Abstract
Licorice, a traditional Chinese medicine, has been widely used for the treatment of COVID-19, but all active compounds and corresponding targets are still not clear. Therefore, this study proposed a deep learning-based network pharmacology approach to identify more potential active compounds and targets of licorice. 4 compounds (quercetin, naringenin, liquiritigenin, and licoisoflavanone), 2 targets (SYK and JAK2) and the relevant pathways (P53, cAMP, and NF-kB) were predicted, which were confirmed by previous studies to be associated with SARS-CoV-2-infection. In addition, 2 new active compounds (glabrone and vestitol) and 2 new targets (PTEN and MAP3K8) were further validated by molecular docking and molecular dynamics simulations (simultaneous molecular dynamics), as well as the results showed that these active compounds bound well to COVID-19 related targets, including the main protease (Mpro), the spike protein (S-protein) and the angiotensin-converting enzyme 2 (ACE2). Overall, in this study, glabrone and vestitol from licorice were found to inhibit viral replication by inhibiting the activation of Mpro, S-protein and ACE2; related compounds in licorice may reduce the inflammatory response and inhibit apoptosis by acting on PTEN and MAP3K8. Therefore, licorice has been proposed as an effective candidate for the treatment of COVID-19 through PTEN, MAP3K8, Mpro, S-protein and ACE2.
Similar content being viewed by others
Introduction
Coronavirus disease 2019 (COVID-19) has had a significant impact on global health systems and economic development due to its highly infectious nature and complex pathogenesis1. Existing therapies, including conventional treatments (e.g., oxygen therapy) and immunomodulators, can only play a preventive role, and the rapid development of specific drugs and vaccines targeting COVID-19 has become the greatest challenge2. Studies have shown that traditional Chinese medicine (TCM) can improve clinical symptoms, delay disease progression, as well as reduce mortality and recurrence rates in patients with COVID-193,4. Of the available formulas, preventive prescriptions, and therapeutic prescriptions for confirmed cases proposed by TCM, licorice is one of the most frequently used for the treatment of COVID-195,6.
Licorice is a perennial herb commonly used in TCM7, leguminous or plant rhizomes and dried roots are frequently used in medicinal preparations8. Many licorice compounds and corresponding targets have been shown to play a central role in the treatment of COVID-19 through network pharmacology analysis, as well as in in vivo or in vitro studies9,10,11. In terms of compounds, for example, Glycyrrhizic Acid, and other compounds found in licorice can bind to Mpro, ACE2 and S proteins, respectively, which could inhibit COVID-19 replication and block virus binding sites12,13,14; in terms of targets, MAPKs, ILs and NF-kB can regulate the MAPK signaling pathway, the IL-17 signaling pathway and the NF-kB signaling pathway, exerting anti-inflammatory and immunomodulatory effects13.
Traditional network pharmacology, namely P1 (Fig. 1a) in this study, an research method is often used to identify active compounds and targets15, has the advantages of being comprehensive, systematic, and holistic, which is consistent with the multi-compound, multi-target, and multi-pathway characteristics of TCM, expanding the potential applications of TCM research16. However, it still presents some challenges, such as the lack of comprehensive data on various drugs, genes, and proteins17. Therefore, we proposed a deep learning (DL)-based network pharmacology method, in which a model integrated with a drug-target interaction (DTI) method was adopted [Highlight the innovative nature of the method].
(a) Flow chart of traditional network pharmacology process (P1); (b) Improved flow chart for two processes: (i) process P2 and (ii) process P3.
Based on above new method, two new processes were proposed, namely P2 (Fig. 1bi) and P3 (Fig. 1bii). (1) For process P2, since the traditional network pharmacology approach does not consider characteristic information about the molecular structure of proposed drugs, integrating GO and KEGG enrichment analysis in process P1 with the DTI prediction, it is possible to predict more compounds and targets. (2) Compared to process P2, which identifies some important compounds and targets after PPI analysis, process P3 takes into account all the active compounds of licorice and COVID-19 targets and performs a prediction of DTI, making full use of all available information.
In fact DTI is one of the most direct and effective methods for discovering active compounds and targets. Several new DTI methods have been developed in recent years, all of which have shown promising results, for example, DeepDTA, a model only extracts the sequence information of targets and compounds by the convolutional neural network (CNN)18. And DeepPurpose, which combined the current optimal models, achieved better results compared to the previous model19. However, these models only considered a single feature of compound information. Therefore, we proposed a new model called DeepDrugTargetInteractionandGraphConvolutional.
(DeepDGC) for DTI prediction, which included two DL algorithms, graph convolutional neural network (GCN) and CNN, to extract more characteristic information from the molecular structure of the compound. GCN and CNN were used to obtain two representations of the compound—the molecular map and the Morgan fingerprint, respectively. In addition, CNN was used to learn the amino acid sequence of the disease targets. After two vectors were generated in the above two steps, they were inserted into a fully connected layer, followed by a regression layer, in which the output was the compound-target affinity value.
According to the above description, this study included a total of three processes, i.e., P1 (Fig. 1a), P2 (Fig. 1bi), and P3 (Fig. 1bii). Here, P1 was the traditional network pharmacology process; P2 and P3 were the improved processes proposed in this study.
Materials and methods
Acquisition of active compounds and licorice targets
We retrieved 249 compounds by searching the TCMSP20 databases with the keyword "licorice". After screening the criteria of oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥ 0.18, 92 active compounds remained. The corresponding targets of the active compounds were obtained from the TCMSP20, SwissTargetPrediction21, PharmMapper22 and GeneCards23 databases, and 1140 targets were named after deleting duplicate items.
Acquisition of disease targets
Searching the GeneCards23, OMIM24, DrugBank25 and other databases with the keywords "COVID-19" produced 13,542 targets for COVID-19 after deleting duplicate values.
Acquisition of overlapping targets
A total of 774 overlapping targets remained after intersection screening using Venny26 from 1140 licorice targets and 13,542 targets for COVID-19, which were considered potential targets of licorice acting in COVID-19.
Acquisition of key compounds and key targets by analysis of PPI networks
The overlapping targets were loaded into the STRING database27 to obtain the PPI network. After the removal of isolated targets, the PPI network was imported into Cytoscape software28, where the Centiscape plug-in was used to screen for the key targets, with three parameters (Degree unDir, Betweenness unDir and Closeness unDir) used as thresholds. The key compounds were then acquired according to the key targets.
Acquisition of core compounds and core targets by GO and KEGG enrichment analysis
Based on the key targets, the Metascape database29 was used to conduct GO and KEGG enrichment analysis (p < 0.01), and a licorice-compound-target-pathway network was built to screen core targets by the Cytoscape software. All compounds corresponding to the core targets on the network were defined as core compounds.
Predictions of compounds and targets based on DeepDGC
The DeepDGC model (Fig. 2) was used in both processes P2 and P3, the input data types being SMILES sequences for compounds and protein amino acid sequences for targets. After compounds were transformed into SMILES strings and targets were converted into protein amino acid sequences, each SMILES string was matched to each amino acid sequence one by one. The output of this model was affinity values that indicated the interaction probabilities.
Detailed structure of the DeepDGC model.
CNN was an architecture containing one or more convolutional layers, usually followed by a pooling layer, and GCN, an optimization of CNN, was a graph neural network using convolutional operations and compensated for the inability of CNN to handle non-Euclidean structured data. Therefore, the model combined GCN and CNN. It comprised two separate CNN blocks and a GCN block. For CNN block, we used DeepDTA’s18 configurations to set up the CNN block for the DeepDGC model, and for GCN block, we used three graph convolutional layers. In each layer, the covalent bonds and node information were extracted according to the molecular map [Highlight the innovative nature of DL model].
The KIBA dataset (pretraining dataset) for DeepDGC comprised primarily of the SMILES strings of 2111 active compounds, the amino acid sequences of 229 targets, and 118,254 binding affinity values. And it was employed to train the model using five-fold-cross-validation. Furthermore, to ensure the generalization ability, we set the activation functions relu and dropout. Finally, CI and MSE were used as evaluation indicators and the average results were reported. And with regard to the parameter settings, some of them were based on relevant studies, while other important parameters were compared experimentally for optimum results. The relevant parameters were set as shown in Table 1.
Subsequent to the evaluation of the pre-training dataset, the DeepDGC model was used to predict affinity values in process P2 and P3. The prediction dataset used in process P2 was composed primarily of the SMILES strings of the key compounds of licorice (obtained from the PPI analysis), and amino acid sequences of the key targets (obtained from the PPI analysis); while the prediction dataset used in process P3 was composed primarily of the SMILES strings of active compounds of licorice (screened by OB ≥ 30% and DL ≥ 0.18), and amino acid sequences of targets (for COVID-19).
Selection of compliant core compounds by ADMET analysis
ADMET prediction is the assessment of five aspects (absorption, distribution, metabolism, excretion, and toxicity), which plays a key role in drug development30. In this study, the physicochemical and pharmacokinetic properties of the active components were predicted through the SwissADME database31 and the pkCSM database32, respectively.
The physicochemical properties considered here include molecular weight (MW), rotatable bond count (RB), H-bond acceptors (HBA), donor count, TPSA, and leadlikeness violations (LSV). The pharmacokinetic properties included absorption (i.e. Caco-2 cell permeability, HIA and skin permeability), distribution (i.e. VDss, unbound fraction, the blood–brain barrier and central nervous system permeability), excretion (i.e. total clearance and renal OCT2 substrate), and toxicity (i.e. AMES toxicity, maximum tolerated dose, hERG I inhibitor, hERG II inhibitor, oral rat acute toxicity (LD50), hepatotoxicity, skin sensitization, and minnow toxicity).
