Introduction: Drug discovery remains one of the costliest and most time-intensive endeavors in the pharmaceutical pipeline. While computational methods have accelerated early-stage discovery, current pipelines remain fragmented, technically demanding, and largely inaccessible. Here we present F.A.D.E. (Fully Agentic Drug Engine), a multi-agent platform that converts natural language queries into ranked small-molecule drug candidates, substantially lowering the barrier to computational drug discovery.

Methods: F.A.D.E. employs a three-branch hierarchical architecture that adapts to the available structural data for any target of interest. When experimentally determined ligand-bound structures and binding site information are available in the RCSB Protein Data Bank, the system proceeds directly to de novo molecule generation. When a structure exists but binding site information is absent or of insufficient resolution, binding pockets are identified computationally using fpocket. In cases where no experimental structure exists, the target sequence is retrieved from UniProt, and the three-dimensional structure is predicted using Boltz-2. All three branches converge on a shared candidate generation and ranking module: drug-like molecules are generated using DiffSBDD and candidates are scored and ranked by the Quantitative Estimate of Drug-likeness (QED), synthetic accessibility (SA) score, and binding affinity.

Conclusion: We validate F.A.D.E. on two structurally distinct protein targets, which are the epidermal growth factor receptor kinase domain (EGFR), a well-established oncology target, and cellular retinol-binding protein 1 (CRBP1), a lipid-binding protein involved in retinoid metabolism. Results across both targets confirm that F.A.D.E. can reliably generate chemically tractable, drug-like hit compounds across diverse protein classes from simple natural language input to drug discovery.