CPSign: Conformal Prediction for Cheminformatics modeling

CPSign is a complete package for cheminformatics modeling, including descriptor calculation and machine learning with conformal prediction for producing valid prediction intervals on a per-compound basis. If you use CPSign please cite:

Arvidsson McShane S, Norinder U, Alvarsson J, Ahlberg E, Carlsson L, and Spjuth O.
CPSign - Conformal Prediction for Cheminformatics Modeling
bioRxiv. 2023.11.21.568108 (2023). DOI: 10.1101/2023.11.21.568108

Download a CPSign release from GitHub. CPSign is open source but commercial use requires a license, contact info@arosbio.com to discuss licensing options. CPSign is trusted by companies such as AstraZeneca, Phenaros, and Prosilico.

Aros Bio AB is a Swedish spin-off from Uppsala University and the Pharmaceutical Bioinformatics research group.

Robust and accurate modeling

CPSign uses Signatures as the default descriptor and SVM as the default modeling method. It has been benchmarked to perform on par with other modeling methods, including Deep Learning. CPSign can easily be extended with other descriptors and modeling methods.


Valid confidence with conformal prediction

Conformal prediction is a well established mathematical framework that delivers object-based predictions, ensuring valid (well-calibrated) prediction intervals instead of point predictions. This is

a compelling alternative to domain applicability estimation.


Superior runtime,

low footprint

CPSign has been optimized and battle-hardened in production environments over 10+ years. The fast modeling and predictions makes it attractive in interactive environments where fast predictions are required.


Visualize predictions in chemistry

Using Signatures as chemical descriptors, CPSIgn allows for visualizing the atom contributions to a specific prediciton in the chemical structure.


Easily integrate via CLI & Java API

CPSign has an extensive and well-documented  Command Line Interface. It also comes with a JAVA API and can be integrated in third party applications.


Easily deploy models as online services

The trained models with CPSign can easily be packaged as containers and served  with an API, making them straightforward to run in containerized environments such as Kubernetes.