Validation of core compounds and core targets by molecular docking
Molecular docking was used to further verify the binding capabilities of the core compounds and related targets outlined above. First, the monomeric component structures of the protein targets and related information were obtained from the UniProb33 and PDB34 databases. Second, AutoDockTools was used to conduct a range of operations, such as hydrogenation, charge addition, removal of water molecules, and removal of metal ions. Third, the PubChem database35 was used to construct the 3D structures of the active compounds. Subsequently, global docking boxes were generated by AutoDockTools while blind docking was performed using qvina-w. The binding score was used to evaluate the ability of a natural compound to bind to the target. Finally, heat maps and 3D docking maps of the docking results were created using Python and Pymol.
Molecular dynamics simulations
The compound-protein target pair with the highest binding energy of molecular docking was subjected to molecular dynamics (MD) simulations to further check the binding stability of the two. MD simulations were then performed using Gromacs software36. To ensure the total charge neutrality of the simulated systems, corresponding amounts of sodium ions were added to the three systems to displace water molecules and produce solvent boxes of appropriate size. Next, periodic boundary conditions (PBC) were applied in each of the three directions of the system, thereby determining the force field parameters for the entire atom. Finally, two complexes were simulated for a 100 ns NPT ensemble (with constant number of particles, pressure and temperature).
Results
Key compounds and key targets
To explore the mechanism underlying the therapeutic effects of licorice against COVID-19, 774 targets were imported into the STRING database to construct a PPI network. After screening according to three thresholds, we obtained 88 key compounds (Table S1) and 156 key targets (Table S2).
Core compounds and core targets
To determine the molecular mechanisms underlying the efficacy of licorice treatment against COVID-19, we used Metascape to perform the biofunctional annotation of GO and KEGG pathway enrichment analysis of the key targets. The GO biofunctional annotation results showed that a variety of terms were identified, including 254 biological process (BP) terms that mainly relate to the positive regulation of protein phosphorylation and the positive regulation of cell migration, 141 cellular component (CC) terms that mainly relate to the lumen of the vesicle, membrane rafts, and receptor complexes, 59 molecular function (MF) terms that were mainly related to kinase binding, protein serine/threonine/tyrosine kinase activity and transcription factor binding. The top 10 considerably enriched terms for BP, CC, and MF were visualized in Fig. 3a.
Enrichment analysis of GO and KEGG. (a) Top 10 biological processes, top 10 cellular compounds, and top 10 molecular functions: (b) top 20 KEGG pathways. The color scale indicates the different thresholds of p values and the size of the dots represents the number of genes corresponding to each pathway.
The KEGG pathway enrichment analysis results showed that the key targets were enriched in 180 pathways, and the top 20 paths with the highest level of enrichment were chosen for visualization in Fig. 3b, which revealed enrichment mainly in the P53 signaling pathway, cAMP signaling pathway, NF-kB, and other related signaling pathways.
Furthermore, to better understand the mechanism by which licorice acts on COVID-19, a licorice compound-target-pathway relationship network was built based on the above KEGG pathway (Fig. S1), which suggests the potential interactions between active compounds and targets, as well as the related pathways of licorice for the treatment of COVID-19. Finally, the top 5 active compounds (quercetin, glypallichalcone, calycosin, vestitol, naringenin) were selected as the core compounds and the top 5 genes (PTGS2, HSP90AB1, PPARG, SYK, ALB) were selected as the core targets. These results derived from process P1.
Complementary core compounds and core targets
To explore more core compounds and core targets, a DeepDGC model was presented in this study. After the model was trained on the pre-training dataset, it was used to predict the affinity values between compounds and targets. We used four models (KronRLS37, SimBoost38, DeepDTA15, WideDTA39, DeepPurpose19 as baseline models and the comparison results were shown in Table 2.
The prediction data set used in the process P2 was primarily composed of 88 SMILES strings of active compounds and 1248 amino acid sequences of targets. The prediction data set used in the P3 process was composed primarily of 92 SMILES strings of active compounds and 58,378 amino acid sequences of targets. The final prediction affinity values of the top 30 for P2 and P3 were shown in Tables 3 and 4.
The combined results of the three processes (P1, P2 and P3), including all core compounds and core targets, were shown in Table 5. Compared with existing studies of formulations or formulae containing licorice, the relevance of the 4 compounds (quercetin40, naringenin40, liquiritigenin41 and licoisoflavanone42) and 6 targets (SYK41, PTGS243, PPARG43, ALB43, HSP90AB144 and JAK245) had been explored previously. Therefore, the other 15 core compounds (glypallichalcone, calycosin and vestitol) and 2 core targets (PTEN and MAP3K8) will be discussed below.
