Artificial Intelligence Meets Bioequivalence: Using Generative Adversarial Networks for Smarter, Smaller Trials

Anastasios Nikolopoulos, Vangelis D. Karalis

This study introduces artificial intelligence as a powerful tool to transform bioequivalence (BE) trials. We apply advanced generative models, specifically Wasserstein Generative Adversarial Networks (WGANs), to create virtual subjects and reduce the need for real human participants in generic drug assessment. Although BE studies typically involve small sample sizes (usually 24 subjects), which may limit the use of AI-generated populations, our findings show that these models can successfully overcome this challenge. To show the utility of generative AI algorithms in BE testing, this study applied Monte Carlo simulations of 2 × 2 crossover BE trials, combined with WGANs. After training of the WGAN model, several scenarios were explored, including sample size, the proportion of subjects used for the synthesis of virtual subjects, and variabilities. The performance of the AI-synthesized populations was tested in two ways: (a) first, by assessing the similarity of the performance with the actual population, and (b) second, by evaluating the statistical power achieved, which aimed to be as high as that of the entire original population. The results demonstrated that WGANs could generate virtual populations with BE acceptance percentages and similarity levels that matched or exceeded those of the original population. This approach proved effective across various scenarios, enhancing BE study sample sizes, reducing costs, and accelerating trial durations. This study highlights the potential of WGANs to improve data augmentation and optimize subject recruitment in BE studies.