Generative Tensorial Reinforcement Learning (GENTRL)

Supporting Information for the paper "Deep learning enables rapid identification of potent DDR1 kinase inhibitors".

The GENTRL model is a variational autoencoder with a rich prior distribution of the latent space. We used tensor decompositions to encode the relations between molecular structures and their properties and to learn on data with missing values. We train the model in two steps. First, we learn a mapping of a chemical space on the latent manifold by maximizing the evidence lower bound. We then freeze all the parameters except for the learnable prior and explore the chemical space to find molecules with a high reward.

Installation

Step 1 :

Make a new conda environment and install RDKit.

conda create -c rdkit -n my-rdkit-env rdkit

Then activate this new environment.

conda activate my-rdkit-env

Note : Make sure that the python3 version is 3.5 or higher and pip3 is installed

Step 2 :

Inside this environment install GENTRL.

cd <Path_to_GENTRL_folder>
python3 setup.py install

Step 3 : (Optional)

Making a new Kernel for jupyter notebook is recommended. For making a new kernel please follow these steps.

python3 -m pip install ipykernel
python3 -m ipykernel install --user --name rdkit_kernel

Now when you open jupyter notebook. Go to Change Kernel > rdkit_kernel

With these the installation is over and now we are ready to run the examples provided in the Repo.

Explanation

pretrain.ipynb

This notebook trains a Variational Auto Encoder (VAE) with TTLP prior distribution to encode SMILES strings onto the latent space. In this step, VAE maps the structural properties of each training molecule to a latent code.

train_rl.ipynb

This notebook applies reinforcement learning to optimize a reward function. In this step, encoder and decoder are fixed, and the model learns only the learnable prior.

sampling.ipynb

This notebook generates new molecules in the form of SMILES strings. Below we show examples of generated molecules (more samples here).