DeepBioisostere

Deep Learning-based Bioisosteric Replacements for Optimization of Multiple Molecular Properties

Table of Contents

• Install Dependencies
• Training data for DeepBioisostere
• MMP Analysis
• Training DeepBioisostere
• Optimize a molecule with DeepBioisostere

Install Dependencies

DeepBioisostere model requires conda environment. After installing conda, you can manually install the required packages as follows:

• rdkit=2022.03.1
• matplotlib
• scipy
• numpy
• scikit-learn
• pytorch>=1.11.0

Or simply you can install the required packages by running

conda env create -f environment.yml
conda install pytorch==1.11.0 cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt
pip install -e .

This will configure a new conda environment named 'Bioiso'.

Training data for DeepBioisostere

1. If you want to re-train DeepBioisostere model without data generation, training data are available at the following Zenodo repository under the Apache License 2.0: https://doi.org/10.5281/zenodo.17804556. After downloading the dataset, go to Training DeepBioisostere.

2. Or, if you want to re-train DeepBioisostere model with data generation by MMP analysis, go to MMP Analysis.

MMP Analysis

All the necessary source code files are in:

./data

Training DeepBioisostere

After getting data for training from the Zenodo repository or manually running MMPA, you can re-train a new model by:

python ./train_main.py

Training arguments that were used to train DeepBioisostere model in our paper can be found in jobscripts/submit_train.sh.

And go to Optimize a molecule with DeepBioisostere.

Optimize a molecule with DeepBioisostere

An example for molecule optimization with DeepBioisostere can be found in ./example.py and ./example.ipynb files. The process can be divided as 1) initializing DeepBioisotere model, 2) initializing Generator class, and 3) molecule optimization.

All the trained model parameters (.pt files) can be found at ./model_save.

For the molecule optimizaiton process, we provide two options about leaving fragment selection; 1) selection by DeepBioisostere model and 2) manual selection. Below are the full descriptions about the overall process and the two options.

import os
from rdkit import Chem
from scripts.conditioning import Conditioner
from scripts.generate import Generator
from scripts.model import DeepBioisostere
from scripts.property import calc_logP, calc_Mw, calc_QED, calc_SAscore


# Setting smiles to optimize
smi1 = "ClC(Cc1c(C(Nc2c(Br)cccc2)=O)cccc1)=O"
smi2 = "Cc1ccc2cnc(N(C)CCc3ccccn3)nc2c1"

# USER SETTINGS
device = "cpu"
num_cores = 4
batch_size = 512
num_sample_each_mol = 100
new_frag_type = "all"      # one of ["test", "train", "valid", "all"]
properties_to_control = ["mw", "logp"]  # You don't need to worry about the order!

# Set model and fragment library paths (based on this project directory)
properties = sorted(properties_to_control)
proj_dir = os.path.dirname(os.path.abspath(__file__))
model_path = f"{proj_dir}/model_save/DeepBioisostere_{'_'.join(properties)}.pt"
frag_lib_path = f"{proj_dir}/fragment_library/"

# Initialize model and generator
model = DeepBioisostere.from_trained_model(model_path, properties=properties)
conditioner = Conditioner(
    phase="generation",
    properties=properties,
)
generator = Generator(
    model=model,
    processed_frag_dir=frag_lib_path,
    conditioner=conditioner,
    device=device,
    num_cores=num_cores,
    batch_size=batch_size,
    new_frag_type=new_frag_type,
    num_sample_each_mol=num_sample_each_mol,
    properties=properties,
)

# Option 1. Generate with DeepBioisostere
print("Option 1. Generate with DeepBioisostere.")
start_time = time.time()
input_list = [
    (smi1, {"mw": 0, "logp": -1}),
    (smi2, {"mw": 0, "logp": -1}),
]
result_df = generator.generate(input_list)
result_df.to_csv("generation_result.csv", index=False)
print("Elapsed time: ", time.time() - start_time)

# Option 2. Generate with a specific leaving fragment
print("Option 2. Generate with a specific leaving fragment.")
start_time = time.time()
input_list = [
    (smi1, "[*]c1ccccc1[*]", 4, {"mw": 0, "logp": -1}),
    (smi2, "[*]c1ccccn1", 12, {"mw": 0, "logp": -1}),
]
result_df = generator.generate_with_leaving_frag(input_list)
result_df.to_csv("generation_result_with_leaving_frag.csv", index=False)
print("Elapsed time: ", time.time() - start_time)