# MIT and BASF release CheMeleon for low-data molecular prediction

Digest AI · Research · published 2026-11-06T19:00:00Z · updated 2026-11-06T19:35:04Z

Canonical: https://digestai.news/story/mit-develops-chemeleon-model-that-outperforms-other-methods

## Summary

MIT and BASF researchers unveiled CheMeleon, an open-source molecular foundation model, on November 6, 2026. The model helps predict chemical properties when experimental data is scarce, aiding drug and materials discovery. It was developed using about one million molecules from PubChem and roughly 1,600 molecular descriptors, pretrained on calculated—not experimental—data like molecular weight and oxygen atom counts.

CheMeleon, with roughly 10 million parameters, outperformed classical models in benchmarks across 58 datasets from the Polaris initiative. It reduces the need for experimental testing by narrowing candidate compounds. The model is open-source via ChemProp and already used by pharmaceutical companies, including one that adopted it as a default tool. Its role remains prioritization, not perfect prediction, requiring further experimental validation.

## Key points

- CheMeleon predicts chemical properties with limited experimental data, aiding drug and materials discovery
- Outperformed classical models in 58 Polaris benchmark datasets; adopted by unnamed pharmaceutical firms

## Why it matters

CheMeleon cuts costs and time in drug discovery by reducing experimental testing needs, leveraging open-source pretraining for low-data scenarios.

## Sources

1. [MIT Details CheMeleon Model for Low-Data Molecular Prediction](https://unite.ai/mit-details-chemeleon-model-for-low-data-molecular-prediction) (Unite.AI, 2026-11-06)
2. [How to predict the behavior of molecules from scant data](https://news.mit.edu/2026/how-to-predict-molecule-behavior-with-scant-data-1106) (MIT News on AI, 2026-11-06, primary source)

## Cite

Digest AI, "MIT and BASF release CheMeleon for low-data molecular prediction", 6 November 2026, https://digestai.news/story/mit-develops-chemeleon-model-that-outperforms-other-methods

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/mit-develops-chemeleon-model-that-outperforms-other-methods.json