Predicting multiple conformations of ligand binding sites in proteins suggests that AlphaFold2 may remember too much
Maria Lazou, Omeir Khan, Thu Nguyen, Dzmitry Padhorny, Dima Kozakov, Diane Joseph-McCarthy, Sandor Vajda · 2024
Read Predicting multiple conformations of ligand binding sites in proteins suggests that AlphaFold2 may remember too much by Maria Lazou, Omeir Khan, Thu Nguyen, Dzmitry Padhorny, Dima Kozakov, Diane Joseph-McCarthy, Sandor Vajda (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-11621821.txt.
Description
Maria Lazou and co-authors test in 2024 whether a famous protein folding program can show the many shapes of a drug binding site. Binding sites can sit open or closed, and a drug may fit only one form. The team sampled shapes from AlphaFold2 for sixteen proteins whose open and closed forms mark the ends of their range.
For proteins with balanced open and closed structures in the public bank, the program produced clusters of both forms, even with default settings. For proteins with lopsided records or many other shapes, it mostly failed to split into both. Subsampling the input helped only one case, and very thin inputs gave wrong structures. The authors suggest the program's success may come from remembering bank patterns rather than learning shape physics, a caution for drug hunters who rely on it.
Catalog record
| CREATOR | Maria Lazou, Omeir Khan, Thu Nguyen, Dzmitry Padhorny, Dima Kozakov, Diane Joseph-McCarthy, Sandor Vajda |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |