Skip to content

Latest commit

 

History

History

customized_example

README.MD

Customized example

We provide a customized example about how to generate compounds for a given pocket.

We choose the pdb 7vh8 as an example.

Steps

  1. Go to folder customized_example

  2. Download 7VH8.cif file into this folder. You can use curl https://files.rcsb.org/download/7VH8.cif > 7vh8.cif

  3. Use get_drug_center.py to get the ligand center. You can open this python script using vscode, and it will look like a jupyter notebook (due to the usage of #%%). You should get 7vh8_out.csv.

  4. Use augment_smiles.py to augment the SMILES of the reference ligand. This means that, given a SMILES string, we use different ways to represent it. Because we use sequence-based method, we can use it to increase diversity. The augmented SMILES strings are in seed_cmpd_7vh8.txt.

This is used for conditional optimization. For unconditional generation, you do not need to use it. We also prepare a common scaffold list from the training dataset, and you can consider using them. They are in misc/scaffold_smiles_top100.txt

  1. Build binary file build_bindata.sh. You should get a folder 7vh8-bin/

  2. Go back to the TamGen repo folder, and run time bash customized_example/generator-7vh8.sh. You should get the results in customized_example/7vh8-results. It requires about 30 minutes, so using tmux is a better choice.

  3. Go to customized_example. Run post_process_7vh8.py to post process the outputs. You should get one pkl and one _flatten.tsv. pkl file has all the information, while the tsv file has two columns: first column is the SMILES, and the second column is the average normalized log probability.

For you reference, we put several intermediate files in the reference_results folder.