Antioxidant peptides from yak milk, composition and application thereof screened based on deep learning

Abstract

The present application relates to the technical field of small molecule peptides, in particular to yak milk-derived antioxidant peptides, compositions and applications based on deep learning screening. The antioxidant peptides are selected from one of the following sequences: LFEEQMPFY, the amino acid sequence is shown as SEQ ID NO:1; YIPQMAGD, the amino acid sequence is shown as SEQ ID NO:2; DAYPSG, the amino acid sequence is shown as SEQ ID NO:3; YPQRD, the amino acid sequence is shown as SEQ ID NO:4. The four antioxidant peptides obtained in the present application have no cytotoxicity, hemolysis or safety risk, and two of them also have obvious umami characteristics. The antioxidant peptides obtained in the present application can be applied in the preparation of food, health products and cosmetics.

Description

Technical Field

[0001] This invention relates to the field of small molecule peptide technology, specifically to yak milk-derived antioxidant peptides, compositions, and their applications screened based on deep learning. Background Technology

[0002] Oxidative stress is a state caused by an imbalance between the generation of reactive oxygen species (ROS) and the body's antioxidant defenses, which can trigger lipid peroxidation, protein damage, and participate in various chronic metabolic abnormalities. Dietary antioxidant peptides, due to their safe origin and compatibility with daily diets, have become an important direction in functional food ingredient research. Hydrolysis of milk proteins can produce a variety of bioactive peptides, but different raw materials, protein composition, enzymatic digestion methods, and sequence structures can significantly affect the final activity.

[0003] Yak milk is a unique dairy resource from the Qinghai-Tibet Plateau, characterized by high nutrient density and a distinctive protein and amino acid composition. Theoretically, it could serve as an important precursor library for discovering novel bioactive peptides. Existing research has reported some antioxidant peptides derived from milk or yak milk, but systematic exploration of short yak milk peptides that simultaneously satisfy antioxidant and umami function remains limited. Especially in low-sodium dairy products and health food formulations, umami peptides can be used to improve palatability, while bifunctional peptides can provide both health-related functions and flavor compensation value as a single ingredient.

[0004] The background technologies that are currently closest to this invention mainly include: 1. Traditional antioxidant peptide discovery process: Protein raw materials are enzymatically digested, then separated by ultrafiltration, gel filtration, reversed-phase chromatography, etc., followed by mass spectrometry identification and in vitro antioxidant evaluation. This method is reliable but time-consuming, low-throughput, and prone to missing short peptides with low abundance but high activity.

[0005] 2. Computer-aided screening techniques: These techniques utilize known antioxidant peptide databases, QSAR models, machine learning, or deep learning models to predict the activity of candidate peptides. While these methods can improve screening efficiency, traditional models typically rely on hand-crafted features and lack generalization ability in scenarios involving short peptide sequences, small numbers of positive samples, and noisy negative samples.

[0006] 3. Flavor-Predicting Peptides and Electronic Tongue Validation: Flavor-predicting is performed using platforms such as the flavor-predicting peptide database and VIRTUOUS, and then validated via electronic tongue or receptor molecular docking. Current technologies primarily focus on flavor-predicting peptides in isolation, with insufficient emphasis on their synergistic development with antioxidant functions.

[0007] 4. Research on the Keap1-Nrf2 antioxidant pathway mechanism: Some dietary peptides can reduce intracellular ROS and increase the expression of downstream antioxidant enzymes by affecting the Keap1-Nrf2-ARE pathway. However, the sequence, site of action, dynamic binding stability, and quantum chemical electron transfer mechanism of short peptides derived from yak milk still need further clarification.

[0008] The main shortcomings of the above-mentioned technologies are: the candidate peptide library is large in scale but the experimental verification cost is high; the short peptide level of yak milk protein resources is not fully explored; the linkage screening of antioxidant and flavor-enhancing dual functional properties is insufficient; and there is a lack of a chain of evidence that connects computational prediction, in vitro free radical scavenging, cell protection, transcriptional regulation, molecular docking / kinetics and electronic tongue evaluation. Summary of the Invention

[0009] The purpose of this invention is to provide yak milk-derived antioxidant peptides, compositions, and their applications based on deep learning screening, in order to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: The yak milk-derived antioxidant peptides screened using deep learning were selected from one of the following sequences: LFEEQMPFY, the amino acid sequence of which is shown in SEQ ID NO:1; YIPQMAGD, the amino acid sequence is shown in SEQ ID NO:2; DAYPSG, amino acid sequence as shown in SEQ ID NO:3; YPQRD, the amino acid sequence is shown in SEQ ID NO:4.

