A few typos and questions

[mkgilson]

Great paper! Here are a few items for your consideration...

page 20, 5 lines above section 7: practise-->practice

page 22, second bullet: cyro-->cryo

page 22, first bullet in Affinity data ("single source"): can this be defined more specifically? In particular, my thought is that data on the same protein in several papers from the same lab/company can be viewed as "single source", but I'm curious if you would agree.

Same list, maybe define "functional assays"; for example, would this exclude enzyme inhibition assays?

page 19, right column: great point about not computing correlations for pairwise relative binding free energies; but isn't it okay to use a set of relative binding FEs to rank compounds and compute Kendall's tau? Granted, the dynamic range of such sets may tend to be too small, but it may not always be, and this entirely avoids the problem of inversion of data points highlighted here

What do you think of the centered RMSD as a metric for relative binding FEs? It's a way of comparing results with experiment that does not rely on choosing a specific reference ligand, with the problems that generates. We used this in D3R and no one seemed to find it problematic. (See eq 1 here: https://link.springer.com/content/pdf/10.1007/s10822-016-9946-8.pdf)

[ppxasjsm]

Great paper! Here are a few items for your consideration...

Thank you for the input! Much appreciated.

page 20, 5 lines above section 7: practise-->practice

page 22, second bullet: cyro-->cryo

Will fix the typos.

page 22, first bullet in Affinity data ("single source"): can this be defined more specifically? In particular, my thought is that data on the same protein in several papers from the same lab/company can be viewed as "single source", but I'm curious if you would agree.

Same list, maybe define "functional assays"; for example, would this exclude enzyme inhibition assays?

Excellent point.

page 19, right column: great point about not computing correlations for pairwise relative binding free energies; but isn't it okay to use a set of relative binding FEs to rank compounds and compute Kendall's tau? Granted, the dynamic range of such sets may tend to be too small, but it may not always be, and this entirely avoids the problem of inversion of data points highlighted here.

I agree with your point here, will add a comment along these lines.

What do you think of the centered RMSD as a metric for relative binding FEs? It's a way of comparing results with experiment that does not rely on choosing a specific reference ligand, with the problems that generates. We used this in D3R and no one seemed to find it problematic. (See eq 1 here: https://link.springer.com/content/pdf/10.1007/s10822-016-9946-8.pdf)

You mean RMSE here right? I think we are using something similar too. Making an explicit statement about this would make sense will update the text.

[mkgilson]

Thanks for the replies! Yes, I meant RMSE.