Multi-task model for prediction of C-H functionalization reactions

Original publication can be found here

This repository contains the trained model of the publication listed above. It also contains the code to make predictions or deploy a web-based GUI.

Setting up enviorment using docker

Build the docker container from site_selectivity directory:

docker build -t sites .

To run one prediction use this command and replace with your molecule of interest:

docker run -t sites bash -c "source activate sites && python selectivity/site_selectivity.py --smiles <smiles>"

To run the web GUI there are two options:\n

• Using docker-compose where using localhost is hard coded run: docker-compose up -d and to kill container: docker-compose down -v

• Using docker run where you can specify the host and the port

docker run -p <host port>:<port> -d sites bash -c "source activate sites && python web/run.py --host <host> --port <port>"

If you are running the default, direct your web browser to localhost:5000 to access the GUI.

Setting up environment using conda

conda create -n sites python=3.7 && conda activate sites

OR

conda create -n sites python=3.7 && source activate sites

Then install dependencies

conda install -n sites -c rdkit rdkit && conda install -n sites pip pip install -r requirements.txt

If installing requirements manually, be sure that the version of tensorflow is 1.1x. Newer versions enable eager execution by default and the code will not run.

Webapp deployment

You can run predictions from a web GUI. To start it run:

python web/run.py --host <IP of host> --port <port>

The default host is localhost and the default port is 5000.

Command line deployment

Run a prediction and output to the terminal. In home folder run (predictions here for toluene):

python selectivity/site_selectivity.py --smiles Cc1ccccc1

Data processing pipeline

Unfortunately, the data used to train this model is not publicly available. However we are putting together a representative pipeline with USPTO open source data.