Random forest algorithms: a tool to identify the impact of arbuscular mycorrhizal fungi inoculation, seed maturation stage and geographic diversity of Pimpinella anisum L. accessions on the physicochemical composition of seeds

Background: A study using random forest (RF) algorithms and principal component analysis (PCA) was proposed to identify the effects of arbuscular mycorrhizal fungal inoculation, the seed maturation stage and the geographic diversity of Pimpinella anisum L. accessions on the physicochemical composition of seeds. Seeds of six anise varieties from North African and Middle Eastern accessions were inoculated or not inoculated with AMF (an arbuscular mycorrhizal fungus) and then grown under controlled conditions. Seeds were harvested at three different maturity stages: mature seeds (157 d after sowing), premature seeds (147 d after sowing), and immature seeds (137 d after sowing). Forty-nine variables related to physical properties, total nutrients, metabolic compounds, essential oils, and biological activity were measured in P. anisum seeds. Results: The RF algorithm allows the differentiation of P. anisum varieties inoculated with AMF from different countries in North Africa and the Middle East. This evidence proves that the geographic origin of P. anisum seeds significantly influences the efficiency of the symbiotic association between anise roots and AMF. In turn, no significant effects of the seed maturation stage on the symbiotic interaction of plants with mycorrhizae were observed. The chemical compounds related to the biological activity of seeds are not influenced by AMF, followed by chemical compounds related to metabolism, total nutrients, and oil components. Conclusions: The performance of classification models using RF is driven primarily by independent variables related to the chemical composition of anise seeds, overshadowing the effects of geographic diversity and the seed maturation stage. Among the chemical constituents of the seed, the variables belonging to the biological activity category best contain information (patterns) on the impacts of AMF inoculation.