“…However, this approach has limitations when the aim is to identify candidate species that may be morphologically dissimilar to relatives known to possess a trait of interest (e.g., domesticated crops and their wild relatives), if the trait has not been systematically characterized in a plant group, and if highly novel candidate species are sought. In those cases, candidate species can be predictively identified from the clades in the tree of life in which that trait is most likely to occur (Ernst et al., 2016; Grace, 2019; Rønsted, Savolainen, Mølgaard, & Jäger, 2008; Zhu et al, 2011). Machine learning analysis, recently successful in the search of new antibiotics (Stokes et al, 2020), could help to separate the search for candidate species from their classification or tree of life altogether, facilitated by curated trait datasets (Kattge et al, 2020).…”