lishuya17/monn

MONN: a Multi-Objective Neural Network for Predicting Pairwise Non-Covalent Interactions and Binding Affinities between Compounds and Proteins

monn's Issues

Could I get all the packages specific version?

if the packages installs in conda?

Instructions on how to make the pre-made files in the directory "data"

Hello

In the preprocessing_and_clustering.py code, I could see the pre-made files in the directory "data" are used.
However, it seems that some of the files are generated from different pdb data.

Traceback (most recent call last):
File "preprocessing_and_clustering.py", line 322, in
idx_list = [ori_protein_list.index(pid) for pid in protein_list]
File "preprocessing_and_clustering.py", line 322, in
idx_list = [ori_protein_list.index(pid) for pid in protein_list]
ValueError: 'O85638' is not in list

Can you explain how to make the pre-made files?
There are 5 files in the data:

• mol_dict
• out7_final_pairwise_interaction_dict
• pdbbind_all_datafile.tsv
• pdbbind_protein_list.npy
• pdbbind_protein_sim_mat.npy

Thank you for the great work.

Testing MONN on individual protein-ligand pairs

Hello,

I really enjoyed the paper and am trying to implement this method to score several novel protein-ligand interactions that are not part of the training data. I have trained the model and am now wishing to test MONN on examples with this model, but can't figure out how to do this. Could you please offer guidance on how I could implement MONN to score protein-ligand binding provided I have the appropriate protein and ligand sequences?

Thank you,
Azim

How to create new dataset ?

For example pdbbind-v2019.

I know there's a Dataset_construction_protocol.txt in the folder create_dataset. But I still don't know what to put into the folder ./pdbbind_files and ./pdbbind_index/.

Can you provide a little example of these files for instructions??

Thank you very much!!

Pre-trained model available?

Hi! I have read with great interest your paper. I wonder if a pre-trained model is available? I can't find it in the repository.

Why there's not a result report for PDBBind database?

I have seen that you contains PDBBind-v2018 in the code. Why there's no result for PDBBind in the paper?

PDBBind is more reliable than DUD-E, since DUD-E decoys are constructed,right?

Thank you very much!!

Hyperparameters setting

Hi,

Thanks for sharing the codes and paper with us.

I was training a model with the default setting, python CPI_train.py IC50 new_compound 0.3.

From the message printed out during training, the val/test RMSE were around ~1.1-1.2 which seems far worse than what you presented in the papaer.

Can you please show us briefly how are your settings to get a better result?

Many thanks!

Why not consider atoms in protein instead of amino acid?

Thank you for your brilliant work in the paper.

But You consider each atom in the compound and each amino acid in the protein, I wonder whether can I consider the atoms in the protein instead of amino acid?

Since we want to know the interactions between compound and proteins, why we don't consider both the atoms within these 2 structures?

I know that the protein may be very long and the amino acid sequence is already lone enough. However, Can we cut off the protein into much smaller sequence, which contains the interaction pocket? Is there an automatic way to accomplish this task?

Thank you very much!!!

when i run cpi_train.py

RuntimeError: cublas runtime error : the GPU program failed to execute at /tmp/pip-req-build-PQwZIe/aten/src/THC/THCBlas.cu:259

Instructions on how to run the code

Hi,

Thank you for your great work, and it is of my interest. Would you mind adding some instructions on:

1. How to run the code (the command I need to use)?
2. The expected output (the results corresponding to that in the paper)?

This may greatly help the followers to know more about your work. Thank you.