Combining Predictive Models and Reinforcement Learning for Tailored Molecule Generation
Document Type
Article
Publication Date
1-1-2024
Abstract
This study introduces a three-fold methodology that harnesses the capabilities of generative artificial intelligence (AI), predictive modelling, and reinforcement learning to craft customized molecules with desired properties. The model seamlessly integrates deep learning techniques with Self-Referencing Embedded Strings (SELFIES) molecular representation, constructing a generative model for producing valid molecules. In the framework, a graph neural network model was used to predict molecular properties and a combined Variational Autoencoder and reinforcement learning model to generate new molecules with specific attributes. Experimental data from a surfactant study validates the effectiveness of the framework. This innovative approach not only streamlines molecular design for surfactant systems but also anticipates transformative advancements in diverse scientific and industrial domains.
Publication Source (Journal or Book title)
Computer Aided Chemical Engineering
First Page
3037
Last Page
3042
Recommended Citation
Nnadili, M., Okafor, A., Akinpelu, D., Olayiwola, T., & Romagnoli, J. (2024). Combining Predictive Models and Reinforcement Learning for Tailored Molecule Generation. Computer Aided Chemical Engineering, 53, 3037-3042. https://doi.org/10.1016/B978-0-443-28824-1.50507-X