Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Ibis
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
---
|
| 4 |
+
|
| 5 |
# Ibis
|
| 6 |
|
| 7 |
+
Ibis is a flow-matching model that generates conformational ensembles of proteins at Cα resolution from their amino acid sequence. It is the model studied in [Ablating Design Choices for Protein Ensemble Generators Across Structural Regimes](https://arxiv.org/abs/2611.05055) (Quinn & Bent, 2026). "Ibis" is our internal name for it, the paper does not use it.
|
| 8 |
+
|
| 9 |
+
## Checkpoints
|
| 10 |
+
|
| 11 |
+
| File | In the paper | Training data for disordered proteins |
|
| 12 |
+
|---|---|---|
|
| 13 |
+
| `base/model.ckpt` | Base | CALVADOS coarse-grained simulations |
|
| 14 |
+
| `mix/model.ckpt` | Fragment 50% | half CALVADOS simulations, half IDRome-o fragment-assembled ensembles |
|
| 15 |
+
|
| 16 |
+
Both models are also trained on AlphaFold DB structures of folded proteins and on CALVADOS simulations of multi-domain proteins. They take sequences of up to 510 residues. In the paper's evaluation against experimental data, `mix` agreed better with NMR chemical shifts and slightly worse with SAXS than `base`.
|
| 17 |
+
|