key: cord-0958544-qebbkr6d authors: Ge, Yiyue; Tian, Tingzhong; Huang, Suling; Wan, Fangping; Li, Jingxin; Li, Shuya; Yang, Hui; Hong, Lixiang; Wu, Nian; Yuan, Enming; Cheng, Lili; Lei, Yipin; Shu, Hantao; Feng, Xiaolong; Jiang, Ziyuan; Chi, Ying; Guo, Xiling; Cui, Lunbiao; Xiao, Liang; Li, Zeng; Yang, Chunhao; Miao, Zehong; Tang, Haidong; Chen, Ligong; Zeng, Hainian; Zhao, Dan; Zhu, Fengcai; Shen, Xiaokun; Zeng, Jianyang title: A data-driven drug repositioning framework discovered a potential therapeutic agent targeting COVID-19 date: 2020-03-12 journal: bioRxiv DOI: 10.1101/2020.03.11.986836 sha: d9c15766ee9c84d154c0f691fc42cbd847668193 doc_id: 958544 cord_uid: qebbkr6d The global spread of SARS-CoV-2 requires an urgent need to find effective therapeutics for the treatment of COVID-19. We developed a data-driven drug repositioning framework, which applies both machine learning and statistical analysis approaches to systematically integrate and mine large-scale knowledge graph, literature and transcriptome data to discover the potential drug candidates against SARS-CoV-2. The retrospective study using the past SARS-CoV and MERS-CoV data demonstrated that our machine learning based method can successfully predict effective drug candidates against a specific coronavirus. Our in silico screening followed by wet-lab validation indicated that a poly-ADP-ribose polymerase 1 (PARP1) inhibitor, CVL218, currently in Phase I clinical trial, may be repurposed to treat COVID-19. Our in vitro assays revealed that CVL218 can exhibit effective inhibitory activity against SARS-CoV-2 replication without obvious cytopathic effect. In addition, we showed that CVL218 is able to suppress the CpG-induced IL-6 production in peripheral blood mononuclear cells, suggesting that it may also have anti-inflammatory effect that is highly relevant to the prevention immunopathology induced by SARS-CoV-2 infection. Further pharmacokinetic and toxicokinetic evaluation in rats and monkeys showed a high concentration of CVL218 in lung and observed no apparent signs of toxicity, indicating the appealing potential of this drug for the treatment of the pneumonia caused by SARS-CoV-2 infection. Moreover, molecular docking simulation suggested that CVL218 may bind to the N-terminal domain of nucleocapsid (N) protein of SARS-CoV-2, providing a possible model to explain its antiviral action. We also proposed several possible mechanisms to explain the antiviral activities of PARP1 inhibitors against SARS-CoV-2, based on the data present in this study and previous evidences reported in the literature. In summary, the PARP1 inhibitor CVL218 discovered by our data-driven drug repositioning framework can serve as a potential therapeutic agent for the treatment of COVID-19. The outbreak of the pneumonia named COVID-19 caused by the novel coronavirus SARS-CoV-2 (2019-nCoV) has infected over 110,000 people worldwide by 8th March, 2020. Apart from China, other countries or regions including South Korea, Iran, and Europe have reported a rapid increase in the number of COVID-19 cases, implying that this novel coronavirus has posed a global health threat. Under the current circumstance of the absence of the specific vaccines and medicines against SARS-CoV-2, it is urgent to discover effective therapies especially drugs to treat the resulting COVID-19 disease and prevent the virus from further spreading. Considering that the development of a new drug generally takes years, probably the best therapeutic shortcut is to apply the drug repositioning strategy (i.e., finding the new uses of old drugs) [1, 2, 3] to identify the potential antiviral effects against SARS-CoV-2 of existing drugs that have been approved for clinical use or to enter clinical trials. Those existing drugs with potent antiviral efficacy can be directly applied to treat COVID-19 in a short time, as their safety has been verified in principle in clinical trials. In this study, we applied a data-driven framework that combines both machine learning and statistical analysis methods to systematically integrate large-scale available coronavirus-related data and identify the drug candidates against SARS-CoV-2 from a set of over 6000 drug candidates (mainly including approved, investigational and experimental drugs). Our in silico screening process followed by experimental validation revealed that a poly-ADP-ribose polymerase 1 (PARP1) inhibitor, CVL218, currently in Phase I clinical trial, may serve as a potential drug candidate to treat Our in vitro assays demonstrated that CVL218 can exhibit effective inhibitory activity against SARS-CoV-2 replication in a dose-dependent manner and with no obvious cytopathic effect. In addition, we found that in human peripheral blood mononuclear cells (PBMCs), CVL218 is able to suppress the CpG-induced production of IL-6, which has been reported previously to be of high relevance to the viral pathogenesis of COVID-19, especially for those intensive care unit (ICU) patients infected by SARS-CoV-2. Further in vivo pharmacokinetic and toxicokinetic studies in rats and monkeys showed that CVL218 was highly distributed in the lung tissue and no apparent sign of toxicity was observed, which makes it an appealing potential drug candidate for the treatment of the novel pneumonia caused by SARS-CoV-2 infection. Moreover, our molecular docking study suggested that CVL218 may bind to the N-terminal domain of nucleocapsid (N) protein of SARS-CoV-2, providing a possible mode of its antiviral action against SARS-CoV-2. Based on the data present in this study and previous known evidences reported in the literature, we also discussed several putative mechanisms of the