A novel method for exploration and prediction of the bioactive target of rice bran-derived peptide (KF-8) by integrating computational methods and experiments

Abstract

The investigation into the bioactive peptide's activity and target action poses a significant challenge in the field of food. An active peptide prepared from rice bran, KF-8, was confirmed to possess antioxidant activity in our previous study, but the specific target was unclear. This study used eight target prediction tools based on artificial intelligence and chemoinformatics to preliminarily screen potential antioxidant targets by integrating different computational methods. Then five different types of docking software were comparatively analyzed to further clarify their interaction sites and possible modes of action. The results showed that SIRT1 and CXCR4 are potential antioxidant targets of KF-8. Different docking software suggested that KF-8 interacts with SIRT1 and CXCR4 as major residues. Meanwhile, the results of Immunofluorescence co-localization experiments showed that the co-localization coefficients of KF-8 with SIRT1 and CXCR4 reached 0.5879 and 0.5684. This study provides new alternative means for the discovery of active peptide targets.

---

1. Introduction

Bioactive peptides are a class of short-chain amino acid sequences that are produced during protein catabolism or synthesis. Their unique amino acid sequences endow them with various biological functions, such as antioxidant, anti-inflammatory, antimicrobial, immunomodulatory, etc. Their diverse functions have garnered academic attention in the health field, and have been studied extensively in drug research and development.^1,2 Active peptides have been found in various foods, plants, animals, and microorganisms.^3,4 An increasing number of polypeptides based on food components are being studied to identify potential bioactivities and health benefits given the soaring demand for health and nutrition.^5,6 Rice bran is a major food processing by-product that has long been recognized as a limited agricultural waste.⁷ However, the focus of attention on rice bran has gradually transitioned to a class of bioactive ingredients present in it – rice bran active peptide. Rice bran active peptide is a bioactive peptide molecule extracted from rice bran. It has various uses and benefits, including prospective applications in the medical area.⁸ A bioactive peptide KF-8 (KHNRGDEF) containing 8 amino acids was extracted from rice bran protein with a molecular weight of 1002.6 Da.⁹ There is no significant difference in the rotation of KF-8 at different resting times and different concentration solutions, indicating its excellent stability.¹⁰ Based on our previous studies, KF-8 in rice bran protein was found to have an excellent anti-inflammatory effect, antioxidant activity and anti-aging activity, but the ways in which KF-8 regulates oxidative stress in vivo, as well as the specific target and mechanism of action remain unclear.¹¹ The comprehension of the exact targets of these active peptides is crucial for drug formulation and biomedical research. However, conventional experimental methods are time-consuming and labor-intensive, and more efficient methods amply warrant research.

Target virtual screening and molecular docking techniques are important components of computer-aided drug design.¹² Among them, the strategy of target virtual screening is to identify potential drug targets with multiple small-molecule pharmacological effects. Artificial intelligence (AI) and cheminformatics approaches have made remarkable progress in the field of target prediction and discovery over the past few decades, yielding new possibilities.^13–18 Chemo-informatics methods, including physicochemical property calculations, quantitative structure–activity relationship (QSAR) studies, and similarity calculations, amongst others, have been applied extensively in drug discovery to address issues like drug absorption, drug target prediction, virtual screening of new molecular entities, and more.^19–22 These methods are based on known compound activity data and predict the biological activity and possible targets of new compounds by calculating similarities or structure–activity relationships between molecules. However, these methods are often limited by the data quality and cannot capture complex non-linear relationships.^23–25 Given the advent of machine learning and deep learning algorithms, it has become possible to learn from large-scale compounds and biological information to identify biologically active pathways and molecular targets that may be affected by food and drug ingredients. These algorithms can handle a wide range of data types, such as molecular structure, chemical properties, biological activity, interaction information, etc., to further improve the accuracy of target prediction.^26–30 Petinrin et al. successfully predicted heterogeneous bioactive molecules during drug discovery using the stacked ensemble of multiple base classifiers (SVM, DT, KNN, and RF).³¹ Méndez-Lucio et al. report DeepDock, a geometric deep-learning-based method capable of predicting the binding conformation of ligands to protein targets.³² Kang et al. developed a new deep learning model (highlighting target sequences; HoTS) to successfully identify 16 unique and novel hit compounds.³³ Alkhadrawi et al. performed a virtual screening of membrane transporter proteins using a strategy combining deep learning prediction of drug–target interactions (DTI) and molecular docking to study the mechanism of glycyrrhetinic acid (GA) transport in intra- or extracellular cells.³⁴ Given the advent of emerging computer platforms for potential target identification and screening, TargetNet is a target prediction tool based on QSAR modeling, and PharmMapper is a target identification tool based on inverse pharmacophore matching that is widely used in drug research and development and in the field of food science.³⁵

The strategy of molecular docking is a theoretical simulation method for studying intermolecular (e.g., ligand and receptor) interactions and predicting their binding modes and binding affinities.^36–38 Individual molecular docking methods allow precise interaction simulations between specific food peptides and targets, thereby providing information on binding modes between molecules, affinities, and potential target sites.^36,39 Some studies have successfully interpreted the mechanism of action and structural information of different food peptides through molecular docking.^22,40,41 Different molecular docking methods based on structure or pharmacophore may be applicable to different situations.^42–44 A series of docking tools using different space search methods and scoring functions are significantly advancing this field.

In this study, the antioxidant targets of KF-8 were predicted and validated by comparing, selecting, and applying multiple deep learning-based target screening and molecular docking tools, combined with in vitro experiments. The mechanism of its antioxidant action was delineated at the molecular level, and a new strategy for exploring the bioactivity of KF-8-based peptides was proposed. First, the potential targets of action of the active peptide were obtained by advanced target prediction methods based on the target screening strategy, and the most likely relevant targets of the active peptide were identified. The most relevant targets were selected for comprehensive study through literature research analysis. The possible active targets of action of the target active peptide were subsequently verified by comparing multiple molecular docking software and validated by in vitro experiments to decipher its mechanism of action at the molecular level. This study provides an effective method to delineate the interactions between bioactive peptides and biological macromolecules based on KF-8. The results suggest that KF-8 may become an active ingredient in functional foods and this method will be potentially applied to the study of more food-derived bioactive peptides, which will subsequently provide a theoretical basis for the development of peptide-type functional products by researchers and guide the development of functional peptide foods.

2. Materials and methods

2.1 Chemoinformatics methods

The literature was reviewed to construct a rational target prediction process. It was found that different target prediction tools have different prediction principles, which were based on the principles of similarity, based on pharmacophore matching, and based on combinations to predict potential targets of compounds. Different methods have different application goals, such as screening efficiency. The advanced AI-driven target prediction tools were found in this study. The tools SEA,⁴⁵ SwissTargetPrediction,^46,47 PPB2,^48,49 TargetNet,²¹ PharmMapper,⁵⁰ SuperPred,^51–53 TargetHunter,²⁶ and PASS Targets⁵⁴ were used for target prediction of KF-8. Daylight molecular fingerprints were used by SEA to calculate the similarity of the compounds, and cluster analysis was performed on the targets of similar compounds to provide a list of potential targets. SwissTargetPrediction combines 2D and 3D similarity to predict targets through multiple logistic regression model training. PPB2 (Polypharmacology Browser 2) uses 2D molecular fingerprint similarity for target prediction. TargetNet builds a large number of QSAR models based on current chemical genomics data to predict interactions between drugs and targets. PharmMapper uses pharmacophore mapping methods to identify potential drug targets. TargetHunter uses 2D molecular fingerprint similarity to predict the target of a compound. SuperPred uses a logistic regression machine learning model for target prediction, which allows the model to learn the relationship between the compound structure and the target. PASS Targets uses statistical methods and known biological activity data to predict the target of a compound. The user selects different methods for target prediction through the input of SMILES of the target compound, the compound structure file (SDF, MOL, etc.) or the custom structure. Although the scoring criteria of different tools differ, combining the results can overcome the application limitations of a single tool and yield more accurate prediction results. Three different screening strategies were adopted: Top20 can yield the top-ranked targets of each prediction tool, and Top50 appropriately relaxes the restriction based on Top20, which was conducive to the inclusion of some targets that are unique to the tool. Moreover, a third strategy was also devised to screen according to the thresholds recommended by the tools to further consider the prediction mechanism of the tools themselves. The comparative analysis of these three strategies can result in more accurate and comprehensive predictions.

2.2 Molecular docking

Molecular docking studies were performed using the small molecule docking tools MOE⁵⁵and LeDock,⁵⁶ the protein–peptide docking programs HPEPDOCK,⁵⁷ HADDOCK,^58,59 and MDockPeP2.⁶⁰ MOE integrates a series of advanced sampling algorithms and scoring functions, which possesses the advantages of increased comprehensive functionality and cross-platform compatibility. LeDock performs conformational sampling through a combination of simulated annealing and evolutionary optimization, and can perform high-throughput molecular docking with multiple targets and ligands with good docking accuracy. Peptide–protein docking tools such as HPEPDOCK, HADDOCK and MDockPeP2 fully consider the flexibility of peptides, local and global docking; therefore, they are apt at predicting the structure of protein–peptide complexes. The difference between the docking algorithm of the small molecule docking tool and the difference of the protein–peptide docking program on flexible peptide docking principle were combined to study ligand receptor interaction more accurately. For protein crystal complex selection, crystal structure 4ZZJ (PDB ID: 4ZZJ)⁶¹ was selected for the SIRT1 protein, while crystal structure 3ODU (PDB ID: 3ODU)⁶² was selected for the CXCR4 protein.

Two different docking strategies in induced fit docking were adopted using MOE software. The first docking strategy involves docking with the active site of the original ligand of the crystal structure of the protein complex as the active pocket and the expansion of the original ligand pocket to 4.5 Å as the active site. Conversely, the second docking strategy involves preferentially performing a conformational search for polypeptide KF-8, followed by docking with the active site of the original ligand as the active pocket as well as the expansion of the original ligand active pocket to 4.5 Å as the active mouth site. Water molecules, ions, and non-canonical amino acid residues were separated from the protein and hydrogen atoms were added under the AMBERT10 force field to treat the proteins. After automatically correcting the protein structure using the “Structure Preparation” module during the protein pretreatment stage, the binding site is selected based on the original ligand, and the active pocket is expanded to 4.5 Å. The prepared peptide KF-8 and the ligand searching for the peptide KF-8 conformation were then flexibly docked into the receptor using the “Triangle Matcher” placement method and “London dG” scoring, as well as other default parameters. Finally, five docking poses were obtained, and the Pose with the highest score and the best pose was selected for subsequent analyses. Other molecular docking tools use the ligand–receptor preparation step followed by docking and the top three best docking poses were selected for analysis.

2.3 Experimental material

Human Umbilical Vein Endothelial Cells (HUVEC) (cat. no. PCS-100-010), Human Renal Epithelial Cell Line (293T) (cat. no. CRL-3216) were procured from ATCC. (Maryland, USA). Rice bran active peptide (KF-8) was synthesized by ChinaPeptides Co., Ltd (Shanghai, China). DMEM medium (cat. no. 11960044), penicillin–streptomycin solution (100×) (cat. no. 15140122), fetal bovine serum (cat. no. 12483020), and 0.25% trypsin with EDTA (cat. no. 25200056) were purchased from Gibco (CA, USA). DMSO (≥99.7%) (cat. no. D8418) was purchased from Sigma-Aldrich Co. (St Louis, MO, USA). TritonX-100 (cat. no. IT9100) and DAPI (cat. no. D8200) dyes were purchased from Beijing Solarbio Science & Technology Co., Ltd (Beijing, China). 4% paraformaldehyde (cat. no. P0099), anti-fluorescence quenching blocking solution (cat. no. P0126) were purchased from Beyotime Biotech Inc. (Shanghai, China). SIRT1 (cat. no. 60303-1-Ig), CXCR4 monoclonal antibody (cat. no. 60042-1-Ig), goat anti-mouse fluorescent secondary antibody (cat. no. SA00013-3) were purchased from Proteintech Group, Inc. (Chicago, USA). Plasmid extraction kit (cat. no. HY-K1081), protamine sulfate (cat. no. HY-107911) and puromycin dihydrochloride (99.89%) (cat. no. HY-B1743A) were purchased from MedChemExpress LLC. (New Jersey, USA). Restriction endonuclease QuickCut™ EcoR I (cat. no. 1611) and QuickCut™ Xho I (cat. no. 1635) were purchased from Takara Bio Inc. (San Jose, CA, USA). Ligase (cat. no. EG21202) was purchased from Yugong Biotech Co., Ltd (Jiangsu, China). Lentiviral LV-KF-8-EGFP, LV-KF-8-GST, and LV-EGFP were purchased from Ori-Biotechnology Co., Ltd (Hunan, China).

2.4 Constructing cell lines

The fusion expression of KF-8 with EGFP and KF-8 with GST was designed, respectively. KF-8 was designed at the N-terminal of the fusion protein, and a link (Gly-Gly-Gly-Gly-Ser) was added to the fusion protein. The related sequences were cloned into the lentiviral vector pLVX-puro, and the recombinant plasmids pLVX-Puro-KF-8-link-EGFP and PLVX-Puro-KF-8-link-GST were obtained. The lentivirus packaging system was then used to prepare lentiviruses LV-KF-8-EGFP and LV-KF-8-GST. LV-EGFP was used as a control lentivirus.

HUVEC was cultured and infected with LV-KF-8-EGFP, LV-KF-8-GST and LV-EGFP, respectively, under identical conditions. Puromycin with a final concentration of 5 μg mL⁻¹ was given for maintenance culture 96 h after infection. The expression of KF-8, EGFP and GST was determined by immunofluorescence detection, qPCR and western-blot, and the polyclonal cell lines HUVEC/KF-8⁺EGFP⁺, HUVEC/KF-8⁺GST⁺ and HUVEC/EGFP⁺ were screened.

2.5 Cell viability assay

Cell viability was determined using the CCK-8 assay. Cells of Control group HUVEC, HUVEC/KF-8⁺EGFP⁺, HUVEC/KF-8⁺GST⁺, and Control strain HUVEC/EGFP⁺ group were inoculated in 96-well culture plates at 4000 cells per well for spreading, the medium was set at 100 μL, and the replicate wells were set in parallel thrice for each group. Hydrogen peroxide was added at 40 μM per well after incubating at 37 °C and 5% CO₂ incubator for 6 h. The cells were incubated for 24 h. 10 μL of CCK-8 was added with 100 μL of culture medium, and the cells were incubated for 1 h in light-deprived conditions. The absorbance value at 450 nm was measured by a multifunctional enzyme labeling instrument. The following formula was used to calculate cell viability:

Cell viability = OD_t − OD₀/OD₀ × 100% (1)

where OD_t stands for the optical density value of the cell culture solution after 6 h, and OD₀ represents the optical density value of the cell culture solution at the beginning of the culture.

2.6 Immunofluorescence co-localization

The cells that were climbed well were fixed, permeabilized, and then closed with 5% bovine serum albumin. They were subsequently incubated and re-stained with nuclei by primary and secondary antibodies, and the images were ultimately collected by observing under a laser confocal microscope. The co-localization factor is calculated as:where x_i and x̄ denotes the i-th pixel intensity and the average pixel intensity after threshold removal in the channel 1 image, respectively. Similarly, y_i and ȳ are the i-th pixel intensity and average pixel intensity of the corresponding channel 2 images.⁶³

2.7 Statistical analysis

The data of this study were analyzed and processed using SPSS statistical analysis software, and plots were analyzed using GraphPad Prism 8. p < 0.01 indicates a statistical difference.

3. Results

3.1 Prediction and selection of potential targets for peptide KF-8

Different tools were used to yield potential targets of active peptide KF-8 and the target prediction results are shown in Fig. 1. Although the criteria and extent of target prediction cannot be standardized across tools, the poll results will provide evidence of protein–peptide interactions from multiple perspectives (e.g., similarity, pharmacophore, and chemogenomics). A series of results were obtained for each of the eight target prediction tools based on different prediction principles, which were analyzed using three different filtering strategies, as shown in Fig. 1A: the three filtering strategies of the top 20 targets (Top20), the top 50 targets (Top50), and the thresholds recommended by the reference tool itself. Using voting, the targets obtained from each of the three screening strategies were filtered, and the top 50 targets for each strategy were selected to summarize the 63 most likely potential targets of action for KF-8. Their voting counts are shown in Fig. 1B. Finally, the 63 targets were analyzed using the GeneCards database and KEGG database, and 12 oxidation-related targets were obtained, as shown in Fig. 1C after reviewing related literature studies in PubMed. SIRT1 was involved in various closely linked cellular processes: senescence, longevity, and oxidative stress, making them particularly important targets. Meanwhile, the existing reported target CXCR4 is linked to apoptosis, autophagy, and inflammation, which are processes that often involve oxidative responses. Whether this target is an important regulatory gene for oxidative stress is less reported. Therefore, CXCR4 and SIRT1 targets were screened for molecular docking for further study to determine whether this target is important for oxidative stress.

Fig. 1 Target prediction results. (A) Target hits of all tools under three different screening strategies. The first screening strategy is Top20 (S1) to keep the predicted targets, the second strategy is Top50 (S2) to keep the predicted targets, and the third strategy is keeping the predicted targets by referring to the threshold recommended by the tool itself (S3). (B) The number of hits on the top 50 targets for each of the three different filtering strategies. (C) The Venn diagram of the intersection of GeneCards, PubMed, and KEGG common target.

3.2 Molecular docking results

Molecular docking was used to investigate the specific structural basis and molecular mechanism of whether peptide KF-8 binds to SIRT1 and CXCR4 and thus acts as an antioxidant. Currently, different docking software have their spatial search methods and scoring functions. The use of multiple molecular docking software to dock the same set of molecules can overcome the limitations of different docking programs and improve the accuracy and reliability of the docking results, thereby better predicting the interaction and binding modes between molecules.

The docking of different proteins (SIRT1 and CXCR4) to KF-8 was first processed by self-docking using the original ligands in the crystal structure. Self-docking results indicate that the system has a set of appropriate docking parameters for KF-8 for MOE, LeDock. Furthermore, there was a 7-amino acid peptide substrate derived from p53 (Ac-p53) in the crystal structure of the SIRT1 protein complex. It would be more practical to dock at the site of Ac-p53 to investigate whether KF-8 would act as a substrate in SIRT1 proteins. KF-8 was docked to the SIRT1 protein using the site of Ac-p53 as the active pocket, with a docking scoring score of −11.33 kcal mol⁻¹. Through a comparison with Ac-p53 self-docking, it was concluded that Ac-p53 has better scoring (−9.55 kcal mol⁻¹) at this site and binds more stably to the protein. KF-8 has a relatively low probability of interacting with SIRT1 by occupying the site of Ac-p53. Therefore, it was reasonably assumed that the KF-8 active site was at the site of reference activator 4TQ. KF-8 was subsequently docked with the protein using two small molecule docking software and three protein–peptide docking tools. The docking results showed that the best pose scoring for SIRT1 protein binding to KF-8 was −7.93 kcal mol⁻¹ and −6.17 kcal mol⁻¹ in the small molecule docking tools MOE and LeDock, respectively. As inferred from the docking results of the small molecule docking software MOE, LeDock in Fig. 2, Asn226, Glu208, Leu205, and Glu230 were the main residues of KF-8 interacting with SIRT1, forming interactions such as hydrogen bonding and ionic bonding. Fig. 2 protein–peptide docking software HPEPDOCK, HADDOCK, and MDockPeP2 in which KF-8 achieved better docking results with the SIRT1 protein. Multiple protein–peptide docking software suggests that Asn226, Glu230, and Gln222 were important residues for interacting peptide KF-8 with proteins. However, in Fig. 3 depicting small molecule docking tools MOE and LeDock, the best pose scoring for CXCR4 protein binding to KF-8 was −15.1 kcal mol⁻¹ and −9.04 kcal mol⁻¹, respectively. KF-8 formed hydrogen-bonding interactions with residues such as Asp97, Glu288, Asp187, Glu32, and Cys186. Moreover, the protein–peptide docking software has a better scoring of peptide KF-8 interacting with CXCR4 protein. Visualization was performed using MOE, and their interactions were shown in the 2D and 3D plots of Fig. 3, where hydrogen bonding interactions between KF-8 and active center residues were observed.

Fig. 2 Docking results of KF-8 and SIRT1 in five different tools. 2D interaction maps for the best poses of KF-8 on SIRT1 (PDB ID: 4ZZJ). 3D binding mode map of the key residues involved in the interaction between KF-8 (green) and SIRT1 (PDB ID: 4ZZJ).

Fig. 3 Docking results of KF-8 and CXCR4 in five different tools. 2D interaction maps for the best poses of KF-8 on CXCR4 (PDB ID: 3ODU). 3D binding mode map of the key residues involved in the interaction between KF-8 (green) and CXCR4 (PDB ID: 3ODU).

