key: cord-0862795-tjy3nyt4
authors: Kumar, Abhinit; Loharch, Saurabh; Kumar, Sunil; Ringe, Rajesh P; Parkesh, Raman
title: Exploiting cheminformatic and machine learning to navigate the available chemical space of potential small molecule inhibitors of SARS-CoV-2
date: 2020-12-29
journal: Comput Struct Biotechnol J
DOI: 10.1016/j.csbj.2020.12.028
sha: 23093e00712b95bb93fd308018e1713a4d35bb59
doc_id: 862795
cord_uid: tjy3nyt4

The current life-threatening and tenacious pandemic eruption of coronavirus disease in 2019 (COVID-19) has posed a significant global hazard concerning high mortality rate, economic meltdown, and everyday life distress. The rapid spread of COVID-19 demands countermeasures to combat this deadly virus. Currently, there are no drugs approved by the FDA to treat COVID-19. Therefore, discovering small molecule therapeutics for treating COVID-19 infection is essential. So far, only a few small molecule inhibitors are reported for coronaviruses. There is a need to expand the small chemical space of coronaviruses inhibitors by adding potent and selective scaffolds with anti-COVID activity. In this context, the huge antiviral chemical space already available can be analysed using cheminformatic and machine learning to unearth new scaffolds. We created three specific datasets called “antiviral dataset” (N= 38,428) “drug-like “antiviral dataset” (N=20,963) and “anticorona dataset” (N= 433) for this purpose. We analyzed the 433 molecules of “anticorona dataset” for their scaffold diversity, physicochemical distributions, principal component analysis, activity cliffs, R-group decomposition, and scaffold mapping. The scaffold diversity of the “anticorona dataset” in terms of Murcko scaffold analysis demonstrates a thorough representation of diverse chemical scaffolds. However, physicochemical descriptor analysis and principal component analysis demonstrated negligible drug-like features for the “anticorona dataset” molecules. The “antiviral dataset” and “drug-like antiviral dataset” showed low scaffold diversity as measured by the Gini coefficient. The hierarchical clustering of the “antiviral dataset” against the “anticorona dataset” demonstrated little molecular similarity. We generated a library of frequent fragments and polypharmacological ligands targeting various essential viral proteins such as main protease, helicase, papain-like protease, and replicase polyprotein 1ab. Further structural and chemical features of the “anticorona dataset” were compared with SARS-CoV-2 repurposed drugs, FDA-approved drugs, natural products, and drugs currently in clinical trials. Using machine learning tool DCA (DMax Chemistry Assistant, we converted the “anticorona dataset” into an elegant hypothesis with significant functional biological relevance. Machine learning analysis uncovered that FDA approved drugs, Tizanidine HCl, Cefazolin, Raltegravir, Azilsartan, Acalabrutinib, Luliconazole, Sitagliptin, Meloxicam (Mobic), Succinyl sulfathiazole, Fluconazole, and Pranlukast could be repurposed as effective drugs for COVID-19. Fragment-based scaffold analysis and R-group decomposition uncovered pyrrolidine and the indole molecular scaffolds as the potent fragments for designing and synthesizing the novel drug-like molecules for targeting SARS-CoV-2. This comprehensive and systematic assessment of small-molecule viral therapeutics' entire chemical space realised critical insights to potentially privileged scaffolds that could aid in enrichment and rapid discovery of efficacious antiviral drugs for COVID-19.

