Plants produce specialized metabolites, among others, to protect against biotic and abiotic stresses. Due to their diversity and bioactivity, these compounds have profound implications for the world economy, especially for the pharmaceutical and agrotechnology sectors. In spite of their importance, their biosynthesis is far from being understood. The automatic prediction of the precursors of these compounds, derived from primary metabolism, is relevant to expediting pathway discovery. Leveraging DeepMol’s automated machine learning engine, we find that regularized linear classifiers provide optimal, accurate, and accountable models for this task. They perform significantly better than state-of-the-art models while chemically explaining their predictions. The pipeline and models are available in the repository https://github.com/jcapels/SMPrecursorPredictor .