In Silico Models for Anti-COVID-19 Drug Discovery: A Systematic Review

The coronavirus disease 2019 (COVID-19) is a severe worldwide pandemic. Due to the emergence of various SARS-CoV-2 variants and the presence of only one Food and Drug Administration (FDA) approved anti-COVID-19 drug (remdesivir), the disease remains a mindboggling global public health problem. Developing anti-COVID-19 drug candidates that are effective against SARS-CoV-2 and its various variants is a pressing need that should be satisfied. This systematic review assesses the existing literature that used in silico models during the discovery procedure of anti-COVID-19 drugs. Cochrane Library, Science Direct, Google Scholar, and PubMed were used to conduct a literature search to find the relevant articles utilizing the search terms “In silico model,” “COVID-19,” “Anti-COVID-19 drug,” “Drug discovery,” “Computational drug designing,” and “Computer-aided drug design.” Studies published in English between 2019 and December 2022 were included in the systematic review. From the 1120 articles retrieved from the databases and reference lists, only 33 were included in the review after the removal of duplicates, screening, and eligibility assessment. Most of the articles are studies that use SARS-CoV-2 proteins as drug targets. Both ligand-based and structure-based methods were utilized to obtain lead anti-COVID-19 drug candidates. Sixteen articles also assessed absorption, distribution, metabolism, excretion, toxicity (ADMET), and drug-likeness properties. Confirmation of the inhibitory ability of the candidate leads by in vivo or in vitro assays was reported in only five articles. Virtual screening, molecular docking (MD), and molecular dynamics simulation (MDS) emerged as the most commonly utilized in silico models for anti-COVID-19 drug discovery.


Introduction
COVID-19 is one of the most signifcant infectious illnesses in the world. It afects hundreds of millions annually and is the main factor behind socioeconomic loss in developing nations. As of 21st March 2023, the World Health Organization (WHO) reported approximately 761,071,826 COVID-19 cases and 6,879,677 confrmed deaths [1]. Te emergence of SARS-CoV-2 variants and the availability of only one FDA-approved anti-COVID-19 drug are the main issues in controlling and treating COVID-19. Terefore, the need to fnd efective anti-COVID-19 drug candidates is crucial. However, fnding and developing new drugs takes time and money [2]. According to the 2016 Tufts Center for the Study of Medicine Development research, developing a new drug typically takes more than ten years and costs more than $2.6 billion [3]. Executing an ideal drug discovery and development method is one of the main issues facing the pharmaceutical research community [4]. In silico drug design and development represents a technique to expedite drug discovery and development procedures efciently as one of the main aims.
In silico drug design and discovery is a rigorous procedure of fnding novel drugs based on understanding a biological target. It is essential to create tiny molecules for complementary drugs in charge and form the biomolecular targets they interact with [5]. Discovering small compounds that preferentially bind to the biological target with the highest binding afnity is crucial. Finding and developing novel drugs faces new challenges and opportunities due to recent advances in bioinformatics and other omics approaches like genomics and proteomics. Several disciplines such as computer science, biological sciences, and information technology or informatics have greatly benefted from the fusion of these technological advances. Protein networks and other fast-developing information on drugtarget interactions (DTI), gene expression, and other topics are becoming more widely available and standardized [6].
Trough the use of numerous readily accessible databases of chemical compounds and proteins like protein data bank (PDB) where SARS-CoV-2 target proteins, such as main protease (M pro ), spike protein, and RNA-dependent RNA polymerase (RdRp), can be retrieved, in silico or computational-based methods can speed up the development process for anti-COVID-19 drugs. During the computer-aided drug discovery process, the costs are often insignifcant because humans are rarely in danger, expenditures are negligible, and biosafety facilities are unnecessary [5]. However, despite discovering new drugs through in silico means, several usually fail during clinical trials due to toxicity and poor pharmacokinetics features. Tese pharmacokinetics characteristics, including ADMET, are crucial for discovering and developing new medicines [5]. Tis is evident in the studies by Adel et al. [7] and Shabaan et al. [8] that performed ADMET properties analysis of potential anti-COVID-19 compounds. Terefore, subjecting newly discovered drug candidates to ADMET or pharmacokinetics properties analysis and prediction can assist in eliminating molecules with unfavorable drug ability characteristics. In silico tools like SwissADME can be utilized for ADMET or pharmacokinetics properties analysis and forecast of drug candidates [2,7,8]. Similarly, molecular modeling can be applied to ADMET or the candidate compounds' pharmacodynamics (toxicity and drug action) characteristics [9].
Computer-aided drug design and discovery have been accomplished via structure-and ligand-based drug development [10]. Te structure-based drug design is called direct drug design [5]. It provides an avenue to create new molecular entities interacting with specifc biological targets using a model of the said targets [11]. Structure-based drug design and development require an in-depth understanding of the biological target's three-dimensional structure. NMR spectroscopy and X-ray crystallography are techniques used to attain the three-dimensional structures of biological targets [12]. Tese three-dimensional structures are often handy when computational approaches like 3D-QSAR involving force feld calculations are performed based on molecular superimposition or protein crystallography. With the help of interactive visuals, a medicinal chemist's intuition, and several automated computational techniques, candidate medications that are anticipated to bind to a particular biological target with high selectivity and afnity can be created [10]. On the other hand, ligand-based drug design and discovery, also called indirect drug design, requires a profound comprehension of other compounds (ligands) that attach to the desired biological target [5]. In some instances, such molecules can create one reference ligand that functions as a pharmacophore model, establishing the minimum requirements for a molecule to bind to a target, as evident in Onyango et al. [2]. Tis systematic review analyzed research publications employing in silico techniques to fnd new anti-COVID-19 drugs, summarized, and presented that information, which is crucial for further discovery and development of efective anti-COVID-19 medications.

