---
title: REINVENT
description: Design molecules with REINVENT 4 reinforcement learning, scoring functions and diversity filters — soft-deprecated here in favour of GenMol.
tags:
  - reinvent
---

REINVENT designs novel small molecules toward a target profile using reinforcement learning, so you get new candidates when existing compounds don't fit. Reach for it when you want to steer generation with configurable scoring and curriculum stages toward a desired property profile; where [GenMol](/docs/genmol) generates fragment-natively via diffusion, REINVENT optimizes toward a scoring objective. See [what you can achieve](/docs/capabilities#chemistry) for the full set.

## Try it

**Try this prompt** — [open in a guest chat](/chat/new?guest=true&prompt=Run%20REINVENT4%20de%20novo%20design%20targeting%20CDK4%2F6%20(scaffold%3A%20pyrido%5B2%2C3-d%5Dpyrimidinone)%2C%20apply%20medchem%20filters%20(PAINS%2C%20Brenk%20structural%20alerts%2C%20lead-likeness)%20to%20the%20generated%20compounds%2C%20then%20run%20ADMET-AI%20on%20the%20passing%20compounds%20%E2%80%94%20report%20the%20attrition%20rate%2C%20identify%20the%20most%20drug-like%20scaffolds%20that%20survive%2C%20and%20recommend%20the%20top%202%20for%20synthesis.).

> Run REINVENT4 de novo design targeting CDK4/6 (scaffold: pyrido[2,3-d\]pyrimidinone), apply medchem filters (PAINS, Brenk structural alerts, lead-likeness) to the generated compounds, then run ADMET-AI on the passing compounds — report the attrition rate, identify the most drug-like scaffolds that survive, and recommend the top 2 for synthesis.

_Generates candidates with REINVENT, then chains RDKit medchem filters and ADMET-AI screening._

> **Soft-deprecated**

> `reinvent4_generate` is soft-deprecated on the platform in favor of `genmol_generate`, which offers better fragment-awareness and SAFE-based generation. REINVENT still runs — reach for it when you explicitly want REINVENT by name, or need its RL-specific features (scoring functions, diversity filters, transfer learning) that GenMol doesn't expose.

REINVENT on the platform supports RL-based molecule generation with configurable scoring
and curriculum stages. Follow the linked guides to stand up reproducible
campaigns and export results downstream.

## Citations

> Loeffler HH, He J, Tibo A, Janet JP, Voronov A, Mervin LH, Engkvist O. Reinvent 4: Modern AI–driven generative molecule design. *Journal of Cheminformatics*. 2024;16:20. doi: [10.1186/s13321-024-00812-5](https://doi.org/10.1186/s13321-024-00812-5)

## Chat Usage

### Example Prompts

- "Generate 100 analogs of gefitinib using REINVENT sampling mode"
- "Use REINVENT to design molecules with reduced H-bond donors for CNS penetration"
- "Run REINVENT in scaffold decoration mode on this kinase inhibitor scaffold"

### Workflow Chaining

- **Feeds into**: ADMET-AI/ADMETLab3 (drug-likeness screening), AiZynthFinder (synthesis planning), Compound Passport
- **Accepts from**: PubChem/ZINC/ChEMBL (seed SMILES)
- **SAFE fragments**: Generated molecules are auto-ingested as SAFE fragments for reuse in fragment library

## FAQ

### How do I keep runs reproducible?

  Pin REINVENT, RDKit, and scoring component versions; store seeds in job metadata.

### What data formats are expected?

  Use CSV or SDF with SMILES in a `smiles` column; include optional score columns for pretraining.

## Read next