The docking results of the Small Molecule Docking Tools and the Protein–Peptide Docking Tools are shown in Fig. 4. Fig. 4A and B display the major interaction residues Gln222, Glu208, Asn226, and Glu230 of KF-8 binding to SIRT1. The binding of KF-8 to SIRT1 was shown to be associated with 12 residues, including Ser229, Thr209, and Pro470, in addition to the same key residues Thr219, Ile223, Asn226, Ile227, Gln222, and Glu230, which have been reported in the literature as key residues of the SIRT1 activator 4TQ.⁶¹Fig. 4C and D show the important residues of KF-8 interacting with CXCR4: Asp97, Glu288, Asp187, Arg188, Glu32, Cys186, Asp181, His113, Asp182, Ser285. Indeed, the key residues of KF-8 interacting with CXCR4 were consistent with the important residues Trp94, Asp97, Tyr116, Arg183, Ile185, Cys186, Asp187, Glu288, His113, Asp171, Arg188, Tyr190, Asp193, and Asp262 reported by CXCR4 antagonists ITD and CVX15,⁶² in addition to 16 residues related to Glu32, His203, and Gln200. The binding potential of protein–peptide should be evident by the interaction residues, docking scoring, number of hydrogen bonds, number of residues, and type of bond. The best binding patterns were obtained for each tool based on the outputs. Overall, the ligand molecule can bind to the protein pockets, which complements the pockets in shape and electrostatic distribution, and be maintained by enough interaction forces.

Fig. 4 Frequency analysis results of interacting amino acids in the docking. (A) Frequency distribution diagram of SIRT1 amino acid residues interacting with KF-8 in the optimal poses (four poses for MOE and three for LeDock). (B) Frequency distribution diagram of SIRT1 amino acid residues interacting with KF-8 in the optimal poses (Three poses for all the peptide docking software). (C) Frequency distribution diagram of CXCR4 amino acid residues interacting with KF-8 in the optimal poses (four poses for MOE and three for LeDock). (D) Frequency distribution diagram of CXCR4 amino acid residues interacting with KF-8 in the optimal poses (Three poses for all the peptide docking software).

3.3 Construction of cell lines and determination of cell viability

HUVEC/KF-8⁺EGFP⁺, HUVEC/KF8⁺GST⁺, and HUVEC/EGFP⁺ stable cell lines were constructed using lentivirus (Table S1†), and the fluorescence expression of the three stable cell lines and the original HUVEC was observed by fluorescence inverted microscope. The results are shown in Fig. 5(A–D). The green fluorescence of HUVEC/KF-8⁺EGFP⁺ and HUVEC/EGFP⁺ cells in both groups was distinguishable, and no green fluorescence signal was detected in the original HUVEC and HUVEC/KF-8⁺GST⁺. Compared with the HUVEC group, the cell survival rate of the HUVEC + 40 μM H₂O₂-treated group was 71.92% (P < 0.01); compared with the HUVEC/EGFP⁺ group, the cell survival rate of the HUVEC/EGFP⁺ + 40 μM H₂O₂-treated group was 67.423% (P < 0.01); there was no statistically significant difference in cell survival in the HUVEC/KF-8⁺EGFP⁺ + 40 μM H₂O₂-treated group compared to the HUVEC/KF-8⁺EGFP⁺ group, and there was no statistically significant difference in cell survival in the HUVEC/KF-8⁺GST⁺ + 40 μM H₂O₂-treated group compared to the HUVEC/KF-8⁺GST⁺ group. The results showed that rice bran active peptide had a protective effect against H₂O₂-induced oxidative damage in HUVEC cells.

Fig. 5 Fluorescence expression of four cell lines of KF-8 and cell viability results. (A–D) Fluorescence detection results of the three stable cell lines and the original HUVEC; (E) cell viability assay of the three stable cell lines and the original HUVEC cells. Note: ****: p < 0.0001 compared with the HUVEC group and ####: p < 0.0001 compared with the HUVEC/EGFP + group.

3.4 Immunofluorescence staining of cells

SIRT1 and CXCR4 proteins were detected by immunofluorescence experiments in primary HUVEC cells (CON group) and HUVEC/KF-8⁺EGFP⁺ cells, respectively Images were captured by laser confocal microscopy, and the results are shown in Fig. 6. The expression of SIRT1 CXCR4 proteins (red fluorescence) was examined in both CON and HUVEC/KF-8⁺EGFP⁺ cells. The co-localization Pearson coefficients (PCCs) of SIRT1 (red fluorescence) with KF-8-EGFP fusion proteins (green fluorescence) in HUVEC/KF-8⁺EGFP⁺ cells were 0.5879; and the co-localization Pearson coefficients (PCCs) of CXCR4 (red fluorescence) with KF-8-EGFP fusion proteins (green fluorescence) in HUVEC/KF-8⁺EGFP⁺ cells were 0.5684. The results show that the KF-8-EGFP fusion protein co-localized with SIRT1 and CXCR4 proteins, respectively.

Fig. 6 Single channel and Merge plots of different HUVEC cell lines after double staining. The red color represents SIRT1 and CXCR4, the green color represents KF-8, and cell nuclei were stained with DAPI (blue). All cells were observed at the same magnification.

4. Discussion

This study used several AI-based algorithmic tools to predict whether a bioactive molecule binds to a target protein. While both experimental and computational methods can be used to predict biological targets that interact with bioactive molecules, traditional experimental methods are typically more costly and slower than computational methods. The use of intelligent algorithms is an effective way to increase the efficiency of bioactive molecule research. Several target prediction tools driven by powerful algorithms were used to predict the possible targets of compounds. Their methodologies include molecular similarity, pharmacophore, and QSAR, etc. In terms of screening efficiency, these tools have different application focuses because different tools have different time and computational power requirements during the screening process. For example, while TargetHunter and PharmMapper generate predictions slower than SEA and TargetNet, these tools can generate predictions much faster. Different coverage of targets of the tools is also a factor, which may lead to the prediction of results that are partial targets of other species. However, this will result in the omission of some protein targets, which is why the Top20, Top50, and other three strategies in this study were used to perform the comprehensive analysis. These strategies can be filtered to complement one another in terms of coverage. Moreover, the wide range of these tools combined with the database size will produce more comprehensive target results. For example, SEA uses compound and target information from databases such as ChEMBL and MDDR, which contain over 65 000 compounds and hundreds of protein targets. Conversely, PASS Targets uses numerous public data sources, e.g., DrugBank, PubChem, ChEMBL, ChemSpider, and ChEBI, and can predict the interactions of 2507 protein targets from different organisms with drug-like compounds.⁵⁴ SuperPred used protein–ligand interaction data obtained in ChEMBL, BindingDB, and SuperTarget to construct reference ligand datasets for protein targets. Moreover, different tools differ in terms of interface friendliness. Some provide more detailed explanations and data, which can help users better understand and evaluate the prediction results, such as SEA, SwissTargetPrediction, PharmMapper, TargetNet, and SuperPred, which provide information related to the potential target of the predicted substance in addition to the reliability of the target through the corresponding numerical values. However, tools such as TargetHunter also directly provides information regarding the biological activity of similar compounds. In terms of the number of predicted targets, SEA, PharmMapper, TargetNet, and SuperPred reported more than others, and it was possible that more targets were covered by these tools. The screening accuracy of various prediction tools is contingent upon the prediction principles and their data sources. Target screening thresholds such as P-value, probability, fit score, similarity score, and confidence will also impact accuracy. Analysis results showed that TargetNet, SuperPred, PharmMapper, and Polypharmacology Browse2 deliver better target prediction results for the active peptide KF-8. Furthermore, the high-scoring values obtained through these tools suggest that their output targets may have a closer association with KF-8. All of these predictive tools have novel theoretical strategies. However, each tool has its own strengths when used individually as mentioned earlier. The accuracy of the tool's predictions is paramount in the study; however, data size, target number, computation costs and user interface should not be overlooked. Moreover, these tools must be used in conjunction the creation of a knowledge-based appropriate strategy to identify targets with a high degree of credibility among the numerous findings.

