Prediction of Plant Uptake and Translocation of Engineered Metallic Nanoparticles by Machine Learning

Machine
learning was applied to predict the plant uptake and transport
of engineered nanoparticles (ENPs). A back propagation neural network
(BPNN) was used to predict the root concentration factor (RCF) and
translocation factor (TF) of ENPs from their essential physicochemical
properties (e.g., composition and size) and key external factors (e.g.,
exposure time and plant species). The relative importance of input
variables was determined by sensitivity analysis, and gene-expression
programming (GEP) was used to generate predictive equations. The BPNN
model satisfactorily predicted the RCF and TF in both hydroponic and
soil systems, with an R2 higher than 0.8
for all simulations. Inclusion of the initial ENP concentration as
an input variable further improved the accuracy of the BPNN for soil
systems. Sensitivity analysis indicated that the composition of ENPs
(e.g., metals vs metal oxides) is a major factor affecting RCF and
TF values in a hydroponic system. However, the soil organic matter
and clay contents are more dominant in a soil system. The GEP model
(R2 = 0.8088 and 0.8959 for RCF and TF
values) generated more accurate predictive equations than the conventional
regression model (R2 = 0.5549 and 0.6664
for RCF and TF values) in a hydroponic system, which could guide the
sustainable design of ENPs for agricultural applications.