MATCHA Logo

MATCHA - Modelling and AI Toolkit for Chemistry in Healthcare Applications

⚠️ Work in progress. MATCHA is under active development and no stable release is available yet. Public APIs, CLI commands, and packaging may change without notice. Pin exact versions if you depend on it, and expect breaking changes between releases.

MATCHA is a general toolkit for molecular property prediction with deep learning models.

Its core design philosophy is to wrap a wide variety of algorithms (e.g. descriptor/fingerprints, SMILES-based, graph-based, conformer-based) into a consistent Python APIs and CLI. Additionally, the package aims to be as agent-friendly as possible, with extensive docs to help LLMs use it and contribute to it.

MATCHA can help solve many typical issues found when developing QSAR models, such as:

⚖️ Quickly comparing many different algorithms to check for statistically meaningful differences
🎯 Pretraining and finetuning models, both supervised and self-supervised
🧩 Multitask learning
🔍 Explainability
📊 Uncertainty estimation

Making models, whether single predictors, ensembles or finetuned foundation models, can be done with two lines of code, starting from RDKIT molecules and numpy arrays:

# training and predicting with a generic model
from matcha.sklearn.graph import ChempropRegressor

model = ChempropRegressor()
model.fit(mols, y)
predictions = model.predict(mols)

# training an ensemble
from matcha.sklearn import Ensemble
from matcha.sklearn.clm import CNNRegressor

ensemble = Ensemble(
  model=CNNRegressor(),
  n_models=10
)
ensemble.fit(mols, y)

# Finetuning a foundation model
from matcha.sklearn.finetuning import FinetuningRegressor

finetuner = FinetuningRegressor(
  path_to_pretrained = "path/to/your/matcha/foundation/model"
)
finetuner.fit(mols, y)

Head over to the installation and getting started pages to begin!

🛠️ How to contribute

Please check out the contribution guide. We strongly recommend using the amazing Mach10 coding workflows to help contributing to the codebase with agents.

👨🏻‍🔧 Contacts

Davide Boldini, (Merck KGaA, Darmstadt, Germany), Contact
Lukas Friedrich (Merck KGaA, Darmstadt, Germany), Contact
Jakub Gunera (Merck KGaA, Darmstadt, Germany), Contact
Christina Schindler (Merck KGaA, Darmstadt, Germany), Contact
Daniel Kuhn (Merck KGaA, Darmstadt, Germany), Contact

📖 Citation

If you use MATCHA, please consider citing our preprint:

@article{matcha_2026,
  author = "Boldini, Davide and Friedrich, Lukas and
            Gunera, Jakub and Schindler, Christina and Kuhn, Daniel",
  title  = "{MATCHA}: a toolkit for streamlining molecular property prediction
            for drug discovery applications",
  journal = "ChemRxiv",
  year = "2026",
  doi = "10.26434/chemrxiv.15006766/v1",
  url = "https://chemrxiv.org/doi/abs/10.26434/chemrxiv.15006766/v1",
}

📄 License

Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Getting started

Contributing

API reference

API reference