anti-SARS-CoV-2 effects for CVL218 or other PARP1 inhibitors to be involved in the treatment of COVID-19. Overall, our results indicated that the PARP1 inhibitor CVL218 identified by our drug repositioning pipeline may serve as an effective therapeutic agent against COVID-19. The overview of our data-driven drug repositioning framework is shown in Figure 1A . We first constructed a virus related knowledge graph consisting of drug-target interactions, protein-protein interactions and similarity networks from publically available databases (Methods). Three different types of nodes (i.e., drugs, human targets and virus targets) within the knowledge graph were connected through edges describing their interactions, associations or similarities to establish bridges of information aggregation and knowledge mining. We then applied a network-based knowledge mining algorithm to predict an initial list of drug candidates that can be potentially used to treat SARS-CoV-2 infection ( Figure 1B and Methods). Next, we further narrowed down the list of drug candidates with the previously reported evidences of antiviral activities based on the text mining results from the large-scale literature texts, which were derived through a deep learning based relation extraction method named BERE [4] ( Figure 1C and Methods), followed by a minimum of manual checking. After that, we used the connectivity map analysis approach [5] with the gene expression profiles of ten SARS-CoV-infected patients [6] to further refine the list of drug candidates against SARS-CoV-2 ( Figure 1D , Table 1, Table S1 and Methods). The above screening process revealed that a poly-ADP-ribose polymerase 1 (PARP1) inhibitor PJ-34 could potentially have the antiviral activities against SARS-CoV-2. To demonstrate that our computational pipeline for drug repositioning can yield reasonably accurate prediction results, we also validated our network-based knowledge mining algorithm ( Figure 1B) using the retrospective data of the two coronaviruses that are closely related to SARS-CoV-2 and had been relatively well studied in the literature, i.e., SARS-CoV and MERS-CoV. With the aid of our developed text mining tool BERE, we found that many of the drugs that had been reported previously in the literature to have antiviral activities against the corresponding coronavirus, were also among the top list of our predicted results ( Table 2) . For example, chloroquine, an FDA-approved drug for treating malaria [7] , which was previously reported to exhibit micromolar anti-SARS-CoV activity in vitro [8] , was also repurposed for targeting the same virus by our prediction framework. Gemcitabine, which was originally approved for treating certain types of cancers [9] , was also predicted for targeting SARS-CoV with validation by previous in vitro studies [10] . Cyclosporine, a calcineurin inhibitor approved as an immunomodulatory drug [11] , was observed to block the replication of SARS-CoV [12] , and also successfully predicted by our approach. Among the predicted top list for MERS-CoV, miltefosine, which was approved for treating leishmaniasis [13] , was previously identified to have anti-MERS-CoV activity [14] . Chlorpromazine and imatinib, which were used for treating schizophrenia [15] and leukemia [16] , respectively, were also selected by our computational pipeline as anti-MERS-CoV drugs and can be validated by previous in vitro experiments [10] . Thus, the above retrospective study illustrated that our computational framework is able to predict effective drug candidates against a specific coronavirus. As PJ-34 is still currently in the pre-clinical trial stage (DrugBank ID: DB08348, [17] ), we selected two PARP1 inhibitors, including olaparib and mefuparib hydrochloride (CVL218) ( Figure S1 ), that are currently FDA-approved and at Phase I clinical trial, respectively, for our initial study. We first conducted a pilot experimental test in vitro (Methods) on the anti-SARS-CoV-2 activities of olaparib, CVL218 and several other related drugs (Figure 2A ). We found that both PARP1 inhibitors olaparib and CVL218 exhibited inhibitory effects against SARS-CoV-2 replication. Nevertheless, CVL218 showed a much higher inhibition rate than olaparib. More specifically, olaparib inhibited SARS-CoV-2 replication by 15.48% at a concentration of 3.2 µM, while CVL218 reached 35.16% reduction at a concentration of 3 µM. Notably, the antiviral efficacy of CVL218 even surpassed arbidol, which is one of the standard treatments for COVID-19 in the Diagnosis and Treatment Protocol for Novel Coronavirus Pneumonia (Trial Version 6) promulgated by the Chinese government (http://www.nhc.gov.cn/yzygj/s7653p/202002/8334a8326dd94d329df 351d7da8aefc2/files/b218cfeb1bc54639af227f922bf6b817.pdf). In particular, arbidol inhibited SARS-CoV-2 replication by 21.73% at 3 µM, much lower than that of CVL218 at the same concentration ( Figure 2A ). In contrast, oseltamivir, zanamivir (drugs used for preventing influenza virus infection) and baricitinib (JAK1/2 inhibitor, which was recommended in [18] to treat showed no inhibitory activities against SARS-CoV-2 at the concentration of 3 µM or 3.2 µM. Based on the above pilot experimental results, we then chose CVL218 for subsequent experimental studies. Our further in vitro assays (Methods) showed that CVL218 exhibited effective inhibitory activity against SARS-CoV-2 replication in a dose-dependent manner, with an EC 50 of 5.12 µM ( Figure 2B ). We also assessed the cytotoxicity of CVL218 by the CCK8 assay (Methods), and found that CVL218 had a CC 50 To systematically assess the inhibitory activities of CVL218 against SARS-CoV-2, we also performed