In this study, tools based on different principles were first used to yield the potential targets of KF-8, and a voting method was then adopted to further improve the accuracy of target prediction through Top20, Top50, and the threshold strategy recommended by the reference tool itself. They were then filtered to yield 12 possible antioxidant-interacting targets of KF-8 through screening. Despite the yielding of a preliminary prediction of this potential role, the mechanism of action in the human body and the potential beneficial effects of the active molecule remained unclear. Therefore, molecular docking was used to confirm the study of biological targets of bioactive molecules, to help identify peptide KF-8 interactions with protein targets, and to understand the mechanism of action of bioactive molecules. However, different docking tools use different scoring functions and spatial search algorithms. KF-8 is a peptide with strong structural flexibility; therefore, small molecule docking tools and protein–peptide docking programs were used for docking to synthesize the docking results of different software and to overcome the limitations of the docking software's searching algorithm and scoring function, to boost the accuracy of the docking results. The molecular docking of KF-8 was analyzed with the protein mainly in terms of docking scoring, number of hydrogen bonds formed, number of residues, and type of bonds. It was found that the flexibility of the peptide during docking makes its optimal conformation in different docking software not identical; therefore, it could be a more reasonable approach to analyze the residue type, bonding type, and statistics in our study. Based on the results of the energy scores, it can be determined that KF-8 interacts with SIRT1 and CXCR4, and the ligand can interact with the receptor at the appropriate site and play the corresponding role. Hydrogen bonding can stabilize intermolecular binding, whereby the number of hydrogen bonds is related to the degree of stability of the intermolecular binding. According to the best poses from small molecule docking and protein–peptide docking results, the maximum number of hydrogen bonds formed by KF-8 with SIRT1 and CXCR4 was 10 and 17, respectively, which exceeded the number of hydrogen bonds formed by the reference activator and antagonist with the protein. The results indicate that the binding of KF-8 to SIRT1 and CXCR4, respectively, may be more stable. The protein–peptide interaction was further illustrated by counting the number of interacting residues in the top best docking poses in the small molecule docking and protein–peptide docking programs, whereby it was found that the maximum number of identical residues prompted by the interaction of KF-8 with SIRT1 and CXCR4 were 6 and 12, respectively. In terms of the type of bonding, there were more hydrogen bonds in the binding pattern of KF-8 with SIRT1 and with CXCR4 proteins, which was a stronger interaction than van der Waals and weaker than covalent and ionic bonds. Different types of residues can form different interactions, such as hydrophobic and electrostatic interactions. More various residue types indicate a greater variety of intermolecular interactions, which can also increase binding diversity and selectivity. Therefore, these evaluation indexes play an important indicative role in the study of protein–peptide binding and can also be directly applied to the analysis and study of other bioactive peptides. Based on the case study of KF-8, it can be speculated that KF-8 binds to SIRT1 and CXCR4 proteins, thus acting as an antioxidant. The case of KF-8 provides a theoretical basis for the study of other peptide activities.

Docking results from small molecule docking tools, and protein–peptide docking programs confirmed the possibility of peptide KF-8 binding to the target proteins SIRT1 and CXCR4, and KF-8 with optimal scores was stable in binding to the proteins. Cellular experiments confirmed that KF-8 had a significant protective effect on H₂O₂-induced oxidative damage cells and reduced the production of oxidative stress factors in vivo. Immunofluorescence co-localization experiments have shown that KF-8 can bind SIRT1 and CXCR4 and may play an antioxidant role by binding these two proteins. However, the antioxidant function of KF-8 is dependent on binding to these two proteins, and only intracellular proteins that directly interact with KF-8. Moreover, SIRT1 and CXCR4 have also been proven to bind to some active substances to exert antioxidant activities in the antioxidant action of some active substances. For example, α-mangostin binds to SIRT1 protein and regulates cellular oxidative damage through the SIRT1/3-FOXO3a axis.⁶⁴ SIRT1 activator SRT2104 enhances SIRT1-mediated deacetylation and inhibits neuronal apoptosis and senescence.⁶⁵ Silibinin blocked chemokine ligand 12 (CXCL12)-induced CXCR4 internalization by binding to CXCR4, thereby inhibiting downstream intracellular signaling, such as inhibition of CXCL12-induced intracellular calcium signaling and phosphorylation of Akt and Erk in breast cancer cells.⁶⁶ It was hypothesized that KF-8 can bind to SIRT1 and cause the expression of SIRT1 downstream proteins and thereby participate in the antioxidant effect, such as through the inhibition of Forkhead box O3 (FOXO3) and the activation of Peroxlsome proliferator-activated receptor-γ coactlvator-1α (PGC1α). It has been shown that α-mangostin significantly promotes the expression of FOXO3a transcription factors, which exert beneficial effects on cell survival and longevity. A molecular docking study predicted that α-mangostin binds directly to the active site of SIRT1 (Ile223, Leu215, and Pro212, as well as Arg446 on the CD-binding site of SIRT1) and that both α-mangostin and memantine hydrochloride can be located in the same protein-binding site of resveratrol.⁶⁷ SIRT1 deacetylates FOXO3 through direct protein interactions, thereby tilting the balance toward cell survival in response to oxidative stress. SIRT1 is able to deacetylate the FOXO3 transcription factor because FOXO3 deficiency reduces the protective effect of the SIRT1 activator SRT1720 (N-[2-[3-(piperazin-1-ylmethyl)imidazo[2,1-b][1,3]thiazol-6-yl]phenyl]quinoxaline-2-carboxamide) against cellular senescence.⁶⁸ However, KF-8 can bind to CXCR4 proteins and may inhibit oxidative stress by regulating the Phosphatidylinositol 3-kinase (PI3K)/protein kinase B (AKT)/mammalian target of the rapamycin (mTOR) pathway, the NADPH oxidase 4 (Nox4) pathway, and the inhibitor of kappa B kinase (IKK)/Nuclear factor kappa-B (NF-κB) pathway. It has been shown that naringin has a preventive effect on HUVEC cell damage induced by oxidative stress and inflammatory responses, which may be mediated by inhibition of the Nox4 and NF-κB pathways as well as inhibition of the activation of the PI3K/AKT pathway.⁶⁹ Hsueh et al. investigated that naringin inhibited the activation of the IKK/NF-κB pathway by Tumor necrosis factor-α (TNF-α). The experimental results showed that TNF-α significantly activated the phosphorylation of inhibitor of Nuclear Factor kappa-B Kinase alpha/beta (IKKα/β), inhibitor of Nuclear Factor kappa-B alpha (IκB-α) and NF-κB itself.⁷⁰ Guangying Lu et al. reviewed that ellagic acid reduces the expression of PIK3CA and PIK3R proteins and inhibits the phosphorylation of PI3K through PI3K/Akt/mTOR signaling pathway, which in turn down-regulates the expression level of Matrix Metalloproteinase-9 (MMP9). Moreover, the inhibition of mTOR activation reduces the expression level of β-catenin protein.⁷¹ Quercetin inhibits proliferation, promotes apoptosis, and reduces invasion and migration of DU145 cells by binding to Akt1, up-regulating Caspase-3 and down-regulating B-cell lymphoma 2 (Bcl-2) expression.⁷² Resveratrol prevented MMP reduction, attenuated neuronal apoptosis, and increased cell viability, thereby effectively protecting PC12 cells from 6-OHDA-induced oxidative stress and apoptosis.⁷³ Moreover, it has been shown that atractylenolide III may inhibit oxidative stress and autophagy by downregulating PI3K/AKT/mTOR expression.⁷⁴ Yoon Hyeun Oum et al. reported the successful application of a computational approach to the discovery and optimization of a new CXCR4-targeting small molecule, Z7R. They demonstrated its efficacy in an in vivo model with better anti-inflammatory activity.⁷⁵ Furthermore, four natural compounds, curcumin, resveratrol, quercetin, and eucalyptus brain, have been reported as inhibitors that interact with the CXCR4 receptor as active compounds in the treatment of coronary artery disease.⁷⁶ In our study, the reliability and accuracy of bioactive peptide target prediction were significantly improved by mutual verification of computational and experimental results. Based on our study, it was observed that investigating the downstream protein expression levels of KF-8 after its interaction with SIRT1 and CXCR4 was conducive to a comprehensive understanding of the biological activity of KF-8 and can be used as a research direction for subsequent in-depth excavation. In conclusion, the successful prediction of KF-8 antioxidant targets provides us with a complete set of applied processes and strategies for future studies in the activity of bioactive peptides. However, computational-based virtual screening of targets and molecular docking also have false-positive results, requiring more experimental validation time. However, using a combination of computational-based and in vitro experimental wet and dry methods to study the activity of substances may be more comprehensive and accurate.

