Abstract Detail

Ecophysiology

Majumder, Sambadi [1], Mason, Chase [1].

Using a Machine Learning approach to study functional trait diversification in Helianthus.

Functional traits indirectly relate to plant growth and fitness and are used to study the differences in ecological strategy across species. Due to natural selection and the role of these traits in the context of plant performance, some traits might diverge more than others enabling in the survival of the species in a new environment. Descriptive and predictive modelling were used to identify functional traits that most strongly delineate the species within the Helianthus genus in a multivariate trait space. This approach was also applied to elucidate whether the same traits strongly diverge at the clade level within the genus, or whether distinct axes of differentiation exist. The results were also examined in the light of established evolutionary paradigms such as the Leaf-Height-Seed spectrum, Competitive-Stress-Ruderal scheme, and the Plant Economic Spectrum and Leaf Economic Spectrum to better understand ecological performance across species from an ecophysiological context. Feature selection approaches like Gini Impurity within a Random Forest (RF) classifier, Mean Decrease of Accuracy within the framework of Recursive Feature Elimination and the Boruta algorithm were used in the descriptive modelling approach. Predictive modelling applied to test datasets were used to evaluate overfitting and validate the findings of descriptive models. The RF and the Gradient Boosting Machine classifier were used to create predictive models. All the feature selection approaches overlapped quite considerably in which traits were deemed the most divergent. At the genus level and for two perennial clades, water resource related traits like leaf area, leaf shape, and trichome density appear to be the most divergent. Within the annual clade, species appear to diverge strongly in leaf trichome density and whole plant total biomass. The findings are consistent with existing knowledge that differences in functional traits due to interspecific ecological strategies appear when a genus occupies a wide variety of biomes which is true in the case of the Helianthus genus. It is also in line with existing hypotheses which pertain to the fast-growing ruderals of the annual clade and longer-lived perennial clades. Furthermore, the results provide some insight into the diversification of Helianthus into three separate clades and development of distinct and overlapping ecological strategies of the species in each. Additionally, the method used in this study can be implemented to identify and rank functional traits according to their biological significance and can facilitate in the formation of consensus traits for global comparison of interspecific ecological strategies, future meta-analysis and forecasting vegetation dynamics under a changing climate.

1 - University Of Central Florida, Department Of Biology, 4110 Libra Drive, Orlando, FL, 32816, United States

Keywords:
sunflower
Helianthus
Machine Learning
functional traits
Evolutionary biology
ecophysiology
Algorithms.

Presentation Type: Oral Paper
Number: EPH1002
Abstract ID:390
Candidate for Awards:Physiological Section Physiological Section Li-COR Prize,Physiological Section Best Paper Presentation