a time-of-addition assay (Methods) to determine at which stage CVL218 inhibits viral infection. Remdesivir, which has entered the clinical trials for the treatment of COVID-19 (https://clinicaltrials.gov/ct2/show/NCT0425765 6), was also tested in this assay for comparison. In particular, as compared to the DMSO control group, both CVL218 and remdesivir showed potent antiviral activities during the full-time procedure of the SARS-CoV-2 infection in Vero E6 cells (Figure 2D ). The results of the time-of-addition assay indicated that CVL218 can partially work against the viral entry and significantly inhibit the replication of virus post-entry, while the remdesivir can only function at the post-entry stage ( Figure 2D , 2E). All together, the results of these in vitro assays indicated that CVL218 can be further evaluated as a potential therapeutic agent for treating COVID-19. Recently it has become evident that interleukin-6 (IL-6) is one of the most important cytokines during viral infection [19] , and emerging clinical studies in humans and animals have linked the excessive synthesis of IL-6 with the persistence of many viruses, such as human immunodeficiency virus (HIV) [20] , foot and mouth disease virus [21] and vesicular stomatitis virus (VSV) [22] . In addition, an in vivo study in the Friend retrovirus (FV) mouse model showed that IL-6 blockage can reduce viral loads and enhance virus-specific CD8+ T-cell immunity [23] . These findings supported a hypothesis that rapid production of IL-6 might be a possible mechanism leading to the deleterious clinical manifestations in viral pathogenesis [24] . Recently published researches on the clinical characteristics of severe patients with SARS-CoV-2 infection showed that IL-6 was significantly elevated especially in those ICU patients, which caused excessive activated immune response [25, 26, 27, 28, 29] . The pathological role of IL-6 in SARS-CoV-2 infection indicated that IL-6 blockade may provide a feasible therapy for the treatment of COVID-19. To test whether CVL218 is able to regulate the IL-6 production in vitro, we stimulated the IL-6 production of the peripheral blood mononuclear cells (PMBCs) by CpG-ODN 1826, which is an effective stimulator of cytokines and chemokines. Incubation of PBMCs with 1 µM CpG-ODN 1826 for 6 h (Methods) induced IL-6 production by 40%, when compared to untreated cells ( Figure 3 ). In the presence of CVL218, the stimulatory effect of CpG-ODN 1826 was counteracted. Further study showed that CVL218 inhibited the CpG-induced IL-6 upregulation in a time-and dose-dependent manner ( Figure 3 ). More specifically, exposure with CVL218 at concentrations 1 µM and 3 µM for 12 h attenuated the CpG-induced IL-6 production by 50% and 72.65%, respectively. These results provided an in vitro evidence to support CVL218 as a potential therapeutic agent for treating pro-inflammatory response caused by SARS-CoV-2 infection. Table S2 ), which was also previously reported in [30] . Among seven tissues (i.e., lung, spleen, liver, kidney, stomach, heart and brain), we observed that lung had the highest CVL218 concentration, which was 188-fold higher compared to that of plasma ( Table 3 ). The observation that lung had the highest concentration of CVL218 was in line with the fact that the SARS-CoV-2 virus has the most pathological impact in lung with high viral loads, which suggested that CVL218 has the potential to be used for the indications of the lung lesions caused by SARS-CoV-2 infection, if its antiviral profile can be established in animal models and clinical trials. Furthermore, we compared the pharmacokinetic data between CVL218 and arbidol, a broad-spectrum antiviral drug commercialized in China since 2016, had been recommended to treat the SARS-CoV-2-infected patients by the Chinese government. We found that the pharmacokinetic parameters of CVL218 and arbidol were comparable, with similar plasma concentrations and drug exposures (Table S3) . Arbidol was mostly distributed in stomach and plasma post administration in rats. In contrast, higher distributions of CVL218 in tissues especially in lung rather than plasma compared to those of arbidol indicated a superior pharmacokinetic profile of CVL218, which may render it as a better potential antiviral treatment of SARS-CoV-2 infection in lung. In rats after being orally administrated 20/60/160 mg/kg of CVL218 for 28 consecutive days and followed by 28 more days without drug administration (Methods), we observed no significant difference in body weight of rats among different dosage and the control groups ( Figure 4A ). We next conducted a toxicokinetic analysis of CVL218 in rats (Methods). In particular, rats were given CVL218 20/60/160 mg/kg by oral gavage once a day for consecutive 28 days, followed by 28 days without CVL218 administration, to investigate the reversibility of the toxic effects of the compound and examine whether there is any potential delayed-onset toxicity of this drug in rats. The results showed that, the maximum tolerable dose (MTD) and the no-observed adverse effect level (NOAEL) were 160 mg/kg and 20 mg/kg, respectively. The exposure of female rats to CVL218 (AUC 0−24 ) was 7605 h·ng/mL in day 1 and 6657 h·ng/mL in day 28, while that of male rats (AUC 0−24 ) was 9102 h·ng/mL in day 1 and 10253 h·ng/mL in day 28 (Table S4) . Based on the toxicokinetic results from the repeated dose studies, all rats survived after a 28-day treatment period and showed no apparent signs of toxicity. Monkeys were administered CVL218 (5, 20 or 80 mg/kg) by nasogastric feeding tubes with a consecutive daily dosing schedule for 28 days, followed by a 28-day recovery period (Methods). Only a slight decrease of body weight was observed in the highdose (80 mg/kg) group, and all changes were reversed after a 28-day recovery period Overall, the above in vivo data showed that CVL218 possesses good pharmacokinetic and toxicokinetic characteristics in rats and monkeys, and its high-level distribution in the therapeutically targeted tissue (i.e., lung) may greatly favor the treatment of SARS-CoV-2 infection. As the previous studies have reported that the PARP1 inhibitor PJ-34 can target the N-terminal domain (NTD) of the coronavirus nucleocapsid (N) protein to reduce its RNA binding and thus impede viral replication [31, 32, 33] , we speculated that olaparib and CVL218 may also interact with the N protein of SARS-CoV-2 to perform the similar antiviral function. To test this hypothesis, we conducted molecular docking to study the potential interactions between these two drugs and the N-NTD of SARS-CoV-2 (Methods). The overall structure of HCoV-OC43 (another coronavirus phylogenetically closely related to SARS-CoV-2) and SARS-CoV-2 share similar compositions of secondary structure elements, including five conserved β strands and flexible loops ( Figure 5A ). In addition, their sequences covering the corresponding binding pocket regions are well conserved ( Figure 5C ). Thus, their structures may provide common molecular features in terms of interactions with small molecules at this binding pocket. Therefore, we also used the experimentally solved structure of HCoV-OC43-N-NTD complexed with PJ-34 as a reference to analyze our docking results. Our docking results (more details can also be found in Supplementary In this study we reported a top down data integration approach by combining both machine learning and statistical analysis techniques, followed by web-lab experimental validation, to identify potential drug candidates for treating SARS-CoV-2 infection. We showed that the PARP1 inhibitor CVL218 discovered by our in silico drug repurposing framework may have the therapeutic potential for the treatment of COVID-19. Although we mainly conducted in vitro assays to experimentally validate the anti-SARS-CoV-2 effects of olaparib and CVL218 due to limited time, it is natural to speculate that other PARP1 inhibitors may also have antiviral activities against SARS-CoV-2 infection, based on our computational prediction and experimental validation results. Based on the data present in this study and the previously known evidences reported in the literature, we propose several potential mechanisms to help understand the involvement of PARP1 inhibitors in the treatment of COVID-19 ( Figure 6 ). First, during the life cycle of the coronavirus, PARP1 inhibitors may inhibit the viral growth through suppressing viral replication and impeding the binding of the nucleocapsid protein to viral RNAs [31, 34, 35, 36] , which can also be supported by our molecular docking results (see Section 2.6). Second, PARP1 inhibitors have been previously reported to play a critical role in regulating inflammatory response by modulating the expression of pro-inflammatory factors such as NF-κB, AP-1, IL-6 and downstream cytokines and chemokines [37, 38, 39, 40] . Also, it has been shown that the overactivation of PARP1 promotes energy (NAD + /ATP) consumption and drives cell death, exacerbating the inflammation response [37, 38, 39, 41] . PARP1 inhibitors thus may repress the NF-κB-mediated pro-inflammatory signals, and reduce energy failure and subsequent cell death induced by necrosis, thus providing a clinical potential for attenuating the cytokine storm and inflammatory response caused by SARS-CoV-2 infection. Third, ADP-ribosylation is a conserved post-translational modification on the nucleocapsid proteins across different coronavirus lineages, implying that it may have an important regulatory role for the structure packing of viral genome. Several previous studies have demonstrated that PARP1 is critical for viral replication [35, 42, 43] . For example, PARP1 has been reported to interact with hemagglutinin (HA) of influenza A virus (IAV) and promote its replication by triggering the degradation of host type I IFN receptor [44] . In addition, the ADP-ribosylation of adenoviral core proteins displays an antiviral defense mechanism [34] . Therefore, intervening the ADP-ribosylation mediated interplay between PARP1 and viral proteins may be another important pathway for PARP1 inhibitors to prevent SARS-CoV-2 infection. Of course, to throughly understand the anti-SARS-CoV-2 roles of PARP1 inhibitors, more experimental studies and direct (clinical) evidences will be needed in the future. Considering the pro-inflammatory role of PARP1, the therapeutic effects of PARP1 inhibitors in inflammatory-mediated diseases have been extensively studied over past two decades [45, 46] . PJ-34, the early generation PARP1 inhibitor, has been suggested in previous studies to have neuroprotective effects in stroke model and protect mice from necroptosis-associated liver injuries by repressing the IL-33 expression [47, 48] . In addition, the FDA-approved PARP1 inhibitor, olaparib, has been reported to protect against the LPS (Lipopolysaccharide)-induced acute lung and kidney injuries in a NF-κB-dependent manner in mice [49] . Numerous pre-clinical studies demonstrated that PARP1 inhibitors play an essential role in a range of inflammatory injuries and related diseases, especially the lung inflammatory disorders including ARDS (Acute Respiratory Distress Syndrome), COPD (Chronic Obstructive Pulmonary Disease) and asthma [40, 46, 50, 51] . All these studies suggest that PARP1 inhibitors are of high relevance to the treatment of