Add compound prediction based on ChEBI class appartenance
Current predictor tries to predict the type of the discovered BGCs using a kNN, but it stops at the level of BGC compound type. Given that MIBiG compounds are well represented within the ChEBI ontology, it could be very much feasible to try to attempt to predict the compound type based on the most-common chemical class between the nearest neighbors.
E.G. if the nearest neighbours are streptomycin and kanamycin BGCs, you could predict the compound as being the closest common ChEBI class between the two, which is aminoglycoside antibiotic.