, Scaffold hunter (20), MACCS (21), and ECFP6 (22) and cheminformatics tools integrated with the 130 KNIME analytical platform (23). Instant JChem was used for structure database management, search, and 131 prediction, Instant JChem 19.21.5, 2020, ChemAxon. Mona (24) was used for curation and compound 132 library preparation. 133 134 2.2 Scaffold analysis. We generated the scaffolds from "anticorona dataset" (N= 433) using the various 135 filters such as atom count, finger print, molecular weight, and pIC 50 to explore fragments, molecular 136 scaffolds, virtual scaffold and their relationship using Scaffold Hunter (20). Scaffold Hunter first reads 137 the scaffold data from an SQL database and automatically constructs and displays the scaffolds as a tree 138 using various properties like atom count, fingerprints, etc. The scaffold tree shows the relationship 139 between the parent and child molecular scaffolds. The chemical scaffolds -both parent and child and other 140 intermediate virtual scaffolds derived from the parent scaffolds are stored in the database. They can be 141 retrieved as scalable vector graphics (SVG) images for further analysis. The fragment scaffold and the 142 virtual scaffolds derived from the fragments were manually retrieved from the scaffold tree analysis. 143 We generated Murcko scaffolds by excluding the exocyclic double bonds and the α attached atom (25) of 144 both the "drug-like antiviral dataset" and "anticorona dataset". The Murcko scaffold was further used to 145 create the skeleton scaffold. The skeleton analysis includes only the ring and replaced the heteroatoms by 146 a carbon atom. We also analyzed structures in "drug-like antiviral dataset" and "anticorona dataset" using 147 the scaffold representation proposed by Bemis and Murcko (26) . In this method, the molecule is dissected 148 into ring systems, linkers, side-chain atoms, and the framework. 149 150 2.3 Activity cliff analysis. The structure-activity landscape index (SALI) calculated by the activity cliff 151 analysis supplies a measure between activity (pIC 50 ) and chemical diversity (1-similarity) for each 152 compound (27). The analysis was carried out using the Skeleton Sphere descriptor, as given by equation 153 1. 154 2) were used to 164 calculate the ECFP6 fingerprints for "antiviral dataset" and "anticorona dataset". The fingerprints were 165 compared between these two datasets. The edges and nodes were generated for the related compounds 166 and were further used to represent these compounds' chemical networks. We used Gephi 0.9.2 for 167 visualization usingvarious algorithmic configurations, for example, Force Atlas, Fruchterman Reingold, 168 Open Ord, Contraction, Force Atlas2, and Yifan, and Yifan Hu Proportional (28). Binning clustering of 169 the compounds in the database was done for "anticorona dataset" using the ChemMine web server (29), 170 as the experimental value for this dataset is known.

171 2.6 QSAR modeling by machine learning. We applied DCA (DMax Chemistry Assistant) software (30) 172 to derive the hypotheses and determine the relationship within the morphological and structural features 173 of the anti-COVID ligands and their bioactivities. DCA is an ILP (inductive logic programming)-based 221 suggest diverse chemical representation. It will still require efforts to populate this library with more novel 222 and unique scaffolds to increase diversity further. In contrast, scaffold diversity analysis of "drug-like 223 antiviral dataset" (Murcko scaffolds (0.23), singleton scaffold (0.16), and skeleton scaffold (0.14)) 224 suggests low chemical diversity, as confirmed by Gini coefficient. Interestingly, we were able to identify 225 some promising scaffolds from the "anticorona dataset" with favorable characteristics for designing novel 226 derivatives by SAR studies ( Figure S1 ). For instance, the scaffolds of N-(2-oxo-2-((2-(2-oxopyrrolidin-227 3-yl)ethyl)amino)ethyl)-1H-indole-2-carboxamide and benzyl (2-oxo-2-((2-oxo-2-((2-(2-oxopyrrolidin-228 3-yl)ethyl)amino)ethyl)amino)ethyl)carbamate showed highest biological activity ( Figure 1A ). 229 230 Table1 Colors depict the Murcko frequency, ranging from blue (lower) to red (higher) 243 244 A scaffold tree of the compounds in the "anticorona dataset" was generated to visualize the standard core 245 structure or scaffold. We first isolated the most active chemical scaffold from the database, then generated 246 the entire possible parent scaffold, followed by selecting one parent-child pair. This process continued 247 until all of the possible successive parent-child scaffold pairs of the "anticorona dataset" were exhausted. 248 The scaffolds were achieved by cutting all of the side chains but keeping the double bonds connected 249 directly to a ring (39). All 433 molecules of the "anticorona dataset" were pruned until a single ring was 250 attained. We identified oxopyrrolidine, indoline, cyclopropylbenzene, thiophene, indole, dioxole, 251 cyclobutylbenzene, azaspiro, pyranone, and phenylsulfane as some of the most frequent fragment by this 252 scaffold analysis (Figure 2 ). Fragment-based analysis of the "anticorona dataset" inhibitors revealed that 253 spiro compounds represent an interesting scaffold-point to develop potent coronaviruses inhibitors. 254 However, so far, only a few spiro compounds had been explored to target coronaviruses (Figure 2 , 255 azaspiro). Further, spiro compounds have inherent three-dimensionality and structural diversity (40). 256 Therefore, it will be promising to include novel spiro scaffolds for targeting coronaviruses, incredibly 257 challenging to treat SARS-CoV-2 infection. 258 259 260 261 Figure 2 . Representative examples of frequent fragments identified from the "anticorona dataset" using 262 scaffold hopping. 263

