Reproduce “Projecting Molecules into Synthesizable Chemical Spaces” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shitong Luo, Wenhao Gao, Zuofan Wu, Jian Peng, Connor W. Coley, Jianzhu Ma
ChemProjector, a framework from researchers at MIT, Tsinghua University, and Helixon Research, introduces a method to project unsynthesizable molecules into a synthesizable chemical space by generating valid synthetic pathways. The system achieved a 97.5% success rate in bottom-up synthesis planning for novel molecules, demonstrating its ability to create synthesizable analogs for AI-generated compounds while largely preserving desired properties like binding affinity.
Key insights
- A postfix notation for synthesis, analogous to Reverse Polish Notation, can effectively represent diverse and complex branched synthetic pathways, including multi-input reactions, while ensuring inherent synthesizability.
- A transformer-based model can successfully learn to generate these postfix notation sequences, enabling accurate bottom-up synthesis planning and the projection of theoretical molecules into practically synthesizable forms.
- Even for molecules initially deemed unsynthesizable or suboptimal, the projection mechanism can yield synthesizable analogs that preserve crucial properties (e.g., binding affinity) and docking poses, making AI-generated designs experimentally actionable.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.