Scenario

In lead optimization the classic trap is whack-a-mole: push potency and break solubility; fix clearance and lose selectivity. The team needs a multi-parameter footprint that shows where each analog sits across every axis so design decisions optimize the whole profile, not one number.

Worked example. A series is being iterated to hold potency while pulling clearance and lipophilicity into range.

Agent workflow

1. Read the current series and assay data.
2. SAR series analysis — map the potency SAR and identify the levers.
3. ADMET liability triage — predict the developability axes for each analog.
4. Combine into a multi-parameter optimization (MPO) footprint and suggest the next design direction.

Demo output

An MPO footprint plot placing each analog across potency and ADMET axes, with the design directions that improve the overall profile. The magic moment: the tool points to an under-explored region of the series that improves clearance without costing potency.

Deterministic vs LLM

• Deterministic — the SAR analysis, ADMET predictions, and MPO scoring.
• LLM — the design-direction narrative and next-analog suggestions.

Every axis is a validated calculation. The agent reads and routes; the engine decides.

Limits

Design-support guidance, not a guarantee of the next analog's measured profile. It sharpens the design hypothesis; the assay confirms it.