PDBBind Dataset

[hkmztrk]

I failed to find the exact data set that's used in in this study (https://arxiv.org/pdf/1703.10603.pdf)
Could you please help me with that?

Thanks!

[lilleswing]

@hkmztrk are you looking for the featurized datasets with the atomicconv featurizer? These scripts will get you the featurized core and refined sets.
https://github.com/deepchem/deepchem/blob/master/contrib/atomicconv/acnn/core/get_acnn_core.sh
https://github.com/deepchem/deepchem/blob/master/contrib/atomicconv/acnn/refined/get_acnn_refined.sh

If you are looking for the raw dataset (protein/ligand co-crystallized pairs) you can use this script.
https://github.com/deepchem/deepchem/blob/master/examples/pdbbind/get_pdbbind.sh

[hkmztrk]

@lilleswing I was looking for the raw data set, thank you! It contains binding affinities of the pairs, right?

[hkmztrk]

Hello again,

Sorry if this is a trivial question!
I downloaded the raw data set, and I realized that (in general set for instance) there are three types of binding affinity types (Ki, Kd, IC50). I remember reading Ki binding is predicted, so how do you deal with other types, do you convert them to Ki?
In PL general data there are total 11,987 complexes, but only 3650 complexes with Ki.

Thanks in advance.

[evanfeinberg]

Great question! Since they are closely related measures with the same units we have been treating all three as identical when training many models. However this is not universally recommended. For example an agonist will have a very poor IC50 but a strong Kd

…

On Sun, Aug 20, 2017 at 7:10 AM hkmztrk ***@***.***> wrote: Hello again, Sorry if this is a trivial question! I downloaded the raw data set, and I realized that (in general set for instance) there are three types of binding affinity types (Ki, Kd, IC50). I remember reading Ki binding is predicted, so how do you deal with other types, do you convert them to Ki? In PL general data there are total 11,987 complexes, but only 3650 complexes with Ki. Thanks in advance. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#715 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABre1gSRdg_-iRliANdQK-aiSavADQRtks5saD5vgaJpZM4OrCuo> .

[hkmztrk]

Thanks for the clarification @evanfeinberg, about the units, they also seem to differ (nM, uM, mM etc.), which unit did you use? Does it matter which unit we convert them all? (e.g. all of them nM)

[rbharath]

All training is done in log-units (log-Kd/log-Ki/log-IC50). These are directly provided in the labels files for PDBBind.

[hkmztrk]

@rbharath what I mean is molar concentration unit, for instance, all three are Kd samples, but they have different molar concentrations (uM, mM, nM), do you perform a conversion among those?

3p9t 2.02 2011 6.20 Kd=0.63uM 3p9t.pdf (TCL)
4ad2 2.10 2012 6.20 Kd=625nM 4acz.pdf (GLC-IFM)
4qsw 1.80 2014 1.68 Kd=21mM 4qsw.pdf (38T)

[evanfeinberg]

Yes, all are converted to units of Molar (i.e., 0.63 uM --> 6.3e-7 M) and then -log10(value) is performed.

…

On Tue, Aug 22, 2017 at 3:10 AM hkmztrk ***@***.***> wrote: @rbharath <https://github.com/rbharath> what I mean is molar concentration unit, for instance, all three are Kd samples, but they have different molar concentrations (uM, mM, nM), do you perform a conversion among those? 3p9t 2.02 2011 6.20 Kd=0.63uM 3p9t.pdf (TCL) 4ad2 2.10 2012 6.20 Kd=625nM 4acz.pdf (GLC-IFM) 4qsw 1.80 2014 1.68 Kd=21mM 4qsw.pdf (38T) — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#715 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABre1sZ1rRznjpPeI7YY1AnrGz7uy-17ks5saqkagaJpZM4OrCuo> .

[hkmztrk]

thank you @evanfeinberg !

[evanfeinberg]

HTH!

…

On Tue, Aug 22, 2017 at 8:53 PM hkmztrk ***@***.***> wrote: thank you @evanfeinberg <https://github.com/evanfeinberg> ! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#715 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABre1qgmRLLvHvAPCK8DeiQODOUQ-ANwks5sas9QgaJpZM4OrCuo> .