[0011] An antioxidant composition comprising any two or more of the antioxidant peptides described above.

[0012] The application of the above-mentioned antioxidant peptides or antioxidant compositions in the preparation of products with antioxidant functions.

[0013] Furthermore, the products with antioxidant functions include food, dairy products, beverages, functional foods, health foods, nutritional supplements, cosmetics, feed or feed additives.

[0014] Furthermore, the antioxidant peptides DAYPSG and YPQRD possess umami properties.

[0015] The above-mentioned antioxidant peptides DAYPSG and / or YPQRD are used in the preparation of foods that enhance umami, improve flavor, or are low in sodium.

[0016] Furthermore, the foods described as having enhanced umami, improved flavor, or low sodium content include dairy products, fermented milk, milk beverages, flavored foods, low sodium foods, or health foods.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention provides four short peptide sequences derived from yak milk protein: LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD. All of these sequences have been verified through in vitro free radical scavenging, cell protection, and safety evaluations, and have good prospects for antioxidant applications. 2. This invention further confirms that DAYPSG and YPQRD possess both antioxidant activity and umami properties, and can be used as antioxidant-umami dual-function peptides for flavor enhancement and antioxidant function improvement in low-sodium dairy products, functional dairy beverages or other health food formulations.

[0018] 3. This invention establishes a comprehensive technical route for yak milk protein multi-enzyme simulated hydrolysis—short peptide library construction—dual-branch deep ensemble model screening—functional and safety verification—flavor verification. It can efficiently and preferentially screen high-confidence candidate peptides from 134,109 short peptide candidate sequences of 2-10 amino acids. The prediction model integrates Energy-Based Model (EBM) and Prototypical Network (PROTO), and combines hard negative sample mining, semi-supervised learning, knowledge distillation of unlabeled candidate peptides, and dynamic weighted ensemble, which can reduce the false positive rate in short peptide screening. 4. All four selected peptides were predicted to be non-toxic, without significant allergen matching, and non-hemolytic; in cell experiments, they showed no significant inhibition of HepG2 and Caco-2 cell proliferation in the range of 0.1-4.0 mg / mL, providing a safe basis for their development as food-derived functional ingredients. Attached Figure Description

