Machine learning-guided production of a nanoemulsion for delivery of anacardic acid

Abstract

Bioactive molecules from plants remain an important source of drug candidates for many diseases. However, such molecules have poor in vivo performance due to low water solubility leading to inadequate distribution. Oil-in-water nanoemulsion drug delivery systems can help counteract these limitations by improving drug distribution even into highly resistant tissues. A limitation preventing the widespread adoption of nanoemulsion drug delivery systems is the expensive, time-consuming development process by trial and error. In this study, we developed a nanoemulsion design strategy guided by machine learning. We retrieved and aggregated data such as average particle size and polydispersity index associated to nanoemulsion composition to construct a comprehensive dataset from available literature. A predictive machine learning model was used to identify improved self-nanoemulsifying system formulations, including olive oil as oily base and combinations of Tween 20, Tween 80, glycerol, and soy lecithin. The predictive power of the model was assessed by estimating the successful self-nanoemulsification through transmittance, and later confirmed by analyzing the formulations using Dynamic Light Scattering. As an experimental model, the nanoemulsions were loaded with an organic extract from Amphipterygium adstringens (a plant native to Mexico known as cuachalalate) containing anacardic acid. Encapsulation efficiency was measured by UHPLC, and the antiproliferative activity of the preparations was evaluated on HEPG2, a human hepatic cancer cell line, and HEK-293, a normal-like human embryonic kidney cell line. The machine learning model was able to accurately predict a successful formulation 81% of the time. The best-performing formulation, a combination of 10% olive oil, 60% Tween 20, and 30% glycerol, exhibited average particle size of 162.8±26 nm, with a polydispersity index of 0.234±0.03, and full encapsulation efficiency given the assay used. The naked nanoemulsion presented no toxicity in the normal-like cell line but exerted an inhibitory effect on the cancer cell line. Moreover, loading the plant extract into the selected formulation increased the cytotoxic effect on the cancer cell line in comparison to the naked nanoemulsion, the extract alone, and pure anacardic acid alone, yielding an IC50 value of 5.9±1.27 µM. These results suggest that the formulation identified by the model was a successful carrier of the plant extract and molecule of interest. This study presents a proof of concept on how artificial intelligence can reduce the development pipeline of nanoemulsified drug delivery systems.