# Ultrasound-Assisted Sustainable Processing of Garden Cress Juice: Enhancing Bioactive Compounds and Bioaccessibility through XGBoost Optimization

## TL;DR

This study uses ultrasound and machine learning to improve the nutritional value and bioaccessibility of garden cress juice.

## Contribution

A novel XGBoost model optimizes ultrasound parameters to enhance bioactive compounds and aroma preservation in juice processing.

## Key findings

- Optimal ultrasound conditions (12 min, 80% amplitude) significantly increased TPC, FRAP, and chlorophyll levels.
- Ultrasound preserved volatile aroma compounds like 1-hexanol and cinnamaldehyde better than heat treatment.
- Bioaccessibility of bioactive compounds improved by 28.81–34.96% in ultrasound-treated samples.

## Abstract

This study aimed
to improve the functional and nutritional properties of garden cress
(Lepidium sativum) juice using ultrasound
and optimize process parameters by modeling them with advanced machine
learning algorithms. Using a Box–Behnken experimental design,
the effects of sonication time (8–16 min) and amplitude (60–100%)
on total chlorophyll, total phenolic content (TPC), and ferric reducing
antioxidant power (FRAP) were investigated. Nonparametric, high-accuracy
estimations were made using the XGBoost algorithm. Optimum conditions
were determined to be 12 min and 80% amplitude. Under these conditions,
TPC (78.44 mg GAE/mL), FRAP (59.80 mg TE/mL), and chlorophyll (7.15
g/100 mL) values were significantly higher than those in control and
pasteurized samples (p < 0.05). HPLC-DAD analysis
showed that ultrasound treatment positively impacted the phenolic
profile by increasing the release of quercetin, quercetin derivatives,
caffeic acid, and chrysin. GC-MS data revealed that volatile aroma
compounds (especially 1-hexanol, benzaldehyde, and cinnamaldehyde)
were preserved mainly by ultrasound. In vitro digestion simulation
showed that total postdigestion recovery rates in ultrasound-treated
samples were 34.96% for TPC, 32.50% for chlorophyll, and 28.81% for
FRAP, demonstrating a significant increase in bioaccessibility. PCA
and hierarchical clustering analyses confirmed a significant biochemical
separation of ultrasound-treated samples. The findings indicate that
ultrasound technology is a superior method for preserving bioactive
compounds, maintaining the aroma profile, and enhancing bioaccessibility
compared to heat treatment. This enables data-driven process design.
The developed model showed a strong predictive performance under optimal
conditions. However, the study is limited by the relatively small
data set used for model training.