Non-toxic and easily absorbed core compounds
Based on the above results, ADMET was used to predict the physicochemical and pharmacokinetic properties according to SwissADME and pkCSM. SwissADME calculations showed that 5 compounds passed the stringent lead-like criteria (250 g/mol ≤ MW ≤ 350 g/mol, XLOGP ≤ 3.5 and rotatable bonds ≤ 746), indicating they could be considered excellent drug candidates against COVID-19 (Table 6). These lead-like compounds were further predicted by pkCSM. Regarding absorption parameters and drug distribution parameters, all 5 compounds were within the acceptable range. However, isotrifoliol, glypallichalcone and calycosin did not satisfy the criteria of hERG II inhibitor. Finally, only the new active compounds glabrone (MOL004912) and vestitol (MOL000500) could be considered eligible core compounds, of which vestitol was identified by P1, and glabrone was identified in P2 and P3 (Table 7).
Stable combination of core compounds and core targets
To support our findings mentioned above, we used molecular docking to evaluate the interaction between the active core compounds and the core targets, in which binding affinity less than − 7.0 kcal/mol indicated a good interaction47. Two new active compounds in licorice (glabrone and vestitol) were docked to COVID-19 binding sites, such as Mpro, S-protein and ACE2. The binding affinities were shown in Table 8, and the binding modes of the selected active compounds and targets with the highest binding values was shown in Fig. 4. According to Tables 3 and 4, two new targets in COVID-19 (PTEN and MAP3K8) were docked to licorice-related compounds, such as glabrone, licoisoflavone B and isotrifoliol. The binding affinity results were shown in Table S3.
Molecular docking analysis of the selected compounds and targets with the largest binding values. (a) The docking mode of glabrone and S-protein. (b) The docking mode of glabrone and Mpro. (c) The docking mode of vestitol and ACE2.
Molecular dynamics simulations
The larger the protein Calpha root-mean-square deviation (RMSD) of the MD simulations, the more violent the fluctuations indicating greater motility and less stability.The RMSD data of licorice, include glabrone and S-protein, as well as licoisoflavone B and MAP3K8, were shown in Fig. 5. The results showed that the RMSD fluctuations for S-protein/glabrone and MAP3K8/licoisoflavone B are within 2 Å, which means that the system is less kinetic and more stable. These findings all showed that a stable conformation has been achieved in the process of MD simulations [MD simulations analysis].
Differences in the root mean square deviation (RMSD) of the plural over time (blue polygonal line means glabrone and S-protein, red polygonal line means licoisoflavone B and MAP3K8).
Discussion
The study design was divided into three processes, including the traditional network pharmacology process (P1) and the two improved processes (P2 and P3). Using this approach 20 core compounds and 6 core targets were predicted, among which 4 active compounds (quercetin, naringenin, liquiritigenin and licoisoflavanone) and 2 targets (SYK and JAK2), were associated with COVID-19 infection, as confirmed by previous studies. Furthermore, after ADMET and molecular docking analysis, two new active compounds were identified: vestitol was identified in P1 and glabrone was jointly identified in processes P2 and P3. and 2 new targets were also predicted, among which PTEN was identified in P2, and MAP3K8 were identified in P3. In addition, this study also revealed the signaling pathways of P53, cAMP, and NF-kB.
Considering the 4 active compounds (quercetin, naringenin, liquiritigenin and licoisoflavanone), their potential roles in the treatment of COVID-19 have been demonstrated in previous studies. Quercetin has anti-inflammatory activity exerted by inhibiting the secretion of pro-inflammatory factors (such as IL-6, IL-1β and TNF-α), by an antiviral effect by blocking the entry of coronavirus into host cells, as well as by an anticoagulant activity exerted by inhibiting plasma protein disulfide isomerase48. Naringenin also induces anti-inflammatory and antiviral activity49. As a flavonoid with anticancer, antioxidant, hepatoprotective, immune regulatory, and antiplatelet aggregation properties, liquiritigenin has served as a therapy for COVID-19. For example, liquiritigenin was used to form a complex with the Mpro of SARS-CoV2 because it was found to inhibit the catalytic activity of the main protease50. Licoisoflavanone is an isoflavonoid compound that plays a role in the reduction of antiviral, cytokine storms, prevention of ARDS and multi-organ damage, and reduction of the severity of inflammatory diseases51.
Previous studies have also indicated that SYK and JAK2 may serve as target proteins related to COVID-19. SYK also plays an important role in the treatment of COVID-19 and has been reported to regulate signal transduction pathways implicated in these complications associated with COVID-1952. JAK2 involves in M2 macrophage polarisation, inflammatory response, pulmonary fibrosis and thrombosis by activating STAT3, a signal transduction and transcriptional activator53. An increasing number of studies have also highlighted that JAK2 is an important gene belonging to the JAK2/STAT3 signaling pathway, and can induce overexpression of IL-6 and IL-18, which can exacerbate the inflammatory response and lung injury46. Furthermore, the SYK inhibitor (fostamatinib)54 and the JAK2 inhibitor (fedratinib)55, evaluated in clinical studies, are highly effective in the prevention and treatment of COVID-19.
