Machine learning prediction of 18 date palm polyphenol protection against biomolecular damage

Abstract

This study investigated the diverse antioxidant and enzyme-inhibiting properties of 18 date palm cultivars, correlating these bioactivities with polyphenol profiles using biochemical methods and machine learning (ML). Maktoomi exhibited the highest phenolic content (759.42 mg GAE/100g), while Fard showed strong ferric-reducing antioxidant power (FRAP) activity (2456.13 mmol/100g). Significant enzyme inhibition variation was observed, Jabri (8.69 % AChE inhibition), Shikat alkahlas (21.06 % α-amylase inhibition), and Barhe (51.39 % tyrosinase inhibition). Maghool provided the highest protein protection (95–100 % BSA). These bioactivity data were integrated into an extreme gradient boosting (XGBoost) ML model to connect chemical features with experimental outcomes. The model demonstrated high predictive capability (R² ∼ 0.9–0.95) for amylase, acetylcholine, and 2,2-diphenyl-1-picrylhydrazyl (DPPH) assays, but lower values (R² < 0.9) for more complex assays involving DNA or superoxide systems, indicating data quality limitations. This highlights targeted method improvements. These findings demonstrate that certain date components offer higher specific bioactivity, and the ML approach validates these methods, revealing benefits and limitations. Date extracts possess therapeutic potential, and the combined approach of experimental testing and ML mapping provides a framework for multi-parameter analysis of complex biological systems. However, further in-vivo validation is needed.