is crowded: large molecules fill a large share of its volume, and each protein is present in its own number of copies, from a few to very many.

No tool yet

Chains working together

“Each chain in a complex works on its own; binding at one site changes nothing elsewhere.”

In haemoglobin, oxygen binding at one site changes how the other sites bind. That is cooperativity.

foldcommons.org/tools/gripchain

How molecules move

“A motor protein walks along its track on purpose, step after step.”

A motor is knocked forward and back at random; the energy from ATP makes forward steps more likely, so only the average drifts one way.

foldcommons.org/tools/ratchetrun

Words about molecules

“A protein folds because it wants to do its job.”

A chain folds because of physics: water, charge and packing. Purpose words describe what a fold does for the cell, not why the chain took it.

foldcommons.org/tools/verbsnap

Five readers, one fold

One model on the screen: The AlphaFold DB prediction of human haemoglobin beta (UniProt P68871). Five people read it five ways. Each reading takes about three minutes aloud. A handout for teacher training; print this section on one page.

  1. The dismisser. “It is not an experiment, so it tells us nothing.”

    Reader one closes the model because nobody measured it. But a prediction is a different kind of evidence, not no evidence: it comes with its own confidence score for every residue, and blind tests show how often models like it match experiments. The honest move is to name it a prediction and read the confidence, not to throw it away.

    The tool that corrects it: HonestReport · CASP, explained

  2. The pocket truster. “There is the drug site.”

    Reader two points at a dip in the surface and calls it the place a medicine goes. She has not checked whether the walls of that dip are drawn in ink (confident) or pencil (low confidence). A pocket in a model is a shape in a prediction; even with confident walls, whether anything binds there needs an experiment.

    The tool that corrects it: PencilOrInk · PocketScout

  3. The haem hunter. “Where is the haem? The model must be broken.”

    Reader three remembers that haemoglobin carries oxygen on an iron-holding haem group and cannot find it. Nothing is broken: an AlphaFold DB model is one protein chain on its own. The haem, the iron and the other three chains of the real haemoglobin are not in the file.

    The tool that corrects it: WhatsMissing

  4. The purpose reader. “The chain wants to fold like this so it can carry oxygen.”

    Reader four reads the fold as a plan. The chain has no plan. It is knocked about by water; oily side chains end up packed away from water, and the fold that results is the one that is most stable. What the fold does for the cell is a separate question from why the chain ends up in it.

    The tool that corrects it: VerbSnap · NotMarbles

  5. The diagnoser. “AlphaMissense says likely pathogenic, so this person has the disease.”

    Reader five sees a red cell on an AlphaMissense map and reads it as a diagnosis. AlphaMissense is a computed prediction about a single letter change, released for research. It does not see the person, the other copy of the gene, or anything a clinic would test. It is a reason to look closer, not an answer.

    The tool that corrects it: TolerancePaint · OneLetterLottery

Fold Commons · class set · updated 2026-10-07 · CC BY 4.0 · paper only, nothing collected.