The examples given above demonstrate the important roles of the 4 active compounds and 2 targets in COVID-19, identified in our analysis, and confirmed by previous studies, suggesting the usefulness of our model in the prediction of active compounds and targets. In fact, we also identified 2 new compounds (vestitol and glabrone) and 2 new targets (PTEN and MAP3K8) that have not yet been demonstrated to be directly associated with the treatment of COVID-19, but have been attested to have a role in the treatment of COVID-19-related diseases (e.g. hepatitis B, influenza A virus).
Many related studies had shown that the newly identified active compounds vestitol and glabrone play a key role in the treatment of COVID-19. Because they not only inhibit viral replication through stable binding to the three viral binding sites of S-protein, ACE2, and Mpro, but also exert antioxidant, anti-inflammatory, and antiviral effects. For example, vestitol can achieve an anti-inflammatory effect by inhibiting the NF-kB signaling pathway and has shown to be a considerable promising new anti-inflammatory agent56. Glabrone can achieve an antioxidant activity by modulating the nuclear factor erythroid 2-related factor 2 (Nrf2) pathway and an anti-inflammatory effect by regulating the NF-kB signaling pathway57. In addition, the antiviral activity of glabrone was demonstrated by cytopathic effect (CPE) inhibition assays targeting the influenza A virus58.
Two new targets identified using our approach, including PTEN and MAP3K8, were potentially key targets for the treatment of COVID-19. Studies had shown that PTEN can activate dendritic cells, B cells and T cells, which are innate immune cells, and secrete pro-inflammatory factors, including interferon (IFN), TNF-α and IL10, thus inducing the formation of the cytokine storm in patients with COVID-19. Therefore, targeting PTEN can inhibit the formation of cytokines storms59. MAP3K8 participates in the pulmonary fibrotic response and the lung inflammatory response. An increasing number of studies had also highlighted the significance of MAP3K8 in suppressing lung inflammation and fibrosis (the main symptom of COVID-19)60.
In terms of the signaling pathway, the P53, cAMP, and NF-kB signaling pathways, discovered by using KEGG pathway analysis, are involved in inflammation, immunomodulation and infection. The P53 signaling pathway is a pathway known to influence immune responses61. Furthermore, p53, an intrinsic host restriction factor of SARS-CoV-2, can reduce virus production62. The cAMP signaling pathway is the most important signaling pathway in EG pathway enrichment, and EG could also act on the PI3K-Akt, JAK-STAT and chemokine signaling pathways, thus reducing responses such as inflammation and apoptosis40. In turn, the NF-kB signaling pathway, considered as an inflammation center63, induces various target genes in inflammatory diseases64,65, as well as regulates cytokine storm syndromes and immunosuppression66,67.
S-protein and ACE2 were key protein targets in the first process of infection (attachment and cell entry)68,69, and Mpro was a key target in the second process of infection (replication and transcription)70. In this paper, we preliminarily concluded that vestitol and glabrone had good binding stability with S-protein, ACE2 and Mpro by molecular docking. Analysis of the S protein and glabrone was then further performed by MD simulations to demonstrate that glabrone may inhibit host cell infection at the first stage of attachment and entry. In addition, we also conducted molecular docking of two new targets (PTEN and MAP3K8) with the compounds of licorice (including vestitol and glabrone), and the results showed good binding stability. We followed by MD simulations of MAP3K8 and licoisoflavone B with the highest binding energy, which demonstrated that licorice may be considered an effective candidate for the treatment of COVID-19 through MAP3K8 [MD simulations analysis].
Based on the above analysis, we knowed that the DL-based network pharmacology method could compensate to some extent for the impossibility of obtaining all compounds and targets in traditional network pharmacology through databases and analysis software, and achieve a promising predictive results. However, due to the predictive performance of DL was very dependent on the quantity and quality of the data, and it also had limitations in feature information extraction. Therefore, there was still room for improvement in future work in terms of the quantity and quality of data and feature extraction [Highlight the innovation of the method and the shortcomings of the new method].
Conclusion
In this study, we proposed a DL-based DeepDGC model that learned from both molecular maps and Morgan fingerprint data representations of drugs, which contains more feature information for drug characterization that can be evaluated and optimize the model. As a result, 2 new compounds and 2 new targets were also found to possess potential effects on COVID-19 treatment. Although our findings are not sufficient to reach more definite conclusions and further validation, using in vivo or in vitro studies, is still encouraged, we believe that this method has a certain translational value that can also be applied to drug and target discovery studies in other diseases.
Data availability
The datasets for this study can be found in the https://github.com/2022-fuyu/COVID-19.
