GAN Structure Synthesis

Our structure synthesis platform is a patent-pending approach that uses generative adversarial networks (GAN) to create 3D digital image volumes replicated from imaging scans (WO2024138171A1). Through training on microscopic images like X-ray micro-CT and FIB-SEM datasets, the platform learns physical characteristics of a product which are screened in generation of a replicated image data. It can then generate new microscopic datasets with controlled modifications, while adhering to the inherent structure of the original sample (eg a different drug amount, but with the same particle size). This approach allows scientists to digitally design new formulations without manufacturing them, exploring parameters like drug load, dispersion patterns, and particle size.

‍

For pharmaceutical scientists, this means a powerful new tool to speed up formulation design, reduce experimental costs, and support regulatory decisions. The software can be used in combination with in-silico dissolution models to simulate drug release profiles and assess how changes in formulation or processing may impact performance. By digitizing and simulating key quality attributes, structure synthesis helps teams make informed decisions earlier, with fewer physical trials, and greater confidence in final product performance.