Join us in welcoming Hossein Tohidi, Senior OR Specialist from SAS, as he discusses molecular structure with Graph Neural Networks. Alums and friends of the program are always welcome.
https://ncsu.zoom.us/j/93445582296?pwd=FykTWtJLHlZkhaVn6aV4AZrwVmMrF7.1
Meeting ID: 934 4558 2296
Passcode: 051656
Graph-Based LogP Prediction: Leveraging Molecular Structure with Graph Neural Networks
Molecular structures can be effectively represented as graphs, with atoms as nodes and bonds as edges, offering rich connectivity patterns crucial for property prediction tasks. LogP, the logarithm of the partition coefficient, serves as a pivotal descriptor reflecting a molecule’s hydrophobicity, influencing its behavior in biological and chemical environments. In this study, we propose a novel approach utilizing Graph Neural Networks (GNNs) to predict LogP values from molecular graphs directly.
Our method exploits the inherent graph representation of molecules to learn informative nodes autonomously, and graph features crucial for LogP prediction. Leveraging SAS Viya capabilities as well as the Torch Geometric Python package, we train our model on diverse datasets, ensuring robustness and generalization across various molecular structures. We conduct extensive experiments, benchmarking our approach against state-of-the-art methods in LogP prediction. Our results underscore the efficacy of our graph-based model in accurately predicting LogP values.
In summary, our work showcases the potential of Graph Neural Networks in elucidating graph properties for LogP prediction. By integrating graph representation learning with molecular property prediction, we offer a promising framework to enhance drug discovery and molecular design processes.
Hossein Tohidi is a Senior Operations Research Specialist at SAS, working in the SAS R&D in the Network Analytics team. He earned his PhD in Industrial and Systems Engineering from North Carolina State University in 2020, with a dissertation titled “Expert Systems for Decision Making in Multistage Healthcare Problems,” supervised by Dr. Osman Ozaltin.