References
Chilamakuri, R. & Agarwal, S. COVID-19: Characteristics and therapeutics. Cells 10(2), 206 (2021).
Aimrane, A. et al. Insight into COVID-19’s epidemiology, pathology, and treatment. Heliyon 8(1), e08799 (2022).
Xiong, X. J. et al. Chinese herbal medicine for coronavirus disease 2019: A systematic review and meta-analysis. Pharmacol. Res. 160, 105056 (2020).
Lu, L. et al. Analysis on herbal medicines utilized for treatment of COVID-19. Acta Pharm. Sin. B 7, 1192–1204 (2020).
Zhou, Z., Zhu, C. S. & Zhang, B. Study on medication regularity of traditional Chinese medicine in treatment of COVID-19 based on data mining. Zhongguo Zhong Yao Za Zhi 45(6), 1248–1252 (2020).
Wang, C. et al. Analysis of medication regularity and pharmacodynamic characteristics of traditional Chinese medicine treatment in 444 severe cases of COVID-19. Zhongguo Zhong Yao Za Zhi 45(13), 3007–3012 (2020).
Shah, S. L. et al. Inhibitory effects of Glycyrrhiza glabra and its major constituent Glycyrrhizin on inflammation-associated corneal neovascularization. Evidence-based Complement Altern Med 45, 1–8 (2018).
Hosseinzadeh, H. & Nassiri-Asl, M. Pharmacological effects of Glycyrrhiza spp. and its bioactive constituents: Update and review. Phytother. Res. 29(12), 1868–1886 (2015).
Zhang, Q. H. et al. Traditional uses, pharmacological effects, and molecular mechanisms of licorice in potential therapy of COVID-19. Front. Pharmacol. 12, 719758 (2021).
Wu, C. Y. et al. Small molecules targeting severe acute respiratory syndrome human coronavirus. Proc Natl Acad Sci USA 101(27), 10012–10017 (2004).
Diomede, L. et al. Can antiviral activity of licorice help fight COVID-19 infection? Biomolecules 11(6), 855 (2021).
Huang, Y. F., Bai, C., He, F., Xie, Y. & Zhou, H. Review on the potential action mechanisms of Chinese medicines in treating Coronavirus Disease 2019 (COVID-19). Pharmacol. Res. 158, 104939 (2020).
Ng, S. L. et al. Licorice: A potential herb in overcoming SARS-CoV-2 infections. J. Evidence-Based Integr. Med. 26, 2515690X21996662 (2021).
Abraham, J. & Florentine, S. Licorice (Glycyrrhiza glabra) extracts-suitable pharmacological interventions for COVID-19? A review. Plants (Basel). 10(12), 2600 (2021).
Li, X. et al. Chemical composition and pharmacological mechanism of ephedraglycyrrhiza drug pair against coronavirus disease 2019 (COVID-19). Aging 11(6), 683–690 (2021).
Luo, T. T. et al. Network pharmacology in research of Chinese medicine formula: Methodology, application and prospective. Chin. J. Integr. Med. 26(1), 72–80 (2020).
Kim, J., Park, S., Min, D. & Kim, W. Comprehensive survey of recent drug discovery using deep learning. Int. J. Mol. Sci. 22(18), 9983 (2021).
Öztürk, H., Özgür, A. & Ozkirimli, E. DeepDTA: Deep drug-target binding affinity prediction. Bioinformatics 34(17), i821–i829 (2018).
Huang, K. et al. DeepPurpose: A deep learning library for drug-target interaction prediction. Bioinformatics 36(22–23), 5545–5547 (2020).
Ru, J. et al. TCMSP: A database of systems pharmacology for drug discovery from herbal medicines. Chemiinformatics 6(13), 1–6 (2014).
Daina, A. et al. SwissTargetPrediction: Updated data and new features for efficient prediction of protein targets of small molecules. Nucl. Acids Res. 47, W357–W364 (2019).
Wang, X. et al. PharmMapper 2017 update: A web server for potential drug target identification with a comprehensive target pharmacophore database (Article). Nucl. Acids Res. 45, W356–W360 (2017).
Safran, M. et al. GeneCards Version 3: The human gene integrator. Database 2010, baq020 (2010).
Amberger, J. S. et al. OMIM.org: Leveraging knowledge across phenotype-gene relationships. Nucl. Acids Res. 47, D1038–D1043 (2019).
Wishart, D. S. et al. DrugBank 5.0: A major update to the DrugBank database for 2018. Nucl. Acids Res. 46, D1074–D1082 (2018).
Oliveros J. C. Venny. An Interactive Tool for Comparing Lists with Venn's Diagrams, 34 (2007).
Szklarczyk, D. et al. STRING v11: Protein–protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucl. Acids Res. 47, D607–D613 (2019).
Franz, M. et al. Cytoscape js: A graph theory library for visualisation and analysis. Bioinformatics 32(2), 309–311 (2016).
