Physics-based generative model of curvature sensing peptides; distinguishing sensors from binders
Problem
Distinguishing membrane curvature-sensing peptides from general membrane binders is difficult because of their chemical diversity.
Why it matters
Correctly telling sensors from binders is needed to understand and discover membrane-recognizing peptides and proteins.
What the AI did
An evolutionary algorithm combined with coarse-grained molecular dynamics (Evo-MD) resolved optimal curvature-recognizing sequences, and a physics-trained neural network enhanced identification and discovery of sensors.
Results
NN benchmarked against experimental data correctly identified known sensors and binders; numeric metrics not stated in source
Impact
Method demonstrated on known sensors/binders; broader adoption or discovery outcomes not stated in source.
Record
| Venue | Science Advances |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Niek van Hilten, Jeroen Methorst, Nino Verwei, Herre Jelger Risselada |
| Citations | 49 (OpenAlex, at import time) |
| References | 56 |
| Model or method | Evo-MD plus a physics-trained neural network |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1126/sciadv.ade8839 |
| Open access | gold |
| Keywords | curvature sensing, peptides, neural network, evolutionary algorithm, molecular dynamics, membranes |
Importance rating
incremental / niche
Narrow basic-science question (curvature sensors vs binders); abstract gives no scale or application impact beyond peptide/protein discovery.