Study Design.
Tis systematic review evaluated the computational in silico models used to discover new anti-COVID-19 drugs.

Literature
Searches. Te following electronic databases were searched for research studies published between January 2019 and the end of December 2022: PubMed, Google Scholar, Cochrane Library, and Science Direct, according to the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA). Te Boolean operators "OR" and "AND" were also used to help in the literature search by combining the following keywords and terms "Drug Discovery," "Anti-COVID-19," "In Silico Models," "Computer-Aided Drug Design," and "Drug Development." Since language restrictions do not afect or alter the results of systematic reviews, all searches were limited to studies published in English. Additional articles were searched by thoroughly examining the reference lists of research publications retrieved from the electronic databases. Duplicate research publications were documented and omitted from the review.

Eligibility Criteria.
All research publications that employed diferent in silico models for discovering novel anti-COVID-19 drugs published in English from 2019 to December 31, 2022, were included in the study. Studies were not fully accessible by virtue of being behind a paywall, publications on diseases other than COVID-19, and duplicate articles were excluded.

Study Selection Process.
Te searched studies from all databases were randomly downloaded to reduce the chances of bias. All downloaded publications were individually reviewed to determine their eligibility. Te titles and authors of each research study were examined, and duplicates were removed. All downloaded publications' titles and abstracts were then screened for potential relevance. Te full-text review was performed on the studies for which there was uncertainty about signifcance. Publications that failed to meet the inclusion criteria were excluded.

Data Extraction and Synthesis.
Te data were manually extracted from the research publications and recorded in a table. Te following data were extracted from each of the research articles included in the review: the title of the article, the reference (names of the authors and years of publication), in silico methods used (2-or 3-D Quantitative Structure-Activity Relationship: QSAR, pharmacophore modeling, MD, homology modeling, and others), software packages and web-based databases and servers utilized, biological targets, lead or hit compounds, and experimental techniques (in vivo or in vitro assays) where applicable. Tematic analysis was used to synthesize the data, and similar information was grouped into columns, as displayed in Table 1.

