Using Machine Learning To Predict the Self-Assembled Nanostructures of Monoolein and Phytantriol as a Function of Temperature and Fatty Acid Additives for Effective Lipid-Based Delivery Systems

Lyotropic
liquid crystalline lipid nanomaterials have shown promise
as delivery vehicles for small therapeutic drugs, protein, peptides,
and in vivo imaging contrast agents. To design effective
lipid-based delivery systems, it is important to understand and be
able to predict their self-assembly processes. In this study, we utilized
a machine learning approach to study the phase behavior of a nanoparticulate
system consisting of a base lipid, monoolein, or phytantriol and varied
the concentration of saturated and unsaturated fatty acids. The experimental
data sets acquired by high throughput characterization techniques
were used to train the “machine” using two separate
models, i.e., multiple linear regression (MLR) and Bayesian regularized
artificial neural networks (ANNs). The models were accurate (>70%)
in predicting the phase behavior for data used to train the neural
networks. The ANN model appeared to be more accurate than the MLR
model in predicting mesophases. We then used the obtained ANN models
to interpolate the phase behavior of various nanoparticles at temperatures
not yet tested. Compared to the experimental result, the prediction
of phase behavior was interpolated with high accuracy, ranging from
66% to 96% for the different phases. The models were capable of interpolating
data for the same fatty acids at temperatures that were not yet tested
as well as extrapolating data for new fatty acid structures. We also
studied quantitatively the contributions of various factors on the
formation of different mesophases and elucidated rules that are useful
for future design of advanced lipid systems for therapeutic delivery.