TopScore Documentation Page

Submitting a Job

The job will be available for 7 days after the calculations are finished. After that, it will be deleted. User entered data will not be shared with any third parties.

Email

Entering an email address is optional. If you provide an email address, a link to the results webpage will be send to that email. Your email address will not be shared with any third parties.


File Upload

To upload models, they have to have a certain file format. The file format must be .zip, .tgz, or .tar.gz. Inside of the archive, there must be .pdb files or a single folder containing .pdb files. All .pdb files must contain the same atoms (they must be models of the same sequence). The maximum length of a .pdb file is 1000 residues, and only 50 models are accepted for one scoring. The archive file must not be larger than 10 MB. An example file can be downloaded here.


Example

An example results page is offered. The models used are from the sequence of hemoglobin beta.


Interpreting the Results

TopScore

TopScore is a prediction of the lDDT error (defined as 1-lDDT score) of the protein. TopScore is both a global score (the estimated error of the whole protein) and a local score (the estimated error of each residue in the model). Low scores (blue/cyan color) correspond to residues and models with a low estimated error, and high scores (yellow/orange/red color) correspond to residues and models with a high estimated error.


TopScore Single

TopScoreSingle is a prediction of the lDDT error just as TopScore. The only difference between TopScore and TopScoreSingle is that TopScore considers clustering information. This makes TopScore sensitive to the model quality of the entire model ensemble, whereas TopScoreSingle is independent of the model ensemble.


Support

If you have any questions or suggestions, please write an email to cpcweb[at]uni-duesseldorf.de.


Literature

Mulnaes, D., Gohlke, H.
TopScore: Using deep neural networks and large diverse datasets for accurate protein model quality assessment.
J. Chem. Theory Comput. 2018, 14, 6117-6126. [PDF]

Mulnaes, D., Koenig, F., Gohlke, H.
TopSuite webserver: A meta-suite for deep learning-based protein structure and quality prediction.
J. Chem. Inf. Model. 2021, DOI: 10.1021/acs.jcim.0c01202

This Server uses the NGL Viewer for visualization.
Rose, A. S., Bradley, A. R., Valasatava, Y., Duarte, J. M., Prlić, A., & Rose, P. W. (2016, July).
Web-based molecular graphics for large complexes.
In Proceedings of the 21st international conference on Web3D technology (pp. 185-186). ACM.

Rose, A. S., & Hildebrand, P. W. (2015). NGL Viewer: a web application for molecular visualization.
Nucleic acids research, 43(W1), W576-W579.