---
title: "Filter Compounds For Drug-Likeness"
description: "Filter Compounds For Drug-Likeness: a medchem recipe chaining admet_ai."
tags:
  - recipes
  - medchem
tools:
  - admet_ai
generated: true
source: quickActionConfig:medchem-admet-filter
---

> **Generated recipe**

> Seeded from a curated quick-action prompt. The prompt below runs the real
> routing — copy it, change the target, and run.

> **Recommended pairing**

> These tools are commonly run together — they cover complementary ground on the same input.

**Best fit:** filter compounds for drug-likeness — a medchem goal.

**Try this prompt** — [open in a guest chat](/chat/new?guest=true&prompt=Search%20ChEMBL%20for%20JAK2%20inhibitors%20with%20IC50%20%3C%20100nM%2C%20run%20medchem%20filters%20(Lipinski%2C%20PAINS%2C%20NIBR%2C%20Lilly%20demerits)%20on%20the%20top%2030%2C%20run%20ADMET-AI%20on%20the%20survivors%2C%20then%20dock%20the%20top%205%20filter-passing%20compounds%20against%20JAK2%20(PDB%3A%204GI3)%20with%20DiffDock%20and%20score%20with%20GNINA%20for%20consensus%20-%20load%20all%20data%20into%20the%20sandbox%20and%20use%20RDKit%20to%20compute%20Murcko%20scaffolds%2C%20generate%20a%20trellis%20plot%20with%20matplotlib%20showing%20each%20scaffold's%20filter%20pass%20rate%20%C3%97%20ADMET%20score%20%C3%97%20consensus%20docking%20confidence%2C%20and%20recommend%20the%20top%203%20compounds%20for%20a%20JAK2-selective%20lead%20series%3F).

> Search ChEMBL for JAK2 inhibitors with IC50 < 100nM, run medchem filters (Lipinski, PAINS, NIBR, Lilly demerits) on the top 30, run ADMET-AI on the survivors, then dock the top 5 filter-passing compounds against JAK2 (PDB: 4GI3) with DiffDock and score with GNINA for consensus - load all data into the sandbox and use RDKit to compute Murcko scaffolds, generate a trellis plot with matplotlib showing each scaffold's filter pass rate × ADMET score × consensus docking confidence, and recommend the top 3 compounds for a JAK2-selective lead series?

_A real `medchem-admet-filter` quick-action prompt — open it in a guest chat, then swap in your own compound or target._

## The prompt

```text
Search ChEMBL for JAK2 inhibitors with IC50 < 100nM, run medchem filters (Lipinski, PAINS, NIBR, Lilly demerits) on the top 30, run ADMET-AI on the survivors, then dock the top 5 filter-passing compounds against JAK2 (PDB: 4GI3) with DiffDock and rescore with MM-GBSA for consensus - load all data into the sandbox and use RDKit to compute Murcko scaffolds, generate a trellis plot with matplotlib showing each scaffold's filter pass rate × ADMET score × consensus docking confidence, and recommend the top 3 compounds for a JAK2-selective lead series?
```

## What each tool contributes

- **medchem filter** — Drug-likeness, PAINS, structural alerts, Lilly demerits
- [admet_ai](/docs/admet-ai) — 50+ ADMET endpoints on filtered compounds

## Why these tools pair

MedChem filters eliminate structural liabilities first, then ADMET predicts pharmacokinetic properties on the survivors — efficient funnel

## Data flow

Every tool operates on the same input: `smiles_array`.

## Read next

- [Which tool?](/docs/which-tool) — match a single goal to a tool.
- [Workflows](/docs/workflows) — every workflow and recipe.
- [How the orchestrator picks tools](/docs/how-orchestration-works) — the same chain from a plain-language goal.