What are we discovering today?

Describe the protein target or PDB ID and your criteria.
Try: Find candidate drugs for PDB ID 3TCT with binding affinity lower than -8.0

Supervisor (auto-route) custom config
objectives set
custom LLM
🧑‍🔬 Chemist review — type your feedback below (or press Enter for no feedback) Waiting for input
Siloam Drug.AI may produce errors — verify results independently.
⚙️ Pipeline Configuration — overrides configs.yaml defaults
🌳 LATS Tree Search
Molecules sampled per roll-out (default 25)
Max selected fragments per branch (tree depth)
Stopping criterion (number of simulations)
UCB exploration weight (default 1.4)
Child fragments created per expansion
Generation
Total molecules to generate
Top candidates selected for refinement
Molecules requested per LLM call
Max LLM generation calls
Max Tanimoto similarity between seeds
Max seeds sharing the same scaffold
Refinement
Critique→Revise iterations per seed
LLM reviser attempts per revision
Min Tanimoto to seed (prevents drift)
Max molecular weight (Da)
Max rotatable bonds
Diversity
Max Tanimoto for deduplication
Max molecules per scaffold cluster
🎯 Objectives & Constraints — injected into every run
Drug-likeness
Valid SMILES
Only accept RDKit-parsable SMILES
Passed Lipinski (Ro5)
MW ≤ 500, HBD ≤ 5, HBA ≤ 10, LogP ≤ 5
Minimum QED
Quantitative estimate of drug-likeness (0–1)
QED and docking are mutually exclusive — enabling QED disables docking.
Max SAScore
Synthetic accessibility (1=easy … 10=hard)
Binding Affinity (Vina Docking)
Enable docking objective
Runs AutoDock Vina by default; cavities are proposed when the box is unchanged
🤖
LLM Provider
🟢 OpenAI
🔶 Anthropic
🔷 Gemini
🦙 Ollama (local)
🧠 GLM / Z.ai
🌙 Kimi
🔹 Qwen
🔎 DeepSeek