Study Selection.
A comprehensive search of the electronic databases yielded 1,105 possibly associated research publications based on the keywords and search terms described previously, published between 2019 and 2022 ( Figure 1). A thorough review of the reference lists of some of these electronically retrieved research articles provided 15 additional publications linked to the topic of interest. Terefore, 1,120 research articles were obtained from the initial phase of the literature search. Te titles of these 1,120 articles were checked to identify duplicates and studies that were not original research, for instance, reviews. From these initial publications, 840 articles were excluded for the following reasons: duplicates (411) and review publications (429). From the remaining 280 articles, 191 were excluded after a review of their titles and abstracts confrmed nonrelevance to this review. Terefore, 89 full-text publications were sought for retrieval. However, only 81 full-text publications could be retrieved. Te full texts of 8 publications were inaccessible. In this regard, 81 articles were examined for eligibility based on the preset criteria, further excluding 48 studies primarily due to their failure to report the outcome of interest. Eventually, 33 publications met the eligibility criteria and were included in the fnal review ( Figure 1 and Table 1). Table 1 summarizes the information collected from the 33 articles included in this current systematic review. Te data were categorized into diferent themes: title of the article, reference (authors and year of publication), in silico methods/software/databases used, drug target, lead candidates, and experimental technique. Table 1 shows that only one author published one article [37]. Te other three articles were published by two authors [23,32,41]. More than two authors published all the remaining twenty-nine studies. All articles had diferent titles directly associated with the topic of interest. Tey were published during diferent periods within the 2019-2022 timeframe. 9.09% of the articles were published in 2020. 57.58% of the reports were published in 2021. 33.33% of the studies were published in 2022.

Study Characteristics.
Te published in silico models were mainly applied for identifying prospective anti-COVID-19 drugs using SARS-CoV-2 proteins and some human proteases as targets. Seven studies undertook pharmacophore modeling, chemical synthesis of ligands, ligand database creation, or homology modeling as preliminary steps of anti-COVID-19 drug discovery. 16 publications performed virtual screening to discover compounds with inhibitory efects on SARS-CoV-2 target proteins. MD emerged as a crucial step in fnding anti-COVID-19 drugs. Twenty-six articles employed MD to test the binding afnities of their lead compounds to SARS-CoV-2 target proteins. Another in silico process that was common was molecular dynamics simulation. Twenty-three studies applied the method to ascertain the stability of their ligandtarget protein complexes. After MDS, 25 publications underscored the need for drug-likeness and physicochemical and pharmacokinetics assessment. Ten reports analyzed the drug-likeness of their lead compounds, while 15 articles assessed their drug candidates' physicochemical and pharmacokinetics properties. Tese two in silico procedures were performed to virtually confrm the lead compounds' drug ability.
Te lead candidates from each publication depended on the drug target and the databases used for virtual screening. Terefore, the lead candidates ranged from diazole, furan, and pyridine to gliquidone, glimepiride, and linagliptin (Table 1). However, the drug targets used were 9: RdRp (5 articles), spike protein (5 articles), main protease (25 publications), nsp16 (2 articles), nsp15 (1 article), nsp12 (1 study), PL pro (3 studies), TMPRSS2 (2 publications), and ACE2 (5 articles). Te most common drug target in the fght against COVID-19 is SARS-CoV-2 main protease. Even though human proteases or enzymes like ACE2 and TMPRSS2 are also used, SARS-CoV-2 proteins are preferred. Some studies (5) performed in silico approaches and in vitro validation of their lead candidates. Although the remaining 28 articles did not undertake in vitro validation of their drug candidates, they recommended additional clinical processes to ascertain the use of their lead compounds as anti-COVID-19 drugs. Figure 2 is a fowchart diagram summarizing the most commonly used in silico models for anti-COVID-19 drug discovery.
Even though all these researchers did not perform in vitro validation of the inhibitory ability of these lead compounds, they utilized relatively comparable in silico approaches to discover them. Gosh et al. [14] employed QSAR/SiRMS tools Publications not retrieved (n = 8) Articles examined for eligibility (n = 81) Studies excluded: Outcome of interest not reported (43) Reported as systematic reviews (5) Articles included in review (n = 33) Records of included articles (n = 33) Identification Screening Included Figure 1: PRISMA chart displaying the diferent phases of the systematic literature review. 1,120 publications were retrieved from electronic databases and reference lists. 840 articles were removed because they were duplicates and reviews. 191 were excluded because of nonrelevance after the screening. Eight of 81 articles could not be recovered, and 48 were excluded because they failed to report the outcome of interest. Terefore, 33 studies were included in the review.
for anti-COVID-19 drug discovery. All other studies utilized molecular docking in addition to chemical-chemical and chemical-protein interactions using the STITCH database and randomization test using SWISSADME [16], molecular dynamics simulation, drug-likeness tests, and protein-protein interactions [20], molecular dynamics simulation [21,27], virtual screening and pharmacokinetic assessment [23], virtual screening by MCCS [31], quantum mechanics and molecular dynamic simulations [33], molecular similarity detection using Discovery Studio software, molecular fngerprint detection using Discovery Studio software, toxicity studies using Discovery Studio 4.0, and molecular dynamics (MD) simulations using the GROningen machine [34], virtual screening of MolPort database, molecular dynamics (MD) simulations, and drug-likeness predictions [35,39,40], and structure-based virtual screening (SBVS) of ASINEX antiviral library, drug-likeness and lead likeness annotations, pharmacokinetics analysis, and molecular dynamics (MD) simulations [36].
Other researchers preferred using SARS-CoV-2 M pro with other SARS-CoV-2 proteins or human proteases as their drug targets. For example, Ongtanasup et al. [22]; Xu et al. [25]; Shahabadi et al. [26]; and Wang et al. [28] used SARS-CoV-2 M pro and ACE2 as their drug targets. Ongtanasup et al. [22] undertook MD, MDS, drug-likeness, and ADMET prediction and found Myristica fragrans compounds as suitable drug candidates against COVID-19. Xu et al. [25] utilized the same in silico techniques in addition to virtual screening and identifed red wine, Chinese hawthorn, and blackberry as substances with anti-COVID-19 compounds. Shahabadi et al. [26] undertook only two in silico processes, MD and MDS to fnd cetilistat, abiraterone, di-iodo hydroxyquinoline, and bexarotene as anti-COVID-19 drug candidates. Among the four groups of scholars, Wang et al. [28] employed seven in silico models in their anti-COVID-19 drug discovery process. Te authors used virtual screening, molecular interaction networks using Cytoscape, protein-protein interaction (PPI) network construction, gene ontology enrichment analysis, KEGG pathway analysis, molecular docking, and molecular dynamics (MD) simulation. Tey found compounds created using the HuaShi XuanFei Formula (HSXFF) as probable anti-COVID-19 drugs.
Muhseen et al. [38] opted for MDS and structure-based virtual screening to obtain NPACT01552, NPACT01557, and NPACT00631 as probable inhibitors of the SARS-CoV-2 spike receptor-binding domain (RBD). Rao and Shetty [41] performed virtual screening, pharmacokinetic and pharmacodynamics properties examination, molecular docking, and MDS and discovered 12,28-oxa-8-hydroxymanzamine A as a potential inhibitor of nsp12. In the last study, Pandey et al. [43] utilized the same in silico models employed by several other researchers: molecular docking, MDS, and ADME analysis. Te authors found fsetin, quercetin, and kaempferol as lead compounds against COVID-19.

