Plant metabolites are important for plant breeders to improve nutrition and agronomic performance, yet integrating selection for metabolomic traits is limited by phenotyping expense and limited genetic characterization, especially of uncommon metabolites. As such, developing biologically-based and generalizable genomic selection methods for metabolites that are transferable across plant populations would benefit plant breeding programs. We tested genomic prediction accuracy for more than 600 metabolites measured by GC-MS and LC-MS in oat (Avena sativa L.) seed. Using a discovery germplasm panel, we conducted metabolite GWAS (mGWAS) and selected loci to use in multi-kernel models that encompassed metabolome-wide mGWAS results, or mGWAS from specific metabolite structures or biosynthetic pathways. Metabolite kernels developed from LC-MS metabolites in the discovery panel improved prediction accuracy of LC-MS metabolite traits in the validation panel, consisting of more advanced breeding lines. No approach, however, improved prediction accuracy for GC-MS metabolites. We tested if similar metabolites had consistent model ranks and found that, while different metrics of ‘similarity’ had different results, using annotation-free methods to group metabolites led to consistent within-group model rankings. Overall, testing biological rationales for developing kernels for genomic prediction across populations, contributes to developing frameworks for plant breeding for metabolite traits.