Artificial intelligence generates novel 3D printing formulations
Moe Elbadawi, Hanxiang Li, Siyuan Sun, Manal E. Alkahtani, Abdul W. Basit, Simon Gaisford
Abstract
Formulation development is a critical step in the development of medicines. The process requires human creativity, ingenuity and in-depth knowledge of formulation development and processing optimization, which can be time-consuming. Herein, we tested the ability of artificial intelligence (AI) to create de novo formulations for three-dimensional (3D) printing. Specifically, conditional generative adversarial networks (cGANs), which are generative models known for their creativity, were trained on a dataset consisting of 1437 fused deposition modelling (FDM) printed formulations that were extracted from both the literature and in-house data. In total, 27 different cGANs architectures were explored with varying learning rate, batch size and number of hidden layers parameters to generate 270 formulations. After a comparison between the characteristics of AI-generated and human-generated formulations, it was discovered that cGANs with a medium learning rate (10−4) could strike a balance in generating formulations that are both novel and realistic. Four of these formulations were fabricated using an FDM printer, of which the first AI-generated formulation was successfully printed. Our study represents a milestone, highlighting the capacity of AI to undertake creative tasks and its potential to revolutionize the drug development process.