The Darwin-Godel Drug Discovery Machine (DGDM): A Self-Improving AI Framework

Abstract

Despite advances such as AlphaFold and modern generative AI models, current drug discovery pipelines lack mechanisms to refine both molecules and the pipelines themselves, limiting their ability to achieve autonomous and reliable self-improvement. To address this, we present the Darwin-Godel Drug Discovery Machine (DGDM), a self-improving artificial intelligence framework that integrates generative molecular design and evolution with adaptive meta-learning. DGDM employs a dual-loop architecture: an inner loop frames molecular optimization as a Darwinian evolutionary process guided by reinforcement learning signals, where candidate molecules generated by generative AI are evolved through search and feedback; an outer loop adaptively modifies the discovery pipeline itself. Unlike the original Godel machine, which demands formal proofs of improvement--rarely attainable in practice--DGDM uses statistical validation to bound risk and ensure reliable progress. The framework is fully compatible with modern structural biology tools, including AlphaFold, and supports evaluation through docking, binding affinity prediction, and ADMET profiling. In a proof-of-concept study, DGDM improved the median binding affinity of candidate ligands from -4.457 to -5.422 kcal/mol while maintaining 100% drug-likeness and novelty. These results suggest that bounded-risk, self-improving AI can accelerate drug discovery by continuously refining both molecular design and discovery processes, extending the Godel machine principle of self-improvement into biomedical research. All code is open-sourced at https://github.com/deep-geo/DGDM.