the novel pneumonia caused by SARS-CoV-2 infection, possibly via their roles in modulating inflammatory response. Notably, current pathological studies have shown that the severe patients infected by SARS-CoV-2 generally have higher plasma levels of IL-2, IL-6, IL-10, TNFα, IFN-γ [25, 27, 28, 29] , implying a high risk of the inflammatory-associated cytokine storm after viral infection. In addition, reduction and functional exhaustion of T cells have also been observed in COVID-19 patients [27] . Therefore, blocking the overactive inflammatory response may be an effective strategy for the treatment of COVID-19, particularly for those ICU patients infected by SARS-CoV-2. Recently, tocilizumab, a monoclonal anti-body drug targeting IL-6, has been recommended for the treatment of COVID-19 in the Diagnosis and Treatment Protocol for Novel Coronavirus Pneumonia (Trial Version 7) promulgated by the Chinese government (http://www.nhc.gov.cn/yzygj/s7653p/20 e964.pdf), which also highlights the vital role of anti-inflammatory response in current therapeutics against SARS-CoV-2. Our in vitro study has showed that CVL218 can effectively inhibit the IL-6 production induced by CpG in PBMCs (Figure 3 ). This finding indicates that CVL218 may also possess the IL-6 specific anti-inflammatory effect that is applicable to those severe patients infected by SARS-CoV-2. PARP1 inhibitors are originally used for targeting homologous recombination repair defects in cancers, and mainly categorized as oncology drugs. Thus, it would generally need more safety data to justify any repurposing of PARP1 inhibitors for non-oncology indications. Fortunately, there are numerous existing pre-clinical and clinical studies on repurposing PARP1 inhibitors into non-oncological diseases, including the aforementioned acute diseases (e.g., acute respiratory distress syndrome (ARDS), stroke) [52] and chronic diseases (e.g., rheumatoid arthritis and vascular diseases) [52, 53] . All these evidences indicate the possibility of repurposing PARP1 inhibitors as a safe therapeutic agent to treat the current acute lung disease caused by SARS-CoV-2 infection. In addition, our pharmacokinetic and toxicokinetic data in rats and monkeys shown in our study indicate that CVL218 may have a relatively acceptable safety profile to be repositioned for the antiviral purpose. Moreover, CVL218 has been approved to enter Phase I clinical trial in 2017 by National Medical Products Administration (NMPA) in China for cancer treatment. The preliminary data from the Phase I clinical trial have shown that CVL218 is well tolerated in ascending dose studies at doses as high as 1000 mg QD and 500 mg BID, and no Grade II and above adverse events have been observed, which indicates that CVL218 is also quite safe and well tolerated in human. Our pharmacokinetic examination in rats has shown that CVL218 has the highest level distribution in the lung tissue, 188-fold higher concentration compared to that in plasma. Such a tissue specific enrichment in lung may bring an extra advantage for CVL218 to be used for the anti-SARS-CoV-2 purpose, as lung is the therapeutically targeted tissue for COVID-19. Moreover, high level distribution in lung may also suggest that only low dosage is needed in order to ensure the therapeutic efficacy of CVL218 against SARS-CoV-2, which may further reduce the risk of adverse events. Thus, CVL218 may have great potential to be repurposed as an effective therapeutic agent to combat SARS-CoV-2 and prevent future epidemic outbreak. The virus-related knowledge graph was constructed for predicting the coronavirus related drugs. In total seven networks were considered in the constructed knowledge graph (Figure 1B) , including a human target-drug interaction network, a virus target-drug interaction network, a human protein-protein interaction network, a virus protein-human protein interaction network, a drug molecular similarity network, a human protein sequence similarity network, and a virus protein sequence similarity network. The human target-drug interaction network was derived from DrugBank (version 5.1.0) [17] . The virus target-drug interaction network was constructed from the integrated data from DrugBank (version 5.1.0) [17] , ChEMBL (release 26) [54] , TTD (last update 11 Nov, 2019) [55] , IUPHAR BPS (release 13, Nov, 2019) [56] , BindindDB [57] and GHDDI (https://ghddi-ailab.github.io/Targeting2019-nCo V/CoV_Experiment_Data/), with a cut-off threshold of IC 50 /EC 50 /K i /K d <10 µM. The human protein-protein interaction network and the virus protein-human protein interaction network were constructed from the integrated data from BioGRID (release 3.5.181) [58] , HuRI [59] , Instruct [60] , MINT (2012 update) [61] , PINA (V2.0) [62] , SignaLink (V2.0) [63] and innatedb [64] . The drug molecular similarity network was obtained by calculating the Tanimoto similarities from Morgan fingerprints with a radius of 2 computed using the rdkit tool [65] . The protein sequence similarity networks of both human and virus were obtained by calculating the Smith-Waterman similarities of the amino acid sequences derived from UniProt [66] using a sequence alignment software provided in [67] . Noted that we collected additional protein sequences of SARS-CoV-2 from UniProt [66] and added them into the corresponding networks for the final prediction. Those drugs without drug-target interactions or outside the Drug-Bank database were removed from the corresponding networks. We then constructed the virus-related knowledge graph by merging together all the nodes and edges of the above seven networks ( Figure 1B) . The constructed knowledge graph G = (V, E) is