mcompchem / fep-benchmark Goto Github PK

This is a great dataset. Thanks for putting it together!

You might consider using Zenodo to create a DOI for this data. This will allow other researchers to recognize your contribution and cite you when they use this data in their work. Zenodo will capture the state of this repository at one time, and if you make changes, correct typos, or add data later, you can make a new release and have a separate DOI assigned to the new version.

We have been very happy with using Zenodo to track data in our project. You can see an example here. I'd be happy to work with you to get this configured. It should take under an hour.

In case any of this data is sensitive, I wanted to point out that files that are uploaded and later deleted are still available in the git history. It is possible to remove this data from the history using some git commands (here is one method)

You can view the commit history of this repository here https://github.com/MCompChem/fep-benchmark/commits/master

It looks like you removed some data from the cmet folder, and I would be happy to work with you to clean that up if the data is sensitive.

Thank you so much for this resource! It seems that the ligand and results files have missed being committed for the cmet system, would it be possible to provide them?

Thanks!

Dear Developers,
I found the results here are not consistent with those in your paper https://pubs.acs.org/doi/pdf/10.1021/acs.jcim.0c00900.
For example, the pairwise RMSE of tnks2 reported in paper is 2.02 kcal/mol , but when I use data here https://github.com/MCompChem/fep-benchmark/blob/master/tnks2/results_edges_5ns.csv , I get 1.20kcal/mol. I wonder if there is some misleading when processing data.

I came to find the publication via the repo, but the readme still says manuscript in preparation. I don't think I can open a branch on this repo but adding the link to your work on in the readme would be helpful!

https://pubs.acs.org/doi/abs/10.1021/acs.jcim.0c00900

Hi,

The IC50 for CHEMBL3265032 is given as 10000 nm in the SDfile, which would give a binding FE of -6.8 kcal/mol, but the data tables give -11.06, which appears to be consistent with the figures as well. Also the value of -11.06 is consistent with the article, which gives an IC50 of 7.7 nM for this compound. See Compound 68 in Table 4.

Regards,
Mike

PS Credit where due: Mike Potter at VeraChem is the one who spotted this.

Hello!

Looking in the .csv for CDK8, it looks like there are repeated values

https://github.com/MCompChem/fep-benchmark/blob/master/cdk8/results_20ns.csv
i.e. -11.22 and -11.46

Looking at the primary citation, it looks like these are 4 nm or 5 nm affinities (or in the last decimal place quoted to) --- is it possible that these compounds are bottoming out the assay? If so, it is possible that these could be higher affinity than reported? And if so on top of that, does that mean that they can be used for free energy calculations or not? Or should they be handled differently, i.e. assessing categorical success, or using larger error bars for some compounds.

cc. @dominicrufa @jchodera @dfhahn

Hello, thank you for this resource!

If it's not too much trouble - could you provide the graph of the FEP calculations that were carried out to obtain these values?

Starting ligand -> ending ligand or a two-column CSV file or any other human-readable format would be fine.

eg:
CHEMBL1078774 -> CHEMBL1093087
...