BindingDB: Amino acid sequences and y_pred values

[jchen634]

Dear DeepPurpose team,

Thank you so much for providing this powerful tool! I am new to this field and would like to ask some basic questions if you don't mind.
Here, I am trying to predict the binding affinity between the small compound of interest and the short peptide sequence using the Transformer+CNN_BindingDB model.

While, even if I change the input of the amino acid sequence to either very long >100 a.a. or very short, just one amino acid, the y_pred value is always around 5.3~5.7, which is not a significant difference. The output is pKd, correct? If yes, then the bin
Also, I have changed the y value, but it does not affect the y_pred at all. Is it normal, and what is the function of y here?

Thank you in advance.

[kexinhuang12345]

Hi thanks for the question, I think it is likely that random sequence is OOD for the bindingdb dataset, so it just gives me some random/average pK score. yeah, y is for when the ground truth is available to calculate metrics. It will not change the predictions. by the way, in the screenshot, you are using DeepDTA model which is CNN+CNN.

[jchen634]

Hi Kexin,

Thank you so much for the explanation and clarification.
I just copied the code from DEMO/Transformer+CNN_BindingDB.ipynb, and added the last cell for prediction. I am sorry if I used it wrong.
https://github.com/kexinhuang12345/DeepPurpose/blob/master/DEMO/Transformer%2BCNN_BindingDB.ipynb

I was wondering if you have any suggestions on the prediction of the binding between small compounds and peptides, since most of the models I found are for proteins.

Many Thanks,
JC