novana: A Python repository from Martinmejia

Introduction

Novana (Novelty Analysis) is a cheminformatics tool that allows decomposing molecules into their scaffolds and shapes in milliseconds. The method extends the functionalities of Bemis-Murcko scaffolds (J. Med. Chem. 1996, 39, 15, 2887–2893) by introducing more granularity on how atoms and bonds are treated during the decomposition. The implementation of these extra functionalities was inspired by the analysis described by Wills and Lipkus in ACS Med. Chem. Lett. 2020, 11, 2114-2119. Novana can be used to cluster molecule data sets across multiple levels of generalisation - as an alternative to similarity methods. The tool can be used for the analysis of data sets or the creation of train/validation sets for machine learning.

Here is an example of structural decomposition using different flavours of Novana also compared with the Bemis-Murcko decomposition in RDKit (MurckoScaffoldSmiles()):

Method

For a given SMILES input (molecule):

A scaffold is generated by removing recursively all terminal atoms until only atoms bonded with at least two different atoms are retained (e.g., cycles and chains). Non-rotatable terminal atoms (e.g., oxygens of a carbonyl/sulfonyl group) are removed. The bad valences of the retained atoms are fixed by adding or removing hydrogens according to some heuristics which can be found in valence.py. The charges of atoms that have been modified by the algorithm are neutralised, whereas the charges of other atoms in the input remain unchanged.
A Bemis-Murcko scaffold from RDKit is also generated to show that the RDKit implementation, in contrast to that in Novana, preserves non-rotatable terminal atoms.
A scaffold with starred heteroatoms can be created to generalise the atom composition of a scaffold. Heteroatoms are replaced by stars (*) and carbons and hydrogens are preserved.
A shape with retained atom types can be generated as a decomposition of a scaffold by preserving all atoms and converting all bonds into single bonds.
A shape is finally produced as a further decomposition by converting all its atoms into single bonded, non-aromatic, neutral carbons.

Novana also deals with mixtures automatically by extracting the largest fragment containing rings and using it as input to the decomposition. If no structures with rings are found in the input SMILES, Novana throws an error.

How to install the tool

Novana can be installed from pypi (https://pypi.org/project/novana).

pip install novana

Usage

# Example of input SMILES
from rdkit import Chem
smiles = "COC(=O)N1CC2(C1)CS(=O)(=Nc1cc(C)c3c(Nc4ccc(F)cc4O[C@H](C)C(=O)NCC(F)(F)F)ncnc3c1)C2"
Chem.MolFromSmiles(smiles)

# Decompose into scaffold
from novana.api import scaffold_from_smiles
scaffold_from_smiles(smiles)

# Decompose into shape with atom types
from novana.api import shape_from_smiles
shape_from_smiles(smiles, retain_atom_types=True)

# Decompose into scaffold with starred heteroatoms
from novana.api import scaffold_from_smiles
scaffold_from_smiles(smiles, generalise_heteroatoms=True)

# Decompose into shape
from novana.api import shape_from_smiles
shape_from_smiles(smiles)

# Molecule and default scaffold and shape can also be obtained efficiently in one run
# (This function can be particularly useful for processing large sets)
from novana.api import molecule_scaffold_shape_from_smiles
mol, sfl, shp = molecule_scaffold_shape_from_smiles(smiles)

License

Distributed under the terms of the MIT license. Novana is free and open-source software.

For developers