A similarity-augmented q-RASPR framework for the antioxidant potential prediction and identification of functional food compounds

In this research, we implemented a quantitative read-across structure-property relationship (q-RASPR) framework to develop a statistically robust predictive model for estimating the antioxidant potential of chemical compounds, an area of growing relevance in the nutraceutical and dietary research. Using an extensive dataset of 1911 structurally diverse molecules with experimentally determined antioxidant activity based on the 1,1-diphenyl-2-picrylhydrazyl (DPPH) radical scavenging assay, we developed a univariate q-RASPR linear regression (LR) model following stringent internal and external validation procedures. Comparative statistical analysis demonstrated that the q-RASPR model markedly outperformed a two-dimensional quantitative structure-property relationship (2D-QSPR) model across several key performance metrics. The generalizability and robustness of the q-RASPR approach were evaluated using three independent external datasets comprising 45, 70, and 62 molecules, respectively. Additionally, to further explore the model's applicability, the validated q-RASPR LR model was employed to predict the antioxidant activity of approximately 70,474 food-related compounds from the Food Database ( https://foodb.ca/ ). • A large and chemically diverse dataset of 1911 antioxidant activities was compiled. • Univariate q-RASPR model enhances accuracy using similarity integration. • External validation confirms robust and reliable model performance. • FoodDB was screened for antioxidant potential and data gap filling. • Introduced a reliable framework for the precise prediction of potent antioxidants.