[0019] Figure 1 The above are experimental results of LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD on DPPH removal in this invention. Figure 2 The diagram shows the ABTS removal experiment using LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD in this invention. Figure 3 The above are experimental diagrams of the superoxide anion scavenging effects of LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD in this invention. Figure 4 This is an experimental diagram showing the total antioxidant capacity of LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD in this invention; Figure 5 This is a diagram showing the results of the LFEEQMPFY assay on the proliferation inhibition of HepG2 cells in this invention. Figure 6 This is a diagram showing the results of the YIPQMAGD inhibition assay on HepG2 cells in this invention. Figure 7 This is a diagram showing the results of the DAYPSG inhibition assay on HepG2 cells in this invention. Figure 8 This is a diagram showing the results of the YPQRD assay on the proliferation inhibition of HepG2 cells in this invention. Figure 9 This is a diagram showing the results of the LFEEQMPFY assay on the proliferation inhibition of Caco-2 cells in this invention. Figure 10 The figure shows the results of the YIPQMAGD inhibition experiment on Caco-2 cells in this invention. Figure 11 This is a diagram showing the results of the DAYPSG inhibition experiment on Caco-2 cell proliferation in this invention; Figure 12 This is a diagram showing the results of the assay for the inhibition of Caco-2 cell proliferation by YPQRD in this invention; Figure 13 This is a graph showing the survival rate of HepG2 cells subjected to oxidative damage by LFEEQMPFY treatment in this invention. Figure 14 This is a graph showing the survival rate of HepG2 cells subjected to oxidative damage by YIPQMAGD treatment in this invention. Figure 15 This is a graph showing the survival rate of HepG2 cells subjected to oxidative damage by DAYPSG treatment in this invention. Figure 16 This is a graph showing the survival rate of HepG2 cells subjected to oxidative damage by YPQRD treatment in this invention. Figure 17 This is a graph showing the ROS content of oxidatively damaged HepG2 cells after LFEEQMPFY treatment in this invention. Figure 18 This is a graph showing the ROS content of oxidatively damaged HepG2 cells after YIPQMAGD treatment in this invention. Figure 19 This is a graph showing the ROS content of HepG2 cells after DAYPSG treatment for oxidative damage in this invention. Figure 20 This is a graph showing the ROS content of oxidatively damaged HepG2 cells after YPQRD treatment in this invention. Figure 21 Plot showing the relative expression levels of HO-1 mRNA in HepG2 cells from different treatment groups; Figure 22 Plot showing the relative expression levels of NQO1 mRNA in HepG2 cells from different treatment groups; Figure 23 Plot showing the relative expression levels of Keap1 mRNA in HepG2 cells from different treatment groups; Figure 24Plot of relative Nrf2 mRNA expression levels in HepG2 cells from different treatment groups; Figure 25 This is a schematic diagram of the molecular docking between LFEEQMPFY and Keap1 in this invention; Figure 26 This is a schematic diagram of the molecular docking between YIPQMAGD and Keap1 in this invention; Figure 27 This is a schematic diagram of the molecular docking between DAYPSG and Keap1 in this invention; Figure 28 This is a schematic diagram of the molecular docking between YPQRD and Keap1 in this invention; Figure 29 The graph shows the quantum chemical calculation results for LFEEQMPFY; Figure 30 The graph shows the results of YIPQMAGD's quantum chemical calculations. Figure 31 The graph shows the results of DAYPSG's quantum chemical calculations. Figure 32 The graph shows the quantum chemical calculation results of YPQRD; Figure 33 The image shows the test results of the electronic tongues DAYPSG and YPQRD in this invention. Figure 34 This is a schematic diagram of the docking of DAYPSG with T1R1 / T1R3 molecules in this invention; Figure 35 This is a schematic diagram of the docking of YPQRD with T1R1 / T1R3 molecules in this invention; In the above image Figure 1-4 The GSH in this sample is a control sample with added glutathione. Figure 5-12 The Control value represents the cell proliferation inhibition result without the addition of any target peptide; Figure 13-16 Damage in the text represents the cell survival rate of damaged cells without the addition of the target peptide. Figure 17-20 In this context, Control represents the ROS content of undamaged cells, and Damage represents the ROS content of damaged cells that have not received the target peptide. Figure 21-24 In the table, Control represents the relative mRNA expression level of undamaged cells without the addition of any target peptide, Damage represents the relative mRNA expression level of damaged cells without the addition of any target peptide, and GSH represents the relative mRNA expression level of damaged cells with the addition of glutathione. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1 to 35 The present invention provides: Example 1: Construction of a candidate library of yak milk protein short peptides (1) Using the complete protein sequences related to yak milk in UniProt as the raw material protein database, the search keywords included “yak” and “milk”; (2) Perform multi-enzyme simulated hydrolysis on yak milk protein sequences, preferably using pepsin, trypsin, chymotrypsin and elastase for individual enzyme digestion simulation; (3) Short peptides with a length of 4-10 amino acid residues were retained, and repetitive sequences were removed to obtain 134,109 candidate short peptide sequences; (4) The above calculation and screening steps can be combined with actual enzymatic hydrolysis, separation and purification, and mass spectrometry identification according to application needs to obtain the same sequence or enriched components in natural enzymatic hydrolysis products. Short peptides with a length of 4-10 amino acid residues were retained, and repetitive sequences were deleted, resulting in 134,109 candidate short peptide sequences. Statistical analysis of the candidate peptides' molecular weight, net charge, pI, hydrophobicity, and Boman index confirmed that multi-enzyme digestion yielded a diverse library of short peptides with complementary physicochemical properties.

[0022] Example 2: Antioxidant peptide prediction model training and candidate peptide screening Antioxidant peptides from the AODB and ChEMBL databases, as well as literature reports, were collected as positive samples, totaling 729 entries. Antimicrobial peptides, anticancer peptides, and antihypertensive peptides were selected as non-antioxidant negative samples, resulting in 159,202 duplicate entries. A lightweight Transformer-based EBM+PROTO dual-branch deep ensemble model was constructed. In the first stage, the EBM and PROTO branches were trained separately, and the optimal weights were saved based on the validation set AUC. In the second stage, the training and validation sets were merged, and knowledge distillation was performed using unlabeled yak milk candidate peptides. A threshold of Score > 0.99 was used to screen the yak milk candidate library, resulting in 205 high-confidence potential antioxidant peptides, as detailed below: A lightweight Transformer encoder is used as the basic feature extractor to map amino acid sequences into embedding representations, and positional encoding, multi-head attention, GELU activation, and Dropout are introduced. The model includes two complementary prediction branches: (1) EBM branch: The peptide activity prediction is transformed into an energy minimization problem. The scalar system energy E is output through the projection head, and the antioxidant probability is represented by P(pos)=σ(-E). During training, NCE loss, pairwise RankLoss, and CVaR tail penalty for unlabeled candidate peptides can be used to suppress high-confidence false positives in unknown backgrounds.

