Protein structure prediction with in-cell photo-crosslinking mass spectrometry and deep learning
Problem
Predicting accurate structures for challenging proteins, such as those undergoing conformational changes or with few homologous sequences, where AlphaFold2 underperforms.
Why it matters
Accurate protein structures are essential for understanding function, yet many proteins remain hard to predict.
What the AI did
AlphaLink modifies the AlphaFold2 network to incorporate experimental distance restraints from in-cell photo-leucine crosslinking mass spectrometry as anchor points, improving predictions and enabling distinct conformations.
Results
Improves on AlphaFold2 performance on challenging targets, confirmed experimentally via photo-leucine crosslinking mass spectrometry; predicts distinct conformations from provided restraints.
Impact
Demonstrates that integrating experimental data improves protein structure prediction where AlphaFold2 alone struggles.
Record
| Venue | Nature Biotechnology |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Kolja Stahl, Andrea Graziadei, Therese Dau, Oliver Brock, Juri Rappsilber |
| Citations | 161 (OpenAlex, at import time) |
| References | 53 |
| Model or method | AlphaLink (modified AlphaFold2) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41587-023-01704-z |
| Open access | hybrid |
| Keywords | AlphaFold2, protein structure prediction, crosslinking mass spectrometry, distance restraints, deep learning |
Importance rating
significant — thousands to millions affected
Extends AlphaFold2 to challenging proteins (conformational changes, few homologs), a foundational protein-science problem; experimental confirmation shown but abstract states no scale of impact.