drug-retrosynthesis

Goal

To predict the retrosynthetic pathways and synthetic accessibility of novel small molecules (such as undocumented fluorinated gases) using the state-of-the-art transformer models provided by IBM RXN for Chemistry.

Instructions

1. Identify Target Molecule

Ensure you have the valid canonical SMILES string for the target material or chemical you wish to synthesize.

2. Set Up the IBM RXN Environment

Because IBM RXN is a cloud-hosted API, you must have an API Key.

1. Sign up for a free IBM RXN account at https://rxn.res.ibm.com/
2. Generate an API Key in your user profile.
3. Export the key in your terminal session before running the skill script:

export RXN_API_KEY="your-api-key-here"

3. Run Retrosynthesis Evaluation

Use the wrapper script to submit the SMILES string to the IBM RXN API. The script will poll the server and return the predicted pathway and a confidence score for synthetic feasibility.

# Env: drugdisc-agent
python .agents/skills/drug-retrosynthesis/scripts/evaluate_ibm_rxn.py "target_smiles" --steps 3

Examples

Evaluating the synthetic pathway for a fluorinated gas analog (e.g., 2,3,3,3-tetrafluoropropene: FC(F)(F)C(F)=C):

# Env: drugdisc-agent
export RXN_API_KEY="api-key-here"
python .agents/skills/drug-retrosynthesis/scripts/evaluate_ibm_rxn.py "FC(F)(F)C(F)=C" --steps 3

Constraints

• Environments: Requires the drugdisc-agent conda environment.
• Dependencies: The script relies on the rxn4chemistry python library (pip install rxn4chemistry). If not installed, the script will gracefully exit with instructions.
• Rate Limits: The IBM RXN free tier has API limits. Do not use this in a high-throughput loop for thousands of molecules without a premium tier.
• Sourcing Constraints: This tool does not directly check commercial availability of the proposed precursors. You must manually verify if the starting materials proposed by IBM RXN are commercially available.

References

• Schwaller, P. et al., "Predicting retrosynthetic pathways using a combined linguistic model and hyper-graph exploration strategy," Chemical Science, 2020. DOI:10.1039/C9SC05033H
• IBM RXN for Chemistry: https://rxn.res.ibm.com/

---

Author: Sathya Edamadaka Contact: GitHub @snme