[0023] (2) PROTO branch: Map sequence features to metric space, maintain positive sample prototypes and negative sample prototypes, calculate the distance between input sequence and prototype through cosine similarity and temperature coefficient, and use supervised contrast loss, BCE loss, CORAL distribution alignment loss and prototype repulsion loss for training to improve the discrimination ability under few sample conditions.

[0024] (3) Two-stage training and integration: In the first stage, EBM and PROTO are trained independently, and the optimal weights are saved based on the validation set AUC; in the second stage, the training set and validation set are merged and retrained for a fixed number of rounds, and knowledge distillation of unlabeled candidate peptides is introduced; finally, the integration score is output using random weighted average (SWA) and dynamic weighting coefficient α.

[0025] The training set includes 729 reported antioxidant peptides as positive samples and non-antioxidant peptides such as antimicrobial peptides, anticancer peptides, and antihypertensive peptides as negative samples, resulting in 159,202 negative samples after deduplication. In model evaluation, the PROTO and Ensemble models showed ROC-AUCs of approximately 0.922 and 0.921, respectively, and APs of approximately 0.963 and 0.962, respectively, significantly outperforming the EBM model alone. This indicates that this method is suitable for prioritizing screening of large-scale yak milk short peptide libraries. Using a prediction score > 0.99 as the high-confidence antioxidant peptide screening threshold, 205 potential antioxidant peptides were obtained.

[0026] Based on the combined results of molecular weight, sequence length, syntheticity, safety prediction, and functional prediction, AOP1, AOP2, AOP3, and AOP4 were selected as the preferred peptides for synthesis and verification. Their sequences are LFEEQMPFY (AOP1), YIPQMAGD (AOP2), DAYPSG (AOP3), and YPQRD (AOP4), respectively.

[0027] Example 3: Synthesis and Quality Control of Four Antioxidant Peptides LFEEQMPFY (SEQ ID NO:1), YIPQMAGD (SEQ ID NO:2), DAYPSG (SEQ ID NO:3), and YPQRD (SEQ ID NO:4) were synthesized using a solid-phase peptide synthesis method. The purity and molecular weight of the products were confirmed by HPLC and MS, with a preferred purity greater than 95%. Peptide solutions with concentrations of 0.1, 0.25, 0.5, 1.0, 2.0, and 4.0 mg / mL were prepared using distilled water or suitable buffer solutions for subsequent free radical scavenging, cell protection, and safety experiments.

[0028] Example 4: Evaluation of in vitro free radical scavenging and reducing capabilities DPPH scavenging assay: 100 μL of peptide sample was mixed with 100 μL of 0.1 mM DPPH solution, and the mixture was shaken in the dark for 30 min before the absorbance was measured at 517 nm. The sample + 95% ethanol group served as the blank group, and DPPH + 95% ethanol served as the control group. The DPPH scavenging rate was calculated as [1-(A1-A2) / (A3-A2)]×100%. The results are as follows: Figure 1 As shown.

[0029] ABTS clearance assay: Dilute ABTS+ working solution with PBS to a absorbance of 0.7±0.05 at 734 nm. Mix 50 μL of peptide sample with 150 μL of ABTS+ working solution, incubate at room temperature in the dark for 10 min, and then measure the absorbance at 734 nm. Calculate the ABTS clearance rate as [1-A1 / A2]×100%. The results are as follows: Figure 2 As shown.

[0030] Superoxide anion scavenging experiment: The sample was mixed with NBT solution and NADH solution, and then PMS was added to start the reaction. After incubation in the dark for 5 min, the absorbance was measured at 560 nm. The scavenging rate was calculated as [1-A1 / A2]×100%. The test results are as follows: Figure 3 As shown. FRAP experiment: Mix 50 μL of sample with 150 μL of FRAP working solution, incubate for 10 min, and then measure the absorbance at 593 nm. The highest concentration positive control is defined as 100%. Calculate the relative reducing power. The test results are shown below. Figure 4 As shown.