an undirected graph, in which each node v ∈ V in the node set V belongs to one of the node types (including drugs, human proteins, and virus proteins), and each edge e ∈ E in the edge set E ⊂ V × V × R belongs to one of the relation types from the relation type set R (including two drug-target interactions, two protein-protein interactions and three similarities). The initial list of drug candidates targeting SARS-CoV-2 was first screened using a network-based knowledge mining algorithm modified from our previous work [68, 69] . The goal was to capture the hidden virus-related feature information and accurately predict the potential drug candidates from the constructed knowledge graph, which was realized through learning a network topology-preserving embedding for each node. More specifically, our model used a graph convolution algorithm [70] to gather and update feature information for each node in the constructed heterogeneous knowledge graph network from neighborhoods so that the network topology information can be fully exploited. Suppose that we perform T iterations of graph convolution. At iteration 1 ≤ t ≤ T , the message m t v passed to node v can be expressed as: where a v,u,r stands for the weight for edge e = (u, v, r), A u,v,r = av,u,r u av,u,r , W t r ∈ R d×d and b t r ∈ R d stand for the learnable parameters, ReLU (x) = max(0, x), and N r (v) = {u, u ∈ V, u = v, (u, v, r) ∈ E} denotes the set of adjacent nodes connected to v ∈ V through edges of type r ∈ R. Then the feature h t v of node v is updated by where W t ∈ R d×d and b t ∈ R d stand for the learnable parameters, and concat(·, ·) stands for the concatenation operation. Finally, the confidence score s u,v of the relation r between node u and node v is derived from the learned node embeddings and the corresponding projection matrices, that is, where G r , H r ∈ R d×k stand for the edge-type specific projection matrices. We minimized the Bayesian personalized ranking (BPR) loss [71] for drug-target interaction reconstruction, by regarding those edges not in the edge set E as missing values rather than negative samples, that is, where, s u,v and s w,x stand for the confidence scores of the relation r between u and v and between w and x, respectively, and σ(·) stands for the sigmoid activation function. Intuitively, in the above loss function, the confidence scores of the node pairs (u, v) in the edge set (i.e., (u, v, r) ∈ E) were encouraged to be higher than those of unseen pairs (w, x) (i.e., (w, x, r) / ∈ E). We predicted the confidence scores under the relation of virus target-drug interactions for each virus target-drug pair using Equation (3). Then the confidence scores were averaged across all the proteins of a certain virus (e.g., SARS-CoV, MERS-CoV or SARS-CoV-2), and the corresponding p-values were obtained by z-test. For each virus, we selected those predictions with a p-value < 0.05 as drug candidates. We used a deep learning based relation extraction method named BERE [4] to extract the coronavirus related drugs from large-scale literature texts. More specifically, the sentences mentioning the two entities of interest, i.e., name (or alias) of a coronavirus or coronavirus target, or name (or alias) of a drug, were first collected using a dictionarybased name entity recognition method (string matching). For each pair of entities (e 1 , e 2 ), there are usually more than one sentence describing the underlying relations. Therefore, we used a bag of sentences S e 1 ,e 2 , denoting the set of all the sentences mentioning both e 1 and e 2 , to predict the relation between these two entities. We first encoded each sentence s ∈ S e 1 ,e 2 in a semantic and syntactic manner using a hybrid deep neural network (h : s → R d ), including a self-attention module [72] , a bidirectional gated recurrent unit (GRU) module [73] and a Gumbel tree-GRU module [4, 74] . Each sentence representation h(s) was then scored by a sentence-level attention module to indicate its contribution to the relation prediction, that is, where β(s) ∈ R stands for the weight score, and W s ∈ R d×1 stands for the learnable weight parameters. Finally, the relation was predicted by a binary classifier, based on the weighted sum of sentence representations, that is, r e 1 ,e 2 = classifier where r e 1 ,e 2 stands for the probability of the relation of interest between entities e 1 and e 2 mentioned by the bag of sentences S e 1 ,e 2 . The training corpus we used was curated automatically from nearly 20 million PubMed (http://www.pubmed.gov) abstracts by a distant supervision technique [75] . In detail, the names (or aliases) of drugs or targets in sentences were first annotated by a dictionary-based named entity recognition method (string matching), in which the name dictionary was derived from DrugBank (version 5.1.0) [17] , with ambiguous names (e.g., common words) removed. Next, the label for each bag of sentences co-mentioning a drug-target pair of interest was annotated automatically by the known drug-target interactions in DrugBank. The unlabeled corpus that we used in this work for text-mining the coronavirus related drugs was obtained from approximately 2.2 million PMC full-text articles, with entities of interest annotated using the aforementioned named entity recognition approach. A coronavirus related drug was extracted as a hit candidate if the model found a bag of sentences describing a relation between this drug and a target in the coronavirus of interest. We used the transcriptome analysis approach to further filter the potential drug candidates for treating the COVID-19 patients infected by SARS-CoV-2. Due to the lack of gene expression data