(460f) Computer Aided Drug Formulation Design with Image-Based Microstructure Modeling

In pharmaceutical formulation development, microstructures, such as porosity and the distribution of active ingredient particles, play an increasingly important role. Three-dimensional microscopic imaging, specifically X-Ray Microscopy (XRM) and focused ion beam scanning electron microscopy (FIB-SEM), provides insight into these microstructures at unprecedented resolution. However, acquiring such images can be challenged by cost, time, and the availability of high-end imaging facilities and required expertise. Even when the images are acquired at desirable quality, there can be uncertainty that the acquired data is a representative elementary volume (REV).

An image-based microstructure modeling framework will be presented in this talk. XRM images, FIB-SEM images, or correlative images using both, are processed with an artificial intelligence-based image analysis engine using supervised machine learning and generative adversarial networks. A matrix of quantitative descriptors such as volume fractions, particle size, particle dispersion, porosity, and transport properties are computed. A computer aided drug formulation (CADF) model can be numerically generated that matches these image-based descriptors. From the CADF model, full 3D microstructures corresponding to variants of the formulation can be numerically generated.

Figure 1 demonstrates one example CADF model for an amorphous drug formulation. Two dimensional SEM images, Figure 1a, were acquired and segmented with a hybrid supervised machine learning and deep learning method (Figure 1b). Mathematical descriptors are extracted and divided into three groups. The first group is used as the generator input in the GAN model, and the second group as the discriminator input in the GAN model. The third group serves as the validation control group to drive the GAN model iteratively toward a microstructure model with desirable properties, Figure 1c. Once the model is validated, three dimensional microstructures of new formulations can be generated numerically. For example, Figure 1d-1f illustrates five new formulations with porosity 10%, 20%, 30%, 40% and 50%. A broader porosity range, at desirable intervals, can be covered. All quantitative descriptors can be computed and correlated with release behavior without making a new sample.

The CADF model provides a numerical framework to study the sensitivity of these quantitative descriptors to drug formulation parameters (such as drug loading, particle size of the active pharmaceutical ingredients, milling conditions, particle orientation alignment, and more) and under various processing conditions (such as hot melt extrusion temperature, polymer properties, compaction force, and more). The numerical model allows a formulation scientist to rapidly traverse multi-variable parameter space, and narrow down to lead formulation parameters and optimal processing conditions. When in vitro release test or dissolution data is available, release simulations can then be conducted on the CADF model to validate, before applying the CADF model to iteratively predict release profiles under different parameters.

The proposed framework has broad applications in formulation development and solid dosage forms, and in material science in general, where microstructures play a critical role.