Why is high-quality training data for machine learning important?
Machine learning is transforming the way we approach problems in analytical chemistry. But there’s a catch: ensuring reliable results requires careful selection of training data to avoid biases that can mislead models.

We explain:
✨ why high-quality training datasets are important for SIRIUS method development
✨ why representing the full “universe” of small molecules is crucial
✨ how widely used datasets fail to evenly represent the diversity of biomolecular structures
✨ which tools can help evaluating dataset quality.

Read the full blog post.

F. Kretschmer et al. Nat Commun (2025) 10.1038/s41467-024-55462-w

We’ve Got You Covered!
We’re committed to ensuring you have the best experience possible with our software.
Here’s how you can find support and get answers:

• SIRIUS Documentation: For comprehensive guides, troubleshooting tips, and detailed information about SIRIUS
• Community Forum on Gitter: For community support and discussions and to connect with other users
• YouTube Tutorials: How to get started with SIRIUS and in-depth background information.
• Bug Reports and Feature Requests on GitHub: If you encounter any issues or have suggestions for new features

OpenMS 3 provides interfaces with SIRIUS for molecular formula and and molecular structure predictions.

OpenMS is an open-source project that boasts a comprehensive set of algorithms, modular tools, and seamless interoperability with open-standard file formats, ensuring flexibility and ease of integration into existing workflows.

J. Pfeuffer et al. Nat Methods (2024). 10.1038/s41592-024-02197-7

CANOPUS, our tool to predict the complete chemical class hierarchy of an unknown compound by assigning ClassyFire compound classes, is among the top 50 cited articles on AI in health and life sciences.

B. S. Glicksberg et al. Applied Sciences (2024) 10.3390/app14020785

Already over 30 years ago, Curry and Rumelhart introduced one of the first machine learning approaches to predict the presence or absence of key chemical substructures from tandem mass spectra. There are noteworthy conceptual similarities between their approach and SIRIUS structure annotation (CSI:FIngerID).
What they did not see was that a large number of weakly accurate predictions can still allow molecules to be unambiguously identified – a crucial conceptual leap that makes CSIFingerID the state-of-the-art method currently available for identifying small molecules from their MS/MS spectra.

M. A. Skinnider. Nat Rev Chem (2024) doi: 10.1038/s41570-023-00570-2