Agent framework splits target planning from molecule generation in structure-based design
- CadreAI separates de novo design into a target-aware planning step and a plan-conditioned generator, with a memory module that converts docking feedback into reusable text heuristics.
- Evaluation is retrospective and computational, on CrossDocked2020 and DrugGen benchmarks, plus Boltz-2 structure-prediction checks; no experimental validation was seen.
- Generated molecules docked better than comparators while keeping reasonable drug-likeness and synthetic accessibility.
💡 Points Based on the abstract
Better docking scores on CrossDocked2020 are the weakest possible win: docking is the very signal the memory module learns from, so the loop can simply overfit the scoring function, and the benchmark's pocket overlap with training data muddies generalization. Ask for comparison against a pocket-conditioned diffusion baseline at matched compute, plus MM-GBSA or FEP rescoring on a few targets. Without that, it is a docking optimizer with an LLM wrapper.
📄 Paper · Journal of Chemical Information and Modeling (ACS) ↗ CadreAI: Agentic AI for Semantically Decoupled Protein–Drug Design and Optimization