Similarity Search Across 40+ Trillion Molecules with CHEESE SynthonGPT2
Introducing SynthonGPT2 — search ultra-large combinatorial chemical spaces from major vendors in seconds, with built-in ADMETox filtering.
Searching Trillion-Scale Chemical Space
We are excited to announce the release of SynthonGPT2, our latest addition to the CHEESE platform. SynthonGPT2 enables similarity search across more than 40 trillion virtual molecules from combinatorial chemical spaces — covering datasets from major vendors including eMolecules, Enamine, Chemspace, XtalPi, and molecule.one.
What SynthonGPT2 Enables
- Search ultra-large combinatorial spaces in seconds
- Apply similarity thresholds for controlling chemical novelty
- Filter results by physicochemical properties and predicted ADMETox
- Download compounds for downstream workflows
- Request vendor quotations directly
Performance and Complementarity
According to user evaluations, SynthonGPT2 identifies non-overlapping distinct chemical matter compared to existing tools like SpaceLight and FTrees. Hit counts matched comparable tools via in-silico docking, while retrieval speed at trillion-scale stood out as superior.
"SynthonGPT2 finds chemical matter that other tools miss — the non-overlapping hits are exactly what you want when building diverse screening libraries."
Custom Combinatorial Spaces
Beyond the built-in vendor libraries, we offer custom-trained models for proprietary combinatorial spaces. We are currently onboarding early partners who want to make their internal virtual libraries searchable at scale.