Generative AI for Drug Discovery and Medical Imaging A Simulation-Based Framework for Personalized Treatment Prediction

This paper presents a reproducible, simulationdriven framework that combines gen- erative artificial intelligence with embedding-based drug representations and lightweight predictive models to demonstrate personalized treatment response estimation and syn- thetic medical-image generation. The design emphasizes modularity, interpretability, and reproducibility. The pipeline integrates synthetic 32-dimensional drug embeddings, a retrain-on-request feedforward classifier for treatment outcome probability estimation, and a compact convolutional generator that synthesizes grayscale medical-like images to accompany numeric predictions. The work is intended as an educational prototype rather than a clinically validated system. Content and architecture draw from the user’s pro- vided project document and are expanded and formalized here for academic presentation.