Skip to content
Faraday

Compound property screening

Use this when you have a list of candidate molecules and want a fast property screen — molecular weight, logP, QED — before committing to more expensive assays or synthesis.

What you get

Section titled “What you get”
  • Physicochemical properties calculated via rdkit for each compound
  • A rule-based drug-likeness assessment against Lipinski and Veber-style criteria
  • A ranked comparison table across all compounds in a single run
  • A short recommendation on which compounds look best for follow-up
  • A reusable trajectory you can extend with additional compounds or analyses

How it works

Section titled “How it works”

You give Faraday a list of compound names and a clear screening brief. The agent fetches structures from PubChem, writes Python code using rdkit, executes it inside a Docker sandbox, collects the output, and assembles a structured comparison. If a compound fails a heuristic check, the agent flags it and explains why.

This fits early-stage triage workflows where you want a fast, reproducible pass over a compound series before a deeper SAR or assay planning discussion.

Setup

Section titled “Setup”

This example assumes:

  • Faraday runs locally on your machine
  • Agent-generated code runs in the Docker sandbox
  • You built the provided sandbox image with Dockerfile.sandbox
  • The sandbox can reach PubChem over the network
  • OPENAI_API_KEY is set

Build the sandbox image once:

Terminal window
docker build -f Dockerfile.sandbox -t faraday-code-sandbox .

If you use the provided Dockerfile.sandbox, both rdkit and pubchempy are already included.

Minimal faraday.yaml:

launch:
  mode: host


execution:
  backend: docker
  workspace_path: /workspace


runtime:
  workspace:
    source: ./

Run faraday --check-tools to verify credentials and backend selection before running a longer analysis.

Typical workflow

Section titled “Typical workflow”

1. Run the screen:

Terminal window
faraday "Use PubChem to retrieve structures for the following compounds, then use rdkit to analyze them for early-stage medicinal chemistry triage.


For each compound, calculate:
- molecular weight
- logP
- QED
- TPSA
- H-bond donors
- H-bond acceptors
- rotatable bonds


Then:
- assess each compound against Lipinski's Rule of Five
- note any obvious permeability or flexibility concerns using TPSA and rotatable bonds
- rank the compounds from most promising to least promising for oral drug-like follow-up
- explain the ranking in plain English
- include the PubChem CID used for each compound
- save a CSV called compound_screen.csv with the calculated properties
- present a final summary table in the answer


Compounds:
- warfarin
- imatinib
- aspirin
- sotorasib"

Faraday fetches the structures, writes and executes a Python script using rdkit.Chem and rdkit.Chem.Descriptors inside the Docker sandbox, saves the CSV into agent_outputs/, and returns a ranked summary table.

2. Extend with additional compounds using the trajectory as context:

Terminal window
faraday --previous-context run_outputs/run_<timestamp>_<chat_id>_<run_id>/run_artifacts/trajectory.json \
  "Add gefitinib and erlotinib to the existing screen, recompute the ranking, and highlight which compounds look most balanced for oral follow-up."

3. Request a plot:

Terminal window
faraday --previous-context run_outputs/run_<timestamp>_<chat_id>_<run_id>/run_artifacts/trajectory.json \
  "Create a scatter plot of MW vs logP for all compounds, label each point, color by QED, and save it as compound_screen.png."

The figure lands in agent_outputs/ inside the run directory.

What the agent does step by step

Section titled “What the agent does step by step”
  1. Looks up each compound in PubChem and retrieves a structure identifier
  2. Converts the PubChem structure into an rdkit molecule
  3. Calculates MW, logP, QED, TPSA, HBD, HBA, and rotatable bonds
  4. Checks each compound against Lipinski’s Rule of Five and simple permeability or flexibility heuristics
  5. Writes a CSV or figure into agent_outputs/ when requested
  6. Formats results into a summary table and ranks the compounds for follow-up

When to use this

Section titled “When to use this”
  • Early hit triage before purchasing or synthesizing compounds
  • Comparing a series of analogs against drug-likeness criteria
  • Generating a property baseline for a SAR analysis
  • Producing a reproducible screening record with a full calculation trace
Section titled “Related pages”