264 We then looked for common single ring scaffolds, that are common to molecules representing 265 coronaviruses targets like main protease, papain-like protease, replicase polyprotein 1ab, and helicase. 266 The results depicted in Figure 3 shows that oxopyrrolidine is the standard basic structure core, present in 267 all these molecules. By scaffold hopping, we were able to construct the promising virtual scaffolds, which 268 can be used as polypharmacological ligands for targeting these proteins. These fragments are the ideal 269 starting point for the fragment-based drug discovery for targeting essential SARS-CoV-2 proteins such as 270 main protease, replicase polyprotein 1b, helicase, etc. The fragments and virtual scaffolds identified in 271 the present study (Figure 3) 370 The Tanimoto coefficient is a chemical fingerprint or feature based similarity metrics to measure the 371 chemical similarity between pairs of the molecules (29). In this study, the similarity is measured between 372 the reference and the database structure. The bin clustering partition grouped the molecules into various 373 similarity groups (Table S4) . For main protease inhibitors, the largest bin cluster (n = 275), represented 374 by oxopyrrolidine scaffold, was observed ( Figure 6 ). Additionally, singleton bin clusters of disulfuram, 375 ebselen, and shikonin scaffolds were also noticed ( Figure 6 ). For papain-like protease, 12 bin clusters 376 were observed, where naphthalene-based scaffold constitutes the largest bin cluster. Furthermore, a 377 singleton bin cluster comprising nitrophenyl-piperazine, nitropyridine-amine, and ethyl (phenyl) 378 carbamo-dithioate scaffold was also observed. For helicase, four bin clusters were observed, with the 379 chromenone scaffold representing the largest bin cluster. Additionally, singleton bin clusters, including 380 triazole and benzothiazole scaffold, were detected. For replicase polyprotein 1ab, four bin clusters were 381 observed with an oxopyrrolidine scaffold representing the largest bin cluster. The phenanthro-furan 495 3.7 Structure-activity relationship and activity cliff analysis. Activity cliffs can be evaluated by 496 investigating the biological landscape using similarity metrics that work under the evidence that 497 structurally similar compounds are inclined to have a similar biological response. We generated affinity 498 scatters plot (pIC 50 value) for each molecule of the "anticorona dataset" (Figure 10 A) . This similarity and 499 activity analysis represent all 957 pairwise comparisons between the 433 compounds of the "anticorona 500 dataset, identified using the cut-off similarity threshold of 86 %. The vast amount of "activity cliff" pairs 501 provides vital information on QSAR models and is also useful for designing virtual molecules libraries to 502 be employed for COVID-19 screening. 503 504 505 506 Figure 10 . (A) SALI plot of compound pairs generated from the "anticorona dataset". X-and Y-axis 507 represents activity values; color indicates the delta activity; higher and lower values are indicated by red 508 and blue, respectively. The size of the scatters is indicated by the SALI value. (B) The activity cliff set 509 was grouped based on neighborhood similarity relationships. Colors indicate pIC 50 value with the higher 510 value represented by red color and lower value represented by blue high. The scatters size suggests a max 511 of SALI pIC 50 value/SkeleSpheres. (C) An example of a compound pair from the "anticorona dataset" 512 (ID: 205 and ID: 206) represents the 'activity cliff'. The red circle shows the structural variation in the 513 'activity cliff' pair.

514 For the paired comparison between pair ID 205 and ID 206 ( Figure 10C ), the determined SALI value is 515 69.707, the similarity is 0.959, and the activity values are 8.9 and 6.1, respectively. A higher SALI value 516 indicates a significant difference between the biological activities of two structurally similar compounds 517 (Table S 5 ). Figure 10B represents groups based on neighbouring similarity and their respective SALI 518 values. A representative example of activity cliff in phenyl cinnamamide derivatives is shown in Figure  519 11. These results confirm that small changes in structures can induce significant potency variations. The 520 design of new focused libraries for targeting SARS-CoV-2 is possible by incorporating potent moieties 521 identified by "activity cliff" analysis as displayed in Figure 11 . 522 523 525 526 527 528 Figure 11 . "Activity cliff" analysis of phenyl cinnamamaide derivatives. The value of pIC 50 is color-529 coded with green color denoting lower activity and blue color denoting the highest activity. The structural 530 variation is highlighted in the red. 