from the SARS-CoV-2 infected patients, we used those from the SARS-CoV infected patients to screen the potential therapeutic drug candidates against COVID-19. Such a strategy is reasonable as SARS-CoV and SARS-CoV-2 are two closely related and highly similar coronavirus. First, the genome of SARS-CoV-2 is phylogenetically close to that of SARS-CoV, with about 79% of sequence identity [76] , and the M (membrane), N (nucleocapsid) and E (envelope) proteins of these two coronaviruses have over 90% sequence similarities [77] . In addition, the pathogenic mechanisms of SARS-CoV-2 and SARS-CoV were highly similar [78] . In particularly, we collected the gene expression profiles of the peripheral blood mononuclear cells (PBMCs) from ten SARS-CoV infected patients (GEO:GSE1739) [6] . The raw gene expression values were first converted into logarithm scale, and then the differential expression values (z-scores) were computed by comparing to those of healthy persons using the same protocol as described in [5] , that is, MAD(X healthy ) =median(|X healthy − median(X healthy )|), The African green monkey kidney Vero E6 cell line was purchased from the Cell Thirty Sprague-Dawley rats were randomly divided into three time point groups (3/sex/group). At 3, 6 and 8h after CVL218 administration, animals were sacrificed, and the brain, heart, lung, liver, spleen, stomach and kidney tissues were collected. Tissue samples were washed in ice-cold saline, blotted with paper towel to remove excess fluid, and weighed. Tissue samples were fluid, weighted and stored at −20 ± 2 • C until the determination of drug concentration by LC-MS-MS. Healthy male and female cynomolgus monkeys aged 3-4 years were purchased from Guangdong Landau Biotechnology, China. The animals were maintained in accordance with the Guide for the Care and Use of Laboratory Animals. Cynomolgus monkey (5/sex/group) were selected using a computerized randomization procedure, and administered CVL218 by nasogastric feeding at dose levels of 0 (control), 5, 20, 80 mg/kg. Individual dose volumes were adjusted weekly based on body weight of monkeys. The monkeys were observed twice daily for viability/mortality and for any change in behavior, reaction to treatment or ill-health. Electrocardiograms, intraocular pressure, rectal temperature and body weight were recorded. For all the groups, 2/3 of the animals were randomly selected and euthanized at day 28. The remaining animals were euthanized after a 28-day drug free period. Blood samples were taken before and at 0. All data represent the means ± standard deviations (SDs) of n values, where n corresponds to the number of data points used. The figures were prepared using GraphPad Prism (GraphPad Software, USA). The statistical significance was calculated by SPSS (ver.12), and two values were considered significantly different if the p-value is < 0.05. The docking program AutoDock4.2 [79] We noticed that the binding surface of HCoV-OC43-N-NTD with PJ-34 consists of the N-terminal loop, β2 and β3 strands ( Figure 5B a. The pharmacokinetic data of arbidol were obtained from [91] . Drug repositioning: identifying and developing new uses for existing drugs Drug repurposing: progress, challenges and recommendations Network-based drug repurposing for human coronavirus BERE: An accurate distantly supervised biomedical entity relation extraction network A next generation connectivity map: L1000 platform and the first 1,000,000 profiles Expression profile of immune response genes in patients with severe acute respiratory syndrome StatPearls [Internet], StatPearls Publishing In vitro inhibition of severe acute respiratory syndrome coronavirus by chloroquine Repurposing of clinically developed drugs for treatment of Middle East respiratory syndrome coronavirus infection Ciclosporin 10 years on: indications and efficacy The SARScoronavirus-host interactome: identification of cyclophilins as target for pancoronavirus inhibitors Miltefosine, an FDA-approved drug for the 'orphan disease', leishmaniasis Antiviral potential of ERK/MAPK and PI3K/AKT/mTOR signaling modulation for Middle East respiratory syndrome coronavirus infection as identified by temporal kinome analysis Chlorpromazine for schizophrenia: a Cochrane systematic review of 50 years of randomised controlled trials Approval summary for imatinib mesylate capsules in the treatment of chronic myelogenous leukemia DrugBank 5.0: a major update to the DrugBank database for Baricitinib as potential treatment for 2019-nCoV acute respiratory disease, The lancet The role of interleukin 6 during viral infections Differences in wound healing in mice with deficiency of IL-6 versus IL-6 receptor IL6 inhibits HBV transcription by targeting the epigenetic control of the nuclear cccdna minichromosome Increased virulence of an epidemic strain of vesicular stomatitis virus is associated with interference of the innate response in pigs TLR ligand induced IL-6 counter-regulates the antiviral CD8+ T cell response during an acute retrovirus infection The expression of IL-6, tNF-α, and MCP-1 in respiratory viral infection in acute exacerbations of chronic obstructive pulmonary disease Aberrant pathogenic GM-CSF+ T cells and inflammatory CD14+ CD16+ monocytes in severe pulmonary syndrome patients of a new coronavirus Clinical characteristics of 140 patients infected by SARS-CoV-2 in Reduction and functional exhaustion of T cells in patients Clinical features of patients infected with 2019 novel coronavirus in Immunoglobulin G/M and cytokines detections in continuous sera from patients with novel coronaviruses (2019-nCoV) infection Novel PARP1/2 inhibitor mefuparib hydrochloride elicits potent