5. Conclusion

In this study, artificial intelligence (AI) and chemoinformatics methods were used to virtually screen the antioxidant potential targets of the active peptide KF-8 from rice bran by studying and comparing multiple target prediction tools and molecular docking software and molecular docking-based mechanism exploration and in vitro experimental activity validations were performed. A multidisciplinary approach was applied to successfully screening KF-8 antioxidant targets and studying the mechanism of activity, providing a new set of fast, accurate, and green target screening strategies for applying other bioactive peptides. It is expected to provide valuable references for drug design studies as well as the application of functional peptides in the food industry. Nevertheless, the virtual screening and activity validation are still in the preliminary stage of our study, and the interaction between KF-8 and other targets and its potential drug delivery performance will be further explored in the future, which will be expanded to more applied studies of bioactive peptides.

Author contributions

Rui Liang: data curation, investigation, software, writing – original draft; Fangliang Song: investigation, validation, writing – original draft; Ying Liang: conceptualization, methodology, writing – review & editing, supervision, funding acquisition, project administration; Yanpeng Fang: formal analysis, validation; Jianqiang Wang: visualization; Yajuan Chen: formal analysis; Zhongxu Chen: formal analysis; Xiaorong Tan: software; Jie Dong: conceptualization, methodology, writing – review & editing, supervision.

Data availability

The data supporting this study's findings are available from the corresponding author upon reasonable request.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This work was supported by funding from the National Key Research and Development Program of China (2022YFF1100203), National Natural Science Foundation of China (No. 32372349), Science and Technology Innovation Talent Project of Hunan Province (No. 2022RC3056), Key Research and Development Project of Hunan Province (No. 2024AQ2020).

