AmelieSchreiber
commited on
Commit
·
7cf9c57
1
Parent(s):
d56fafb
Update README.md
Browse files
README.md
CHANGED
@@ -28,7 +28,7 @@ is also an attempt to make deep learning techniques like LoRA more accessible an
|
|
28 |
of simple models and techniques. Moreover, since most proteins still do not have a predicted 3D fold or backbone structure, it is useful to
|
29 |
have a model that can predict binding residues from sequence alone. We also hope that this project will be helpful in this regard.
|
30 |
It has been shown that pLMs like ESM-2 contain structural information in the attention maps that recapitulate the contact maps of proteins,
|
31 |
-
and that single sequence masked language models like ESMFold can be used in atomically
|
32 |
AlphaFold2 on proteins up to about 400 residues long. In our approach we show a positive correlation between scaling the model size and data
|
33 |
in a 1-to-1 fashion provides competative and possibly even SOTA performance, although our comparison to the SOTA models is not as fair and
|
34 |
comprehensive as it could be (see [this report for more details](https://api.wandb.ai/links/amelie-schreiber-math/0asqd3hs)).
|
|
|
28 |
of simple models and techniques. Moreover, since most proteins still do not have a predicted 3D fold or backbone structure, it is useful to
|
29 |
have a model that can predict binding residues from sequence alone. We also hope that this project will be helpful in this regard.
|
30 |
It has been shown that pLMs like ESM-2 contain structural information in the attention maps that recapitulate the contact maps of proteins,
|
31 |
+
and that single sequence masked language models like ESMFold can be used in atomically accurate predictions of folds, even outperforming
|
32 |
AlphaFold2 on proteins up to about 400 residues long. In our approach we show a positive correlation between scaling the model size and data
|
33 |
in a 1-to-1 fashion provides competative and possibly even SOTA performance, although our comparison to the SOTA models is not as fair and
|
34 |
comprehensive as it could be (see [this report for more details](https://api.wandb.ai/links/amelie-schreiber-math/0asqd3hs)).
|