[0031] Experimental results showed that all four target peptides exhibited enhanced free radical scavenging / reduction capabilities with increasing concentration. Considering the results of DPPH, ABTS, superoxide anion, and FRAP, LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD showed relatively balanced and strong antioxidant performance, making them suitable as preferred antioxidant peptides of this invention.

[0032] Example 5: Evaluation of cell safety, hemolytic activity, and protection against oxidative damage Cell safety: HepG2 and Caco-2 cells were seeded in 96-well plates. After adhesion, different concentrations of the target peptides were added for 24 h. Cell absorbance at 450 nm was measured using the CCK-8 assay, and cell viability was calculated. Results showed that within the range of 0.1–4.0 mg / mL, the four target peptides did not significantly inhibit the proliferation of either cell type, and cell viability remained close to 100% at the highest concentration. Figure 5-8 The figures show the results of assays inhibiting the proliferation of HepG2 cells using LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD, respectively. Figure 9-12 The figures show the results of experiments on the inhibition of Caco-2 cell proliferation by LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD, respectively.

[0033] Hemolytic activity: 100 μL of peptide sample was mixed with 100 μL of 4% erythrocyte suspension and incubated at 37 ℃ for 1 h. After centrifugation, the absorbance of the supernatant at 540 nm was measured. PBS was used as a negative control, and 0.1% Triton X-100 was used as a positive control. The hemolysis rate was calculated as [(A1-A2) / (A3-A2)]×100%. The hemolysis rates of the four target peptides were all below the 5% safety threshold in the range of 0.1-4.0 mg / mL, indicating good blood compatibility.

[0034] Oxidative damage protection: An oxidative damage model was established by treating HepG2 cells with 200 μM H2O2, reducing cell viability to approximately 54%. Pretreatment with different concentrations of the target peptide for 24 h was followed by H2O2 damage. CCK-8 results showed that the target peptide could restore cell viability in a dose-dependent manner, and at 4.0 mg / mL, cell viability was restored to near-normal control levels. Figure 13-16 The figures show the survival rates of HepG2 cells after treatment with LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD for oxidative damage.

[0035] Example 6: Intracellular ROS Scavenging and Keap1-Nrf2 Pathway Regulation HepG2 cells were seeded in 6-well plates. After cell adhesion, they were pretreated with the target peptide for 24 h, followed by treatment with 200 μM H2O2 for 24 h. Subsequently, ROS probes were added and incubated for 30 min. After washing with PBS, fluorescence intensity was measured. The results showed that H2O2 induced a significant increase in ROS, while the target peptide effectively reduced ROS fluorescence intensity, bringing it close to normal levels. Figure 17-20 The figures show the ROS content of HepG2 cells after treatment with LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD, respectively, to induce oxidative damage.

[0036] Total RNA was extracted from treated cells and reverse transcribed into cDNA. The mRNA expression levels of Keap1, Nrf2, HO-1, and NQO1 were detected by RT-qPCR, and the relative expression levels were calculated using 2^-ΔΔCt with GAPDH as an internal control. The results showed that the target peptides could remodel the Keap1-Nrf2 related antioxidant network; LFEEQMPFY and YPQRD significantly downregulated Keap1 expression, LFEEQMPFY promoted HO-1 expression, and YPQRD significantly promoted NQO1 expression.

[0037] Figure 21-24 The figures show the relative mRNA expression levels of HepG2 cells after treatment with LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD, respectively, to cells oxidatively damaged.

[0038] Example 7: Keap1 binding mechanism, molecular dynamics and quantum chemical verification Using the Keap1 Kelch domain crystal structure (PDB ID: 2FLU) as the target protein, molecular docking was performed with LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD as ligands to analyze their interactions with key residues in the Keap1 active pocket. After docking, 50 ns molecular dynamics simulations were performed using GROMACS with a CHARMM36 force field, an all-atom model, and a TIP3P water model. RMSD, RMSF, Rg, SASA, number of hydrogen bonds, ligand-acceptor centroid distance, and free energy landscape were analyzed. The results showed that all four peptides could form stable bindings in the Keap1 pocket, mainly interacting with residues such as Arg380, Asn382, Arg415, Tyr334, Ser363, and Pro384. The molecular docking diagrams for LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD are shown below. Figure 25-28 As shown.

