Exploring machine learning for untargeted metabolomics using molecular fingerprints

Christel Sirocchi, Federica Biancucci, Matteo Donati, Alessandro Bogliolo, Mauro Magnani, Michele Menotta, Sara Montagna

Abstract: Highlights•Machine learning models are trained on molecular fingerprints of metabolites.•Resolution and interpretability of fingerprinting structural encoding are analysed.•Challenges, including class imbalance and high dimensionality, are addressed.•The chemical structure of metabolites is predictive of their metabolic response.•Feature importance analysis aligns with prior knowledge of affected pathways.