Research
Jun 16, 2025
2 min

CHEESE Search: A Billion-Scale Vector Molecular Search

This is my billion-scale vector engine for 3D shape & electrostatic molecular similarity. It turns ligand-based screening into nearest-neighbour lookups—so you can *search, rank, and visualise* huge chemical spaces in seconds.
Demo Preprint
Visualisation of chemical spaces using CHEESE embeddings.

What I built:

  • Architected and trained the large AI models and led the R&D behind CHEESE.
  • Implemented a from-scratch vector database (no FAISS/Pinecone) that now indexes 30B+ vectors and is optimized for very low operating cost.
  • Designed and shipped the Python FastAPI backend as well as the Vite + React UI.

👉 Try the demo: cheese.deepmedchem.com 📄 Read the paper: CHEESE: 3D Shape and Electrostatic Virtual Screening in a Vector Space

For a bit of the backstory on my frontend journey, see My Flight from Flask to React, via Streamlit and Dash.

Visualisation of chemical spaces using CHEESE embeddings.

Check out this ACS Spring presentation about it:

Prev
Do Stereoisomers Cause Leakage? A Closer Look at the LIT-PCBA Audit
Next
What Can Neural Network Embeddings Do That Fingerprints Can’t?