[0039] DFT quantum chemical calculations were performed using the ORCA program, and structure optimization and single-point energy calculations were performed based on the B3LYP functional and def2-SVP basis set. The results showed that the four peptides have low HOMO-LUMO band gaps, suggesting that they have certain electron donor capabilities, which are beneficial for free radical scavenging. Figure 29-32 The figures show the quantum chemical calculation results for LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD, respectively.

[0040] Example 8: Verification of the freshness-preserving function of DAYPSG and YPQRD LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD were input into the VIRTUOUS platform for umami prediction. The results showed that DAYPSG had an umami probability of 88%, and YPQRD had a umami probability of 100%. Further validation was performed using an electronic tongue system. The reference solution was simulated saliva containing 30 mM KCl and 0.3 mM tartaric acid. The peptide sample concentration was 0.1 mg / mL. Each sample underwent four measurement cycles, discarding the first cycle and taking the average of the remaining three cycles. The results showed that both DAYPSG and YPQRD produced positive responses in the umami dimension, with YPQRD showing a slightly higher umami response. The electronic tongue results are as follows: Figure 33 As shown.

[0041] A T1R1 / T1R3 fresh receptor homology model was constructed, and DAYPSG and YPQRD were molecularly docked to the receptor active cavity, respectively. The results showed that both peptides could stably anchor to the receptor active cavity. The molecular docking diagrams of DAYPSG and YPQRD with T1R1 / T1R3 are shown below. Figure 34 and Figure 35 As shown, it forms an interaction network dominated by hydrogen bonds, which supports its umami recognition mechanism at the molecular level.

[0042] This invention provides four short peptide sequences derived from yak milk protein: LFEEQMPFY, YIPQMAGD, DAYPSG, and YPQRD. All of these sequences have been verified through in vitro free radical scavenging, cell protection, and safety evaluations, and show promising prospects for antioxidant applications. They can be used to prepare antioxidant functional products, which can be food, dairy products, beverages, functional foods, health foods, nutritional supplements, cosmetics, feed, or feed additives.

[0043] This invention further confirms that DAYPSG and YPQRD possess both antioxidant activity and umami properties, and can be used as antioxidant-umami bifunctional peptides. They can be applied in the preparation of umami-enhancing, flavor-improving, or low-sodium foods, which can be dairy products, fermented milk, milk beverages, flavored foods, low-sodium foods, or health foods.

[0044] This invention establishes a comprehensive technical route for yak milk protein multi-enzyme simulated hydrolysis—short peptide library construction—dual-branch deep integration model screening—functional and safety verification—flavor verification, which can efficiently and preferentially screen high-confidence candidate peptides from 134,109 short peptide candidate sequences of 4-10 amino acids.

[0045] The prediction model of this invention integrates Energy-Based Model (EBM) and Prototypical Network (PROTO), and combines hard negative sample mining, semi-supervised learning, knowledge distillation of unlabeled candidate peptides, and dynamic weighted ensemble, which can reduce the false positive rate in short peptide screening.

[0046] This invention not only obtains the sequence itself, but also provides multi-level verification evidence, including free radical scavenging, protection against H2O2-induced oxidative damage in HepG2 cells, reduction of intracellular ROS, transcriptional regulation of the Keap1-Nrf2 pathway, Keap1 binding stability, T1R1 / T1R3 fresh receptor binding, and fresh response of the electronic tongue.

[0047] All four selected peptides were predicted to be non-toxic, without significant allergen matching, and non-hemolytic; in cell experiments, they showed no significant inhibition of HepG2 and Caco-2 cell proliferation within the range of 0.1-4.0 mg / mL, providing a safe basis for their development as food-derived functional ingredients.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. YPQRD, a yak milk-derived antioxidant peptide screened based on deep learning, is characterized by: Its amino acid sequence is shown in SEQ ID NO:

4.

2. Application of yak milk-derived antioxidant peptides DAYPSG and / or YPQRD, screened based on deep learning, in the preparation of products with antioxidant functions.

3. The application according to claim 2, characterized in that, The products with antioxidant functions include food, cosmetics, or animal feed.

4. Application of yak milk antioxidant peptides DAYPSG and / or YPQRD selected based on deep learning in the preparation of flavor-enhanced or low-sodium foods.

5. The application according to claim 4, characterized in that, The flavor-enhanced or low-sodium foods include dairy products or flavored foods.

6. An antioxidant composition, characterized in that, This includes antioxidant peptides with the sequence DAYPSG and YPQRD.

Images