Drug Targets.
Several drug targets were identifed and validated using in silico approaches. In the fght against COVID-19, this study's fndings align with information in existing literature on the drug targets being SARS-CoV-2 proteins. Te most common is SARS-CoV-2 M pro , also referred to as 3-chymotrypsin-like proteases (3CLpro). It is a highly conserved cysteine hydrolase in β-coronaviruses with an essential function in viral replication. It is a key target for treating and preventing infectious diseases caused by coronavirus, including COVID-19 [46]. Other SARS-CoV-2 proteins utilized as drug targets include SARS-CoV-2 ribonucleic acid (RNA)-dependent RNA polymerase (RdRp), SARS-CoV-2 spike protein, nsp16, SARS-CoV-2 PL pro , and nsp12. SARS-CoV-2 RdRp is a viral enzyme responsible for viral RNA replication in host cells [47]. Zhu et al. [47] explain that SARS-CoV-2 RdRp has no host cell homologs. Terefore, its inhibitors can be created with improved potency and fewer of-target impacts on human host proteins and thus are more efective and safer therapeutics for treating COVID-19. SARS-CoV-2 RdRp has a catalytic subunit called nonstructural protein 12 (nsp12). Hillen et al. [48] outline that with the help of conserved residues, the active-site cleft of nsp12 attaches to the frst turn of RNA and regulates RdRp action. Terefore, nsp12 can also be used as a drug target because it mediates the SARS-CoV-2 RdRp function.
Among all human coronaviruses, this study's fndings conform with information in existing literature that the SARS-CoV-2 spike protein is highly conserved and takes part in the recognition of the receptor, attachment of the virus, and viral entry into host cells. Due to its vital role, it embodies one of the most signifcant targets for COVID-19 therapeutic and vaccine research [49]. Even though nsp16 and PL pro are drug targets, they are rarely used in anti-COVID-19 drug development. Between the two, the 2′-Omethyltransferase nonstructural protein 16 (Nsp16) is crucial for immunological evasion. Nsp16 does this by imitating CMTr1, its human ortholog, which methylates mRNA to improve translation efcacy and diferentiate itself from each other [50]. One of the two SARS-CoV-2 protease antivirals that could potentially target is a papain-like protease (PL pro ). Because it is crucial for viral polyproteins' cleavage and maturation, the construction of the replicase-transcriptase complex, and interference with host defenses, PL pro is also a desirable target [51]. Te last two drug targets are human proteases called TMPRSS2 and ACE2. SARS-CoV-2 requires the serine protease TMPRSS2 for S protein priming and the SARS-CoV receptor ACE2 for entry [52]. Terefore, TMPRSS2 and ACE2 inhibitors can restrict entry and be a therapy option.

