Integrating Trained Neural Networks with RDKit

RDKit is a collection of cheminformatics and machine learning tools written in C++ and Python. Which is more important, it allows you to work with many representations of chemical data and has a power to extract almost each chemical descriptor from the data you have. An added functionality to this would be to be able to find chemical insights from the molecules. This includes being able to find properties of the molecules which are required quite often.
This project aims to add using trained neural networks for property prediction of molecules. Quantum Chemical calculations / Molecular Dynamics / Monte Carlo Simulations for finding out molecular properties have proven to be extremely time consuming and alternate “Neural Network Potentials” have been proposed instead. I aim to add property prediction NNPs like ANI which provide accuracy comparable to Quantum Mechanical Approaches like Density Functional Theory / MD / MC but do the job in a fraction of the time to the existing ML package of RDKit. Having a dependable and accurate tool as a part of the library would remove a significant bottleneck from the pipeline.