Such a rule can form the basis for splitting 175 the molecules into two alternative arguments

The background knowledge in DCA is defined by electrostatic, type of 178 elements (e.g., carbon, sulfur, nitrogen), functional group and rings, and linkage (fused, linked, the 179 positional topology of the ring, how different functional groups and rings are connected) between the 180 substructures of the chemical molecule. DCA can relate this background knowledge of the chemical 181 molecules to correlate with their experimental biological activities, such as inhibition or activation

Data visualization. R studio was used to process, analyze, and visualize the plots. The boxplots were 185 generated using the boxplot package of the R language. To generate the 3D-PCA scatter plot

We analyzed the scaffolds with 192 reported experimental biological activity (pIC 50 value). Murcko scaffold analysis revealed 227 and 4779 193 unique scaffolds from "anticorona dataset" and "drug-like antiviral dataset" respectively, with varying 194 degrees of frequencies

Singleton scaffold frequency was observed to be 152 in the "anticorona dataset" 198 and 3031 in the "drug-like antivirial dataset". Subsequent Murcko Skeleton analysis resulted in further 199 identification of 143 and 3034 skeleton scaffolds in the

The Gini coefficient's value varies between 0 to 1, where 0 implies an equal 206 distribution of income, whereas 1 implies complete inequality, that is, few wealthy individuals represent 207 the major percentage of the total income of the population. In drug discovery, the Gini coefficient has 208 been used to evaluate the diversity of compounds from a sizeable dataset (37)

We could not 217 calculate the Gini coefficient for the "anticorona dataset" as the dataset size (N = 433) is not sufficient for 218 statistical analysis. The Murcko scaffold diversity calculated in terms of the scaffold ratio and total 219 molecules (Ns/M) (38) was computed for both datasets (Table 1). The scaffold diversity analysis of the 220 "anticorona dataset"(Murcko scaffold (0.52), singleton scaffold (0.35), and skeleton scaffold (0.33) 304 shaded area represents the parameters off-limit to the rules for favorable oral bioavailability (49). The 305 analysis revealed that the almost all inhibitors of replicase polyprotein 1ab exhibit MW value 'greater 306 than 500', nRotB count 'greater than 10', PSA 'more significant than 120 Å' and HBA count 'more than 307 10' (Figure 4A-C, 4E) suggesting negligible oral bioavailability. Similarly, the majority of main protease 308 inhibitors display MW value

Whereas, more than ~50% inhibitors of helicase have PSA 'greater than 120 Å'. Few 310 outliers were observed in the case of the HBA and HBD distribution for the main protease inhibitors 311 (Figure 4E and 4F). In general, it can be concluded that most coronavirus inhibitors so far unearthed lack 312 the general criteria for oral bioavailability and, thus

The second PC is primarily contributed by cLogP and secondarily 326 by nRotB and MW. These results indicate that the property space is governed by different molecular 327 descriptors and differs substantially among "anticorona dataset". Figure 5 shows a three-dimensional 328 illustration of the property space scatterplot of 'anticorona dataset" (yellow spheres), 'repurposed drugs-329 COVID-19 pipeline' (red spheres), 'natural products and drugs' (blue spheres), 'repurposed drugs-330 COVID-19-clinical trials' (black spheres) and 'FDA-approved drugs' (green spheres). As observed, many 331 of the 'anticorona dataset (yellow spheres) expand along both the PC1 and PC3 axes, indicating that they 332 vary significantly from the 'approved drugs' (green spheres). Some of 'natural products and drugs' expand 333 majorly along the PC1 axis and PC3 axis, implying that they differ marginally in their property space. 382 scaffold represented singleton bin clusters. The clusters identified by this analysis can be further explored 383 to design scaffold derivatives by using medicinal chemistry and SAR information. For example, for 384 targeting main protease of SARS-Cov-2, it will be good idea to design and synthesize molecules, based 385 on the disulfuram, ebselen and shikonin based molecular scaffolds, as the population of these molecules 386 is underrepresented in the already know inhibitors of coronaviruses. On the other hand, many different 387 derivatives of oxopyrrolidine are well known for targeting

and 138) of the "anticorona dataset" show high inhibitory activity against coronavirus main protease and 412 replicase polyprotein 1ab' proteins. Similarly, AV18965 from the "antiviral dataset" exhibits similarity to 413 "anticorona dataset" compounds 248 and 350, which are active against main protease and spike 414 glycoprotein proteins. Further, we also generated chemical networks to investigate the similarity of 415 inhibitors, individually for each protein target such as

However, some of the prominent examples are AV43733, AV10045, AV9951, AV14839, and AV9998 422 for Helicase

We also identified many other known antiviral compounds that 424 showed significant similarity to "anticorona dataset" compounds but exhibited different chemical features 425 (Supplementary Table S2). Collectively, these antiviral compounds identified from the "antiviral dataset" 426 hint towards their potential use against COVID. The similarity of multiple inhibitors from different protein 427 targets to a single antiviral compound inspires the multi