Zhou, Y. et al. Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nat. Commun. 10, 1523 (2019).
Ortiz, C. L. D. et al. Nellas. Potential inhibitors of galactofuranosyltransferase 2 (GlfT2): Molecular docking, 3D-QSAR, and in silico ADMETox studies. Science 9(1), 1–28 (2019).
Daina, A., Michielin, O. & Zoete, V. SwissADME: A free web tool to evaluate the pharmacokinetics, drug-likeness and medicinal chemistry friendliness. Sci. Rep. 7, 42717 (2017).
Pires, D. E., Blundell, T. L. & Ascher, D. B. pkCSM: Predicting small-molecule pharmacokinetic and toxicity properties using graph-based signatures. J. Med. Chem. 58, 4066–4072 (2015).
Newburger, D. E. & Bulyk, M. L. UniPROBE: an online database of protein binding microarray data on protein-DNA interactions. Nucl. Acids Res. 37, D77-82 (2009).
wwPDB consortium. Protein Data Bank: the single global archive for 3D macromolecular structure data. Nucl. Acids Res. 47(1), 520–528 (2019).
Kim, S. et al. PubChem in 2021: New data content and improved web interfaces. Nucl. Acids Res. 49, D1388–D1395 (2021).
Cao, J. F. et al. Exploring the mechanism of action of licorice in the treatment of COVID-19 through bioinformatics analysis and molecular dynamics simulation. Front Pharmacol. 13, 1003310 (2022).
Pahikkala, T. et al. Toward more realistic drug-target interaction predictions. Brief. Bioinform. 16, 325–327 (2014).
He, T. et al. Simboost: A read-across approach for predicting drug–target binding affinities using gradient boosting machines. J. Cheminform. 9(24), 1–14 (2017).
Öztürk, H., Ozkirimli, E., et al. WideDTA: Prediction of drug-target binding affinity. ArXiv abs/1902.04166, (2019).
Qin, L. et al. Glycyrrhiza uralenesis on novel coronavirus infection complicated with myocarditis based on network pharmacology. Adv. Cardiovasc. Dis. 41(11), D1215–D1221 (2020).
Li, X. et al. Chemical composition and pharmacological mechanism of ephedra-glycyrrhiza drug pair against coronavirus disease 2019 (COVID-19). Aging 13(4), D4811–D4830 (2021).
Xu, J. H. et al. Study on mechanism of Shufeng Jiedu Capsule in treating COVID-19 based on network pharmacology. Chin. Tradit. Herb. Drugs 51, D2015–D2023 (2020).
Wanqing, Z. et al. Study on the molecular mechanism of Shufeng Jiedu capsules in treatment for COVID-19 based on network pharmacology. Modernization Tradit. Chin. Med. Mater. Med.-World Sci. Technol. 22(10), D3515–D3525 (2020).
Ma, D. et al. Single-cell RNA sequencing identify SDCBP in ACE2-positive bronchial epithelial cells negatively correlates with COVID-19 severity. J. Cell Mol. Med. 25(14), D7001–D7012 (2021).
Zhou, H. et al. Network pharmacology and bioinformatics analysis identified essential genes of Jingulian in the treatment of rheumatoid arthritis and COVID-19. Ann. Transl. Med. 10(11), 635 (2022).
Teague, S. J., Davis, A. M., Leeson, P. D. & Oprea, T. The design of leadlike combinatorial libraries. Angew. Chem. Int. Edit. 38(24), 3743–3748 (1999).
Hsin, K. Y., Ghosh, S. & Kitano, H. Combining machine learning systems and multiple docking simulation packages to improve docking prediction reliability for network pharmacology. PLoS ONE 8(12), e83922 (2013).
Manjunath, S. H. & Thimmulappa, R. K. Antiviral, immunomodulatory, and anticoagulant effects of quercetin and its derivatives: Potential role in prevention and management of COVID-19. J. Pharm. Anal. 12(1), 29–34 (2022).
Alberca, R. W. et al. Perspective: The potential effects of naringenin in COVID-19. Front. Immunol. 11, 570919 (2020).
Srivastava, V. & Yadav, A. Molecular docking and ADMET study of bioactive compounds of Glycyrrhiza glabra against main protease of SARS-CoV2. Mater. Today Proc. 49, 2999–3007 (2020).
Liu, C. et al. Screening of active ingredients of the Tibetan medicine Wu Wei Sha ji San for the treatment of novel coronavirus pneumonia (COVID-19) based on network pharmacology and molecular docking method. World Sci. Technol. Modernization Chin. Med. 22(3), 632–641 (2020).
Dummer, W., Markovtsov, V. V., Tong, S. & Masuda, E. Clinical trial to evaluate an approved ITP therapy targeting spleen tyrosine kinase (SYK) for prevention and treatment of COVID-19. Blood 136, 35 (2020).