References

1. M. Akbarian, A. Khani, S. Eghbalpour and V. N. Uversky, Bioactive Peptides: Synthesis, Sources, Applications, and Proposed Mechanisms of Action, Int. J. Mol. Sci., 2022, 23(3), 1445.
2. S. K. Ulug, F. Jahandideh and J. Wu, Novel technologies for the production of bioactive peptides, Trends Food Sci. Technol., 2021, 108, 27–39.
3. A. A. Zaky, J. Simal-Gandara, J.-B. Eun, J.-H. Shim and A. M. Abd El-Aty, Bioactivities, applications, safety, and health benefits of bioactive peptides from food and by-products: A review, Front. Nutr., 2021, 8, 815640.
4. S. Zhang, L. Luo, X. Sun and A. Ma, Bioactive peptides: a promising alternative to chemical preservatives for food preservation, J. Agric. Food Chem., 2021, 69(42), 12369–12384.
5. N. Abd-Talib, E. L. A. Yaji, N. S. A. Wahab, N. Razali, K. Y. T. Len and J. Roslan, et al., Bioactive peptides and its alternative processes: A review, Biotechnol. Bioprocess Eng., 2022, 27(3), 306–335.
6. J. Stadnik and P. Kęska, Meat and fermented meat products as a source of bioactive peptides, Acta Sci. Pol., Technol. Aliment., 2015, 14(3), 181–190.
7. B. L. Tan, M. E. Norhaizan and L. C. Chan, Rice bran: from waste to nutritious food ingredients, Nutrients, 2023, 15(11), 2503.
8. Y. Q. Liu, P. Strappe, W. T. Shang and Z. K. Zhou, Functional peptides derived from rice bran proteins, Crit. Rev. Food Sci. Nutr., 2019, 59(2), 349–356.
9. Y. Liang, Q. Lin, P. Huang, Y. Wang, J. Li and L. Zhang, et al., Rice Bioactive Peptide Binding with TLR4 To Overcome H₂O₂-Induced Injury in Human Umbilical Vein Endothelial Cells through NF-κB Signaling, J. Agric. Food Chem., 2018, 66(2), 440–448.
10. J. Cai, Z. Chen, Y. Wu, Y. Chen, J. Wang and Q. Lin, et al., The rice bran peptide KF-8 extends the lifespan and improves the healthspan of Caenorhabditis elegans via skn-1 and daf-16, Food Funct., 2022, 13(5), 2427–2440.
11. Y. Wang, X. Cui, Q. Lin, J. Cai, L. Tang and Y. Liang, Active peptide KF-8 from rice bran attenuates oxidative stress in a mouse model of aging induced by d-galactose, J. Agric. Food Chem., 2020, 68(44), 12271–12283.
12. T. Seidel, D. A. Schuetz, A. Garon and T. Langer, The Pharmacophore Concept and Its Applications in Computer-Aided Drug Design, Prog. Chem. Org. Nat. Prod., 2019, 110, 99–141.
13. B. Liu, H. He, H. Luo, T. Zhang and J. Jiang, Artificial intelligence and big data facilitated targeted drug discovery, Stroke Vasc. Neurol., 2019, 4(4), 206–213.
14. Z. Wang, H. Li, C. Carpenter and Y. Guan, Challenge-enabled machine learning to drug-response prediction, AAPS J., 2020, 22(5), 106.
15. J. Augen, The evolving role of information technology in the drug discovery process, Drug Discovery Today, 2002, 7(5), 315–323.
16. K. Imai, K. Shimizu and H. Honda, Machine learning screening of bile acid-binding peptides in a peptide database derived from food proteins, Sci. Rep., 2021, 11(1), 16123.
17. Q. Liu, X. Gao, D. Pan, Z. Liu, C. Xiao and L. Du, et al., Rapid screening based on machine learning and molecular docking of umami peptides from porcine bone, J. Sci. Food Agric., 2023, 103(8), 3915–3925.
18. Q. Wu, H. Ke, D. Li, Q. Wang, J. Fang and J. Zhou, Recent progress in machine learning-based prediction of peptide activity for drug discovery, Curr. Top. Med. Chem., 2019, 19(1), 4–16.
19. Y. Chen and J. Kirchmair, Cheminformatics in natural product–based drug discovery, Mol. Inf., 2020, 39(12), e2000171.
20. Y. Chen, J. Qian, R. Liang, W. Zeng, J. Dong and Q.-L. Lin, Precise hapten design of sulfonamides by combining machine learning and 3D-QSAR approaches, Food Anal. Methods, 2022, 15(4), 1085–1097.
21. Z.-J. Yao, J. Dong, Y.-J. Che, M.-F. Zhu, M. Wen and N.-N. Wang, et al., TargetNet: a web service for predicting potential drug-target interaction profiling via multi-target SAR models, J. Comput. Aided Mol. Des., 2016, 30(5), 413–424.
22. R. Xiao, R. Liang, Y.-H. Cai, J. Dong and L. Zhang, Computational screening for new neuroprotective ingredients against Alzheimer's disease from bilberry by cheminformatics approaches, Front. Nutr., 2022, 9, 1061552.
23. I. L. Ruiz, Gómez-Nieto MÁ. Prototype selection method based on the rivality and reliability indexes for the improvement of the classification models and external predictions, J. Chem. Inf. Model., 2020, 60(6), 3009–3021.
24. L. Zhao, W. Wang, A. Sedykh and H. Zhu, Experimental errors in QSAR modeling sets: what we can do and what we cannot do, ACS Omega, 2017, 2(6), 2805–2812.
25. K.-K. Oh, H. Gupta, B. H. Min, R. Ganesan, S. P. Sharma and S. M. Won, et al., The identification of metabolites from gut microbiota in NAFLD via network pharmacology, Sci. Rep., 2023, 13(1), 724.
26. L. Wang, C. Ma, P. Wipf, H. Liu, W. Su and X.-Q. Xie, TargetHunter: an in silico target identification tool for predicting therapeutic potential of small organic molecules based on chemogenomic database, AAPS J., 2013, 15(2), 395–406.
27. Y. Zhou, Y. Jiang and S.-J. Chen, RNA-ligand molecular docking: advances and challenges, Wiley Interdiscip. Rev.: Comput. Mol. Sci., 2022, 12(3), e1571.
28. H. Li, K.-H. Sze, G. Lu and P. J. Ballester, Machine–learning scoring functions for structure–based virtual screening, Wiley Interdiscip. Rev.: Comput. Mol. Sci., 2021, 11(1), e1478.
29. Q. U. Ain, A. Aleksandrova, F. D. Roessler and P. J. Ballester, Machine-learning scoring functions to improve structure-based binding affinity prediction and virtual screening, Wiley Interdiscip. Rev.: Comput. Mol. Sci., 2015, 5(6), 405–424.
30. Y. Gu, J. Li, H. Kang, B. Zhang and S. Zheng, Employing molecular conformations for ligand-based virtual screening with equivariant graph neural network and deep multiple instance learning, Molecules, 2023, 28(16), 5982.
31. O. O. Petinrin and F. Saeed, Stacked ensemble for bioactive molecule prediction, IEEE Access, 2019, 7, 153952–153957.
32. O. Méndez-Lucio, M. Ahmad, E. A. Del Rio-Chanona and J. K. Wegner, A geometric deep learning approach to predict binding conformations of bioactive molecules, Nat. Mach. Intell., 2021, 3(12), 1033–1039.
33. K. M. Kang, I. Lee, H. Nam and Y.-C. Kim, AI-based prediction of new binding site and virtual screening for the discovery of novel P2X3 receptor antagonists, Eur. J. Med. Chem., 2022, 240, 114556.
34. A. M. Alkhadrawi, Y. Wang and C. Li, In-silico screening of potential target transporters for glycyrrhetinic acid (GA) via deep learning prediction of drug-target interactions, Biochem. Eng. J., 2022, 181, 108375.
35. Z. Shuli, L. Linlin, G. Li, Z. Yinghu, S. Nan and W. Haibin, et al., Bioinformatics and computer simulation approaches to the discovery and analysis of bioactive peptides, Curr. Pharm. Biotechnol., 2022, 23(13), 1541–1555.
36. G. M. Morris and M. Lim-Wilby, Molecular docking, Methods Mol. Biol., 2008, 443, 365–382.
37. L. Pinzi and G. Rastelli, Molecular docking: shifting paradigms in drug discovery, Int. J. Mol. Sci., 2019, 20(18), 4331.
38. L. H. S. Santos, R. S. Ferreira and E. R. Caffarena, Integrating molecular docking and molecular dynamics simulations, Methods Mol. Biol., 2019, 2053, 13–34.
39. S.-Y. Huang and X. Zou, Advances and challenges in protein-ligand docking, Int. J. Mol. Sci., 2010, 11(8), 3016–3034.
40. S. Liu, X.-L. Cao, G.-Q. Liu, T. Zhou, X.-L. Yang and B.-X. Ma, The in silico and in vivo evaluation of puerarin against Alzheimer's disease, Food Funct., 2019, 10(2), 799–813.
