Ig-VAE: Generative modeling of protein structure by direct 3D coordinate generation
Problem
Designing class-specific protein backbones, such as immunoglobulins, for structure-based design tasks like binding interfaces was lacking generative methods.
Why it matters
Computational antibody and binder design underpins therapeutics for diseases such as COVID-19.
What the AI did
A variational auto-encoder directly generates 3D coordinates of immunoglobulin backbones; combined with Rosetta via latent space sampling it produced a computational model of a SARS-CoV2-RBD binder.
Results
not stated in source
Impact
Motivates a new paradigm treating backbone design as constrained optimization in a generative model's latent space.
Record
| Venue | PLoS Computational Biology |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Raphael R. Eguchi, Christian A. Choe, Po‐Ssu Huang |
| Citations | 141 (OpenAlex, at import time) |
| References | 64 |
| Model or method | Ig-VAE (variational auto-encoder) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1371/journal.pcbi.1010271 |
| Open access | gold |
| Keywords | protein backbone generation, variational autoencoder, immunoglobulin, Rosetta, SARS-CoV-2 binder |
Importance rating
significant — thousands to millions affected
Generative backbone design for immunoglobulins enables binder design (SARS-CoV2-RBD binder shown), a genuinely new capability for therapeutics design; no scale stated.