Abstract
Developing a new drug is a lengthy and expensive process, often requiring over ten years of research and investment in billions of dollars before reaching the market. The current development in generative artificial intelligence have enabled the automated design of novel molecular structures with desirable drug-like properties. This work presents a system that leverages MolGAN (Molecular Generative Adversarial Network) combined with SELFIES (Self-Referencing Embedded Strings) to generate 100% chemically valid molecules. Generated molecules are further filtered based on Quantitative Estimate of Drug-likeness (QED), lipophilicity (LogP), and toxicity prediction. Visualization and evaluation are performed using RDKit, with an optional Streamlit interface and automated PDF report generation. The proposed system provides an efficient framework for accelerating early-stage drug discovery, offering researchers and students a powerful tool for molecular design.