Bonjour @pascalnotin,
As explained in the PR, we would like to add our model StructureDCA to the ProteinGym benchmark repository in the "DMS zero-shot" category.
I also added the model metadata in the PR so it is easier for you to include it.
Our main idea is to:
- Leverage structural information to build much smaller and more accurate Direct Couplings Analysis (DCA) models
- Build on our observations from the previously submitted RSALOR model and include the RSA into an epistatic model
The PR contains 2 versions of the model (details explained in the publication):
- StructureDCA: Structure-informed DCA
- StructureDCA[RSA]: Structure-informed DCA with RSA-based reweighting
The installation of the software can be done with the command pip install structuredca, and no other dependencies are required.
There are no files or precomputed coefficients to download.
On the global DMS zero-shot benchmark, the models StructureDCA and StructureDCA[RSA] achieve Spearman's rank correlation coefficients of 0.471 and 0.482, respectivey.
We appreciate your time and consideration.
All the best,
Matsvei 🙂
Bonjour @pascalnotin,
As explained in the PR, we would like to add our model StructureDCA to the ProteinGym benchmark repository in the "DMS zero-shot" category.
I also added the model metadata in the PR so it is easier for you to include it.
Our main idea is to:
The PR contains 2 versions of the model (details explained in the publication):
The installation of the software can be done with the command
pip install structuredca, and no other dependencies are required.There are no files or precomputed coefficients to download.
On the global DMS zero-shot benchmark, the models StructureDCA and StructureDCA[RSA] achieve Spearman's rank correlation coefficients of
0.471and0.482, respectivey.We appreciate your time and consideration.
All the best,
Matsvei 🙂