in vitro and in vivo anticancer activity, characteristic of high tissue distribution Structural basis for the identification of the N-terminal domain of coronavirus nucleocapsid protein as an antiviral target Structure-based virtual screening and experimental validation of the discovery of inhibitors targeted towards the human coronavirus nucleocapsid protein Coronavirusesdrug discovery and therapeutic options Possible role of ADP-ribosylation of adenovirus core proteins in virus infection Resolution of the cellular proteome of the nucleocapsid protein from a highly pathogenic isolate of porcine reproductive and respiratory syndrome virus identifies PARP-1 as a cellular target whose interaction is critical for virus biology Regulation of Epstein-Barr virus OriP replication by poly (ADP-ribose) polymerase 1 Transcriptional reprogramming and resistance to colonic mucosal injury in poly (ADP-ribose) polymerase 1 (PARP1)-deficient mice The rheumatoid arthritis risk variant CCR6DNP regulates CCR6 via PARP-1 PARP-1 inhibitors DPQ and PJ-34 negatively modulate proinflammatory commitment of human glioblastoma cells Poly (ADP-ribose): novel functions for an old molecule PARP-1 and YY1 are important novel regulators of CXCL12 gene transcription in rat pancreatic beta cells Host-and strainspecific regulation of influenza virus polymerase activity by interacting cellular proteins Herpes simplex virus 1 infection activates poly (ADP-ribose) polymerase and triggers the degradation of poly (ADP-ribose) glycohydrolase PARP1 enhances influenza A virus propagation by facilitating degradation of host type I interferon receptor Therapeutic applications of PARP inhibitors: anticancer therapy and beyond ADP-ribose) polymerase and the therapeutic effects of its inhibitors Protective effects of PJ34, a novel, potent inhibitor of poly (ADPribose) polymerase (PARP) in in vitro and in vivo models of stroke The chemical inhibitors of cellular death, PJ34 and Necrostatin-1, down-regulate IL-33 expression in liver PARP inhibitor, olaparib ameliorates acute lung and kidney injury upon intratracheal administration of LPS in mice Multifaceted role of PARP-1 in DNA repair and inflammation: Pathological and therapeutic implications in cancer and non-cancer diseases Poly (ADP-ribose) polymerase-1 in lung inflammatory disorders: a review Opportunities for the repurposing of PARP inhibitors for the therapy of non-oncological diseases The role of poly (ADP-ribose) polymerase-1 in rheumatoid arthritis ChEMBL: towards direct deposition of bioassay data Therapeutic target database 2020: enriched resource for facilitating research and early development of targeted therapeutics The IUPHAR/BPS Guide to PHARMACOLOGY: an expert-driven knowledgebase of drug targets and their ligands BindingDB in 2015: a public database for medicinal chemistry, computational chemistry and systems pharmacology The BioGRID interaction database: 2019 update Mapping the human protein interactome INstruct: a database of high-quality 3D structurally resolved protein interactome networks MINT, the molecular interaction database: 2012 update 0: mining interactome modules COBRApy: COnstraintsbased reconstruction and analysis for python InnateDB: systems biology of innate immunity and beyondrecent updates and continuing curation RDKit: Open-source cheminformatics UniProt: the universal protein knowledgebase SSW library: an SIMD Smith-Waterman C/C++ library for use in genomic applications A network integration approach for drug-target interaction prediction and computational drug repositioning from heterogeneous information NeoDTI: neural integration of neighbor information from a heterogeneous network for discovering new drug-target interactions Neural message passing for quantum chemistry BPR: Bayesian personalized ranking from implicit feedback Attention is all you need On the properties of neural machine translation: Encoder-decoder approaches Categorical reparameterization with gumbel-softmax Modeling relations and their mentions without labeled text Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor binding Preliminary identification of potential vaccine targets for the COVID-19 coronavirus (SARS-CoV-2) based on SARS-CoV immunological studies Systematic comparison of two animal-to-human transmitted human coronaviruses: SARS-CoV-2 and SARS-CoV AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility Protein structure and sequence re-analysis of 2019-nCoV genome does not indicate snakes as its intermediate host or the unique similarity between its spike protein insertions and HIV-1 MUSCLE: multiple sequence alignment with high accuracy and high throughput Middle East respiratory syndrome and severe acute respiratory syndrome: current therapeutic options and potential targets for novel therapies How prepared are we to control severe acute respiratory syndrome in future Therapeutic roles of curcumin: lessons learned from clinical trials Plantas de Santa Marta con posible actividad biológica antimicrobiana The therapeutic effects of PJ34 [N-(6-oxo-5, 6-dihydrophenanthridin-2-yl)-N, N-dimethylacetamide. HCl], a selective inhibitor of poly experimental allergic encephalomyelitis are associated with immunomodulation Ambient audio transformation using transformation audio, uS Patent 8 Lipid-lowering efficacy of hesperetin metabolites in high-cholesterol fed rats Anti-SARS coronavirus 3C-like protease effects of isatis indigotica root and plant-derived phenolic compounds Saracatinib inhibits middle east respiratory syndrome-coronavirus replication in vitro Pharmacokinetics of arbidol hydrochloride in rats Distribution and excretion of arbidol hydrochloride in rats