Jafarzadeh, A., Nemati, M. & Jafarzadeh, S. Contribution of STAT3 to the pathogenesis of COVID-19. Microb. Pathog. 154, 104836 (2021).
Li, M. L. et al. The role of JAK2/STAT3 signaling pathway in lung injury in rats with severe acute pancreatitis. PLA Med. J. 6, P611-613 (2011).
Wu, D. & Yang, X. O. TH17 responses in cytokine storm of COVID-19: An emerging target of JAK2 inhibitor Fedratinib. J. Microbiol. Immunol. Infect. 53(3), 368–370 (2020).
Bueno-Silva, B., Rosalen, P. L., Alencar, S. M. & Mayer, M. P. A. Vestitol drives LPS-activated macrophages into M2 phenotype through modulation of NF-κB pathway. Int. Immunopharmacol. 82, 106329 (2020).
Liu, D. Y. et al. Anti-inflammatory and antioxidant effects of licorice flavonoids on ulcerative colitis in mouse model. Chin. Herb. Med. 9(4), 358–368 (2017).
Grienke, U. et al. Computer-guided approach to access the anti-influenza activity of licorice constituents. J. Nat. Prod. 77(3), 563–570 (2014).
Besset, V., Scott, R. P. & Ibanez, C. F. Signaling complexes and protein- protein interactions involved in the activation of the Ras and phosphatidylinositol 3-kinase pathways by the c-Ret receptor tyrosine kinase. J. Biol. Chem. 275(50), 39159–39166 (2000).
Emameh, R. Z., Nosrati, H., Eftekhari, M., Falak, R. & Khoshmirsafa, M. Expansion of single cell transcriptomics data of SARS-CoV infection in human bronchial epithelial cells to COVID-19. Biol. Proced. Online 22(1), 1–10 (2020).
Rahaman, M. et al. Decrypting the role of predicted SARS-CoV-2 miRNAs in COVID-19 pathogenesis: A bioinformatics approach. Comput. Biol. Med. 136, 104669 (2021).
Kumar, A., Grams, T. R., Bloom, D. C. & Toth, Z. Signaling pathway reporter screen with SARS-CoV-2 proteins identifies nsp5 as a repressor of p53 activity. Viruses 14(5), 1039 (2022).
Lawrence, T. The nuclear factor NF-kappaB pathway in inflammation. Cold Spring Harb. Perspect. Biol. 1(6), a001651 (2009).
Taniguchi, K. & Karin, M. NF-kappaB, inflammation, immunity and cancer: coming of age. Nat. Rev. Immunol. 18(5), 309–324 (2018).
Yuan, G. et al. Clock mutant promotes osteoarthritis by inhibiting the acetylation of NFkappaB. Osteoarthr. Cartil. 27(6), 922–931 (2019).
Ye, Q., Wang, B. & Mao, J. The pathogenesis and treatment of the `Cytokine Storm’ in COVID19. J. Infect. 80(6), 607–613 (2020).
Catanzaro, M. et al. Immune response in COVID19: addressing a pharmacological challenge by targeting pathways triggered by SARS-CoV-2. Signal Transduct. Target Ther. 5(1), 84 (2020).
Wrapp, D. et al. Cryo-EM structure of the 2019-nCoV spike in the prefusion conformation. Science 367, 1260–1263 (2020).
Xu, X. et al. Evolution of the novel coronavirus from the ongoing Wuhan outbreak and modeling of its spike protein for risk of human transmission. Sci. China Life Sci. 63, 457–460 (2020).
Chen, S. et al. Feline infectious peritonitis virus Nsp5 inhibits type I interferon production by cleaving NEMO at multiple sites. Viruses 12, 43 (2019).
Acknowledgements
We thank Charlesworth for its linguistic assistance during the preparation of this manuscript.
Funding
This work was supported by the National Natural Science Foundation of China (No. 61873080) and the Natural Science Foundation of Zhejiang Province of China (LY16F020030).
Author information
Authors and Affiliations
Contributions
J.H., P.W. and L.S. conceived the idea. Y.F. used network pharmacology and deep learning methods to predict the core components and targets and drafted the manuscript. S.G. and T.X. performed the code debugging. Y.Y.F. performed the molecular docking analysis. J.H. and Y.F. reviewed the draft and submitted the manuscript for publication. All authors approved the final version.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Fu, Y., Fang, Y., Gong, S. et al. Deep learning-based network pharmacology for exploring the mechanism of licorice for the treatment of COVID-19. Sci Rep 13, 5844 (2023). https://doi.org/10.1038/s41598-023-31380-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-023-31380-7
This article is cited by
-
Network pharmacology and in vitro experimental verification unveil glycyrrhizin from glycyrrhiza glabra alleviates acute pancreatitis via modulation of MAPK and STAT3 signaling pathways
BMC Complementary Medicine and Therapies (2024)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
