A Machine Learning‐Driven 3D‐QSAR Approach for Developing Antioxidant Preservatives From Bovine Hemoglobin and Tryptophyllin L for Meat Products

ABSTRACT This study investigates lipid antioxidants derived from bovine hemoglobin as alternative preservatives for meat products using a joint statistical modeling and experimental approach. A machine learning‐enabled three‐dimensional quantitative structure–activity relationship (3D‐QSAR) model, integrating gradient‐boosting regression (GBR) with a recursive feature elimination (RFE) framework, was employed to screen 259 tripeptides from bovine hemoglobin for their antioxidant potential. A key feature of this model was the explicit definition of its Mahalanobis applicability domain and an evaluation of its impact on prediction errors. The antioxidant activities of promising peptides were assessed using multiple assays, including ferric thiocyanate (FTC), ferric reducing antioxidant power (FRAP), 2,2′‐azino‐bis(3‐ethylbenzothiazoline‐6‐sulfonic acid) (ABTS˙ + ) radical scavenging, and thiobarbituric acid reactive substances (TBARS) assays. Among the peptides, Phenylalanine‐Proline‐5HTP‐Leucine (F‐P‐5HTP‐L, where 5HTP represents 5‐hydroxytryptophan) showed significant efficacy in delaying pork meat oxidation, achieving a TBARS value of 2.519 ± 0.113 mg malondialdehyde (MDA)/kg, comparable to BHT, which had a TBARS value of 2.472 ± 0.455 mg MDA/kg at 0.1% (w/w). Additionally, peptides Glycine‐Arginine‐Leucine (GRL), Leucine‐Arginine‐Valine (LRV), Histidine‐Arginine‐Tyrosine (HRY), and Glutamine‐Arginine‐Phenylalanine (QRF) demonstrated significant activity against linoleic acid oxidation, though less effectively than F‐P‐5HTP‐L and BHT in meat products. This study highlights the potential of F‐P‐5HTP‐L and HRY as promising natural alternatives to synthetic antioxidants, with significant implications for reducing lipid oxidation and improving the quality of meat products.