Lead Identifcation Process.
Most studies in this review, as evident in existing literature as well, employed in silico modeling for ligand-based and structure-based design for probable anti-COVID-19 drug candidates using known targets already described. Te lead identifcation process encompassed all necessary anti-COVID-19 drug discovery processes. As shown in Figure 2, the frst in silico processes that most studies employ include pharmacophore modeling, chemical synthesis, database creation, homology modeling, and literature search. Pharmacophore modeling involves using several ligands to create a model with common pharmacophore features. Te model, popular as a pharmacophore, is a collection of electronic and steric characteristics that ascertain optimal supramolecular interactions during virtual screening on large-scale compound databases [53]. Before virtual screening of the databases, other researchers opt to perform homology modeling or chemical synthesis of their ligands of interest. Homology modeling involves predicting the 3D structure of a ligand using its amino acid sequence. At the same time, chemical synthesis refers to using one or more chemical reactions to convert a starting material into a desired ligand or compound. Since several antiviral compounds are well-known and their structures already elucidated, some scholars prefer performing literature searches and creating a database of such molecules that they use for further drug discovery processes.
Te next three in silico processes usually include virtual screening, molecular docking, and molecular dynamics simulation. Virtual screening is an in-silico technique used in drug development to fnd the structures most probable to bind to a therapeutic target, often a protein, enzyme, or receptor. Te selected hit molecules obtained by virtual screening are subject to molecular docking, which estimates the binding energy and interaction afnity involved in the interaction between a receptor and a ligand [2]. Te ligandreceptor complexes with the best binding afnity undergo molecular dynamics simulation, enhancing the comprehension of a system's dynamic performance. It measures the stability of a complex. Determining stable complexes is a step further toward developing a drug. However, druglikeness and ADMET properties analysis are essential to determine the desirability of a lead compound as an anti-COVID-19 drug. Evaluating the absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of lead compounds is one of the signifcant criteria before developing a compound into a drug because they shed some light on the molecules' solubility, GIT absorption, and bioavailability profles. Tese processes underscore the basic in silico models that several researchers employ. However, not all researchers adhere to the methodology described previously. Te procedure they adopt depends on numerous factors such as preference, objectives, databases used, and others. Combining two or more of these techniques must be employed during anti-COVID-19 drug discovery.

Conclusion
A practical approach for identifying possible anti-COVID-19 drug targets and probable lead compounds during the drug discovery phase is in silico modeling. Clarifying their mechanisms of action and possible medical usefulness is another beneft of using in silico models. An existing literature agrees that using in silico and other computation techniques to investigate possible medications is a secure, afordable, and efcient way to fnd, create, or repurpose potential remedies. Even though medical and nonclinical validation employing some in vivo and in vitro assays is still required to further confrm the antiviral activity of these possible candidate molecules, discovering those particular lead compounds using in silico means is a step in the right direction when drug design, development, and discovery are concerned. Tis study in unison with existing literature confrms that the in silico methods that use several drug targets have the best chance of succeeding because of the broad scope of potential lead candidates. Additionally, the in silico methods should be used concurrently to forecast ADMET and drug-like features of the candidate compounds.

Data Availability
Te data supporting the fndings of this study are available upon request from the corresponding author.

Conflicts of Interest
Te authors declare that they have no conficts of interest.