41. Z. Zhang, Y. Shang, S. Li, Z. Chen, J. Xia and Y. Tian, et al., Molecular docking revealed the potential anti-oxidative stress mechanism of the walnut polypeptide on HT22 cells, Foods, 2023, 12(7), 1554.
42. P. Polishchuk, A. Kutlushina, D. Bashirova, O. Mokshyna and T. Madzhidov, Virtual screening using pharmacophore models retrieved from molecular dynamic simulations, Int. J. Mol. Sci., 2019, 20(23), 5834.
43. B. Shaker, K. M. Tran, C. Jung and D. Na, Introduction of advanced methods for structure-based drug discovery, CBIO, 2021, 16(3), 351–363.
44. E. Yuriev, Challenges and advances in structure-based virtual screening, Future Med. Chem., 2014, 6(1), 5–7.
45. M. J. Keiser, B. L. Roth, B. N. Armbruster, P. Ernsberger, J. J. Irwin and B. K. Shoichet, Relating protein pharmacology by ligand chemistry, Nat. Biotechnol., 2007, 25(2), 197–206.
46. D. Gfeller, A. Grosdidier, M. Wirth, A. Daina, O. Michielin and V. Zoete, SwissTargetPrediction: a web server for target prediction of bioactive small molecules, Nucleic Acids Res., 2014, 42(Web Server issue), W32–W38.
47. A. Daina, O. Michielin and V. Zoete, SwissTargetPrediction: updated data and new features for efficient prediction of protein targets of small molecules, Nucleic Acids Res., 2019, 47(W1), W357–W364.
48. M. Awale and J.-L. Reymond, The polypharmacology browser: a web-based multi-fingerprint target prediction tool using ChEMBL bioactivity data, J. Cheminf., 2017, 9, 11.
49. M. Awale and J.-L. Reymond, The Polypharmacology Browser PPB2: Target Prediction Combining Nearest Neighbors with Machine Learning, 2018.
50. X. Wang, C. Pan, J. Gong, X. Liu and H. Li, Enhancing the Enrichment of Pharmacophore-Based Target Prediction for the Polypharmacological Profiles of Drugs, J. Chem. Inf. Model., 2016, 56(6), 1175–1183.
51. M. Dunkel, S. Günther, J. Ahmed, B. Wittig and R. Preissner, SuperPred: drug classification and target prediction, Nucleic Acids Res., 2008, 36(Web Server issue), W55–W59.
52. K. Gallo, A. Goede, R. Preissner and B.-O. Gohlke, SuperPred 3.0: drug classification and target prediction-a machine learning approach, Nucleic Acids Res., 2022, 50(W1), W726–W731.
53. J. Nickel, B.-O. Gohlke, J. Erehman, P. Banerjee, W. W. Rong and A. Goede, et al., SuperPred: update on drug classification and target prediction, Nucleic Acids Res., 2014, 42(Web Server issue), W26–W31.
54. P. V. Pogodin, A. A. Lagunin, D. A. Filimonov and V. V. Poroikov, PASS Targets: Ligand-based multi-target computational system based on a public data and naïve Bayes approach, SAR QSAR Environ. Res., 2015, 26(10), 783–793.
55. Chemical Computing Group (CCG) | Computer-Aided Molecular Design; 2023 [cited 2023 Mar 23]. Available from: URL: https://www.chemcomp.com/.
56. Z. Wang, H. Sun, X. Yao, D. Li, L. Xu and Y. Li, et al., Comprehensive evaluation of ten docking programs on a diverse set of protein-ligand complexes: the prediction accuracy of sampling power and scoring power, Phys. Chem. Chem. Phys., 2016, 18(18), 12964–12975.
57. P. Zhou, B. Jin, H. Li and S.-Y. Huang, HPEPDOCK: a web server for blind peptide-protein docking based on a hierarchical algorithm, Nucleic Acids Res., 2018, 46(W1), W443–W450.
58. C. Dominguez, R. Boelens and A. M. J. J. Bonvin, HADDOCK: a protein-protein docking approach based on biochemical or biophysical information, J. Am. Chem. Soc., 2003, 125(7), 1731–1737.
59. G. C. P. van Zundert, J. P. G. L. M. Rodrigues, M. Trellet, C. Schmitz, P. L. Kastritis and E. Karaca, et al., The HADDOCK2.2 Web Server: User-Friendly Integrative Modeling of Biomolecular Complexes, J. Mol. Biol., 2016, 428(4), 720–725.
60. X. Xu and X. Zou, MDockPeP: A Web Server for Blind Prediction of Protein-Peptide Complex Structures, Methods Mol. Biol., 2020, 2165, 259–272.
61. H. Dai, A. W. Case, T. V. Riera, T. Considine, J. E. Lee and Y. Hamuro, et al., Crystallographic structure of a small molecule SIRT1 activator-enzyme complex, Nat. Commun., 2015, 6, 7645.
62. B. Wu, E. Y. T. Chien, C. D. Mol, G. Fenalti, W. Liu and V. Katritch, et al., Structures of the CXCR4 chemokine GPCR with small-molecule and cyclic peptide antagonists, Science, 2010, 330(6007), 1066–1071.
63. J. S. Aaron, A. B. Taylor and T. L. Chew, Image co-localization–co-occurrence versus correlation, J. Cell Sci., 2018, 131(3), jcs211847.
64. A. Cereto-Massagué, M. J. Ojeda, C. Valls, M. Mulero, S. Garcia-Vallvé and G. Pujadas, Molecular fingerprint similarity search in virtual screening, Methods, 2015, 71, 58–63.
65. N. Chang, J. Li, S. Lin, J. Zhang, W. Zeng and G. Ma, et al., Emerging roles of SIRT1 activator, SRT2104, in disease treatment, Sci. Rep., 2024, 14(1), 5521.
66. Y. Wang, W.-C. Liang, W.-L. Pan, W.-K. Law, J.-S. Hu and D. T.-M. Ip, et al., Silibinin, a novel chemokine receptor type 4 antagonist, inhibits chemokine ligand 12-induced migration in breast cancer cells, Phytomedicine, 2014, 21(11), 1310–1317.
67. W. Ruankham, W. Suwanjang, K. Phopin, N. Songtawee, V. Prachayasittikul and S. Prachayasittikul, Modulatory Effects of Alpha-Mangostin Mediated by SIRT1/3-FOXO3a Pathway in Oxidative Stress-Induced Neuronal Cells, Front. Nutr., 2021, 8, 714463.
68. H. Yao, S. Chung, J. Hwang, S. Rajendrasozhan, I. K. Sundar and D. A. Dean, et al., SIRT1 protects against emphysema via FOXO3-mediated reduction of premature senescence in mice, J. Clin. Invest., 2012, 122(6), 2032–2045.
69. W. Li, C. Wang, J. Peng, J. Liang, Y. Jin and Q. Liu, et al., Naringin inhibits TNF-α induced oxidative stress and inflammatory response in HUVECs via Nox4/NF-κ B and PI3K/Akt pathways, Curr. Pharm. Biotechnol., 2014, 15(12), 1173–1182.
70. T.-P. Hsueh, J.-M. Sheen, J.-H. S. Pang, K.-W. Bi, C.-C. Huang and H.-T. Wu, et al., The Anti-Atherosclerotic Effect of Naringin Is Associated with Reduced Expressions of Cell Adhesion Molecules and Chemokines through NF-κB Pathway, Molecules, 2016, 21(2), 195.
71. G. Lu, X. Wang, M. Cheng, S. Wang and K. Ma, The multifaceted mechanisms of ellagic acid in the treatment of tumors: State-of-the-art, Biomed. Pharmacother., 2023, 165, 115132.
72. Y. Ning, Y. Wu, Q. Zhou and Y. Teng, The Effect of Quercetin in the Yishen Tongluo Jiedu Recipe on the Development of Prostate Cancer through the Akt1-related CXCL12/CXCR4 Pathway, Comb. Chem. High Throughput Screening, 2024, 27(6), 863–876.
73. J. Zhang, W. Fan, H. Wang, L. Bao, G. Li and T. Li, et al., Resveratrol Protects PC12 Cell against 6-OHDA Damage via CXCR4 Signaling Pathway, J. Evidence-Based Complementary Altern. Med., 2015, 2015, 730121.
74. M. Wang, R. Hu, Y. Wang, L. Liu, H. You and J. Zhang, et al., Atractylenolide III Attenuates Muscle Wasting in Chronic Kidney Disease via the Oxidative Stress-Mediated PI3K/AKT/mTOR Pathway, Oxid. Med. Cell. Longevity, 2019, 2019, 1875471.
75. Y. H. Oum, S. A. Kell, Y. Yoon, Z. Liang, P. Burger and H. Shim, Discovery of novel aminopiperidinyl amide CXCR4 modulators through virtual screening and rational drug design, Eur. J. Med. Chem., 2020, 201, 112479.
76. H. A. S. Murad, T. M. A. Alqurashi and M. A. Hussien, Interactions of selected cardiovascular active natural compounds with CXCR4 and CXCR7 receptors: a molecular docking, molecular dynamics, and pharmacokinetic/toxicity prediction study, BMC Complementary Med. Ther., 2022, 22(1), 35.

---

Footnotes

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d4fo02493a

‡ The first two authors equally contribute to this paper.

---