17 Yield prediction

0. Relevant packages

RDChiral

RDChiral is a wrapper for RDKit’s functionalities for reaction handling, that improves stereochemistry handling. This package will allow us to extract reaction templates from a reaction dataset, which are a standard way of encoding transformation rules.

RDChiral then also lets us apply the reaction template to a target molecule, to discover the reactants that will afford the target molecule under the given transformation.

Learn more from the code and the paper.

1. Obtaining the atom mapping

To obtain the atom mapping of a reaction, you can go to this site and paste your reaction SMILES. The application will then show you the mapped reaction smiles, as well as some visualization options, including:

The atom mapping of the reaction: which atoms in the reactants correspond to each atom in the products.
The attention maps: What the underlying model is computing, that is the conection between each pair of tokens.

NOTE: This model is also accessible through a programming interface. For this, follow the instructions here.

TODO:

Get a reaction and mapped rxn
Get dataset of molecules
Obtain possible reactant sets
Get all reactant sets that can react like this, visualize

import pandas as pd

df = pd.read_csv

#! pip install rdkit rdchiral
! mkdir data/
! curl -L https://www.dropbox.com/sh/6ideflxcakrak10/AADN-TNZnuGjvwZYiLk7zvwra/schneider50k -o data/uspto50k.zip
! unzip data/uspto50k.zip -d data/

  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100    17    0    17    0     0      8      0 --:--:--  0:00:02 --:--:--     8
100   276  100   276    0     0     94      0  0:00:02  0:00:02 --:--:--     0
  0     0    0     0    0     0      0      0 --:--:--  0:00:21 --:--:--     0

from utils import load_data, visualize_chemical_reaction

train_df, val_df, test_df = load_data()

1. Reaction templates

Let’s take as an example the following coupling reaction.

rxn_example = train_df.iloc[5,0]

visualize_chemical_reaction(rxn_example)

To extract the reaction template, use the `extract_template` function from utils.py

A reaction template describes a general transformation of some type. It describes what bonds form and break in a transformation, as well as the chemical environment of these bonds.

from utils import extract_template

tplt_example = extract_template(rxn_example)

# A reaction template looks like this
print(tplt_example)

Now we can use this reaction template. Use the `apply_template` function from utils.py

If we use it on the same product, we should get the same reactants as above.

# Apply the extracted template to the product above.
from utils import apply_template, visualize_mols

prod_1 = rxn_example.split('>>')[1]
pred_reactants = apply_template(tplt_example, prod_1)

# This is the result of applying the template.
visualize_mols(pred_reactants[0])