CAVEAT

Open-source fragment-based bioisosteric replacement tool.

A modern reimplementation of the CAVEAT algorithm (Lauri & Bartlett, J. Comput. Aided Mol. Des. 1994), which identifies bioisosteric fragment replacements by matching 3D attachment vector geometry across a database of protein-bound ligand fragments.

The CAVEAT approach: fragment a molecule at retrosynthetically meaningful bonds, compute geometric descriptors from the 3D attachment vectors, then search a fragment database for replacements with compatible geometry — enabling scaffold hops grounded in protein-ligand interaction patterns.

---

Why Reproduce CAVEAT?

The original CAVEAT was a closed-source tool that pioneered the idea of geometry-aware fragment replacement for drug design. The concept — that fragments occupying similar 3D vector space in a binding site can substitute for each other — remains fundamental to bioisosteric design, but the tool itself has been unavailable for decades. This reimplementation makes the method:

• Transparent: every step from fragmentation to geometric matching to assembly is readable Python
• Extensible: swap in different fragmentation schemes, distance metrics, or scoring functions
• Reproducible: pip-installable with tests, usable as a library or CLI
• Scalable: SQLite + KDTree indexing handles databases of millions of fragments

How It Works

1. Fragment molecules at retrosynthetically meaningful bonds (BRICS decomposition via RDKit)
2. Compute geometric descriptors for each fragment's attachment vectors: (distance, angle₁, angle₂, dihedral) — following the original CAVEAT parameterization
3. Index fragments in a KDTree over normalized 4D descriptor space for fast nearest-neighbor search
4. Query: given a molecule and a substructure to replace, find database fragments with compatible attachment vector geometry
5. Assemble: stitch replacement fragments into the parent molecule, generating novel analogs

Quick Start

# Setup
conda env create -f environment.yml
conda activate caveat
pip install -e ".[dev]"

# Build a fragment database from ChEMBL molecules
caveat build --input data/chembl_sample.smi --db fragments.db --n-confs 10

# Query: find replacements for N-methylpiperazine in imatinib
caveat query --db fragments.db \
  --mol "CC1=C(C=C(C=C1)NC(=O)C2=CC=C(C=C2)CN3CCN(CC3)C)NC4=NC=CC(=N4)C5=CN=CC=C5" \
  --replace "CN1CCNCC1" \
  --top 10

# Database statistics
caveat info --db fragments.db

Repository Structure

caveat/
├── caveat/
│   ├── fragment.py      # BRICS fragmentation engine
│   ├── geometry.py      # 3D bond vector descriptors (CAVEAT parameterization)
│   ├── database.py      # SQLite fragment database with KDTree indexing
│   ├── query.py         # Geometric nearest-neighbor query engine
│   ├── assemble.py      # Fragment stitching into parent molecules
│   └── cli.py           # CLI: build, query, info
├── data/
│   ├── chembl_sample.smi   # ~150 diverse drug-like molecules
│   └── chembl_5k.smi       # 5,000 ChEMBL drug-like molecules
├── scripts/                 # Screening and visualization workflows
├── tests/                   # pytest suite
└── environment.yml

Dependencies

• Python 3.10+
• RDKit, NumPy, SciPy, Click
• SQLite (built-in)

References

• Lauri & Bartlett. "CAVEAT: A Program to Facilitate the Design of Organic Molecules." J. Comput. Aided Mol. Des. 8, 51–66 (1994).
• Degen et al. "On the Art of Compiling and Using 'Drug-Like' Chemical Fragment Spaces." ChemMedChem 3, 1503–1507 (2008). [BRICS]

License

MIT