Chebifier 2: An Ensemble for Chemistry

Presentation given at the Ontologies4Chem Workshop 2025. Abstract: Chemical classification is a complex task without a one-size-fits-all solution. Rule-based approaches are explainable and reliable, but require expert knowledge. Deep Learning methods are easier to scale, but depend on the presence of sufficient data and are black boxes to the user. Chebifier 2 is an ensemble approach that combines deep learning, rule-based and LLM-based approaches to leverage their individual strengths, covering a total of more than 1,700 ChEBI classes.