With the advancement in Artificial Intelligence, it has been possible to predict things helpful in
chemistry and biology, including drug discovery. Supervised learning has been very fruitful in
predicting the physicochemical properties of small molecules. These models are trained using a
labeled dataset, where the model learns about each data type. The dataset here has 133k stable
small organic molecules with nine heavy atoms (CONF) out of the GDB-17 chemical universe.
A message-passing neural network (MPNN) has been discussed in detail in this paper, along with
novel variations to increase the accuracy of the supervised learning models. The MPNN works
on the idea of the message, update, and readout functions that operate on different graph nodes.
These variations include revamping the input system and adding a semi-master node in the
architecture to increase the overall accuracy to create a robust model. Improvements in the model
will lead to a scalable system for choosing compounds to improve and discover the drugs.
"Efficient Integration of Molecular Representation and Message-Passing Neural Networks for Predicting Small Molecule Drug-like Properties" by Shreyas Bhat Brahmavar, Mrunmay Mohan Shelar, Revanth Harinarthini, Bandaru Hemantha Sai krishna, Nahush Harihar Kumta, Ojas Wadhwani, Raviprasad Aduri (Manuscript submimtted)