Deep Generative Inverse Design of Biofunctional Polymer Coatings Using Conditional GANs

Abstract

Designing biofunctional surface coatings for biomedical implants requires balancing multiple biological objectives, including cell viability, antibacterial activity, and controlled drug release. Conventional experimental optimization of such multi-objective systems is time-intensive and explores only a limited portion of the feasible design space. Here, we present a constraint-aware conditional generative adversarial network (cGAN) framework for the inverse design of polymer-based coating compositions conditioned on desired biological performance targets. The model was trained on a curated dataset combining experimentally derived and synthetically augmented compositions, and evaluated using independent surrogate predictors of biological response. The forward predictive model demonstrated high predictive performance, achieving R² values ranging 0.90 and 0.94 across the evaluated biological endpoints, while generated candidates satisfied compositional feasibility constraints and achieved reduced mean distance-to-target relative to baseline sampling and optimization strategies. All evaluations are conducted in silico within a surrogate modeling framework; therefore, results should be interpreted as computational prioritization of candidate formulations rather than experimentally validated performance. Overall, this study establishes a reproducible computational foundation for constraint-guided inverse design of multifunctional biomaterial coatings and provides a structured pathway toward future experimental validation.