DCA is an ILP 441 (Inductive Logic Programming) based approach that uses the existing knowledge such as electron flow, 442 element, moiety, and substructure relationship, to generate the hypotheses that best corroborates with the 443 given data. It starts by reading the functional groups and rings and then constructs the hypotheses to 444 determine the building blocks and their relation to each other. To deduce a significant outcome, we needed 445 a reasonable number of inhibitors with varying high and low activity measurements. Thus, we selected 446 the coronaviruses main protease inhibitors for this investigation. We were successful in generating the 447 inhibitor model hypothesis. The hypothesis suggested that inhibitors with specific patterns containing a 448 benzene ring and a five-membered hetero-aromatic rings

The results 459 showed an aromatic ring and a five-membered hetero-aromatic ring, in agreement with the generated 460 hypothesis model for "anticorona dataset". The selected examples of FDA approved drugs that showed 461 high predicted pIC 50 values are displayed in Figure 9

MK-0518), Azilsartan (TAK-536), Acalabrutinib (ACP-196)

Similarly, Sitagliptin, which is used 467 traditionally to treat diabetes, has been proposed by another group to reduce the severity of COVID-19 468 patients (58). The findings of this analysis are quite intriguing and can be widely applied to repurpose the 469 existing drugs. Considering the hour of need, our machine learning results, together with established 470 literature, recommends the immediate need for a detailed study on these antiviral drugs to understand their 471 mechanisms about protein targets of SARS-CoV-2

Scatter plot of Core fragments vs. pIC 50 generated from "anticorona dataset". The value of the 536 pIC 50 is color coded, with red and blue colors showing higher and lower pIC 50 values, respectively

The 542 analysis revealed that compounds with the same core fragment showed different activity (Figure 12), 543 implying that different R-groups influence biological activity differentially. The analysis revealed that 544 pyrrolidine and the indole core fragment yielded the highest biological activity in the dataset

The results from the SAR study are expected to help design different derivatives with the desired activity 549 against COVID-19. These core fragments are structurally similar to the fragments generated based on 550 scaffold-hoping, as shown in Figure 2

Although researchers are working diligently to find a 558 cure or vaccine for this deadly virus, no successes have been found. Furthermore, because vaccine 559 development might take a long time to enter the market, finding a drug or inhibitor is optimistic and can 560 impede the further spread of the virus. Keeping this in mind

We have determined common fragments and generated promising virtual scaffolds which 565 have not been described before and can be further explored for targeting SARS-CoV-2. We also identified 566 oxopyrrolidine based scaffolds that can be used as polypharmacological ligands (63)

show the highest biological 571 activity and may be used as a framework to design novel SARS-CoV-2 inhibitors. Indole and pyrrolidine 572 based scaffolds have been extensively used in drug discovery and have resulted in the development of 573 many approved drugs. Additionally indole scaffold is widely used in the design and synthesis of the 574 antiviral inhibitors

BILB 577 1941, BMS-791325, MK-8742 and Enfuvirtide. Commercial availability of indole and pyrroldine based 578 building blocks and their significant interactions within the active site of the Mpro protein suggest (61) 579 these pyrrolidine-based derivatives as promising candidates for further investigation. Experimentally 580 validation of our results will be really useful and we are currently synthesizing the molecules

so it is 584 complicated and challenging for an anti-COVID drug to work against the virus. Therefore, in our study, 585 we focused on essential targets of the COVID virus and conducted machine learning and cheminformatics 586 based research to predict the specific scaffold responsible for inhibiting a target. Machine learning resulted 587 in the hypothesis generation of structural patterns and identification of FDA-approved drugs

In conclusion, this analysis provides the groundwork for designing diverse chemical libraries, fragment 591 libraries, virtual scaffolds for shape and ligand-based screening, and identification of essential FDA drugs

It thus will be useful for the discovery of small molecule therapeutics for COVID-19

Coronavirus disease; SARS-CoV-2, Severe acute respiratory 606 syndrome coronavirus-2

610 Clinical features of patients infected with 2019 novel coronavirus in Wuhan

The spike protein of SARS-CoV-a target for 613 vaccine and therapeutic development

Isolation of a novel 615 coronavirus from a man with pneumonia in Saudi Arabia

Niu, 618 P. A novel coronavirus from patients with pneumonia in China

A pneumonia outbreak associated with a new coronavirus of probable bat origin

Epidemiology, genetic 624 recombination, and pathogenesis of coronaviruses

Remdesivir, Sofosbuvir, Galidesivir, and Tenofovir against SARS-CoV-2 RNA 626 dependent RNA polymerase (RdRp): A molecular docking study

The Modern Drug Discovery Process

Journal of chemical information and 633 modeling

Current Prevention of COVID-19: Natural 635 Products and Herbal Medicine

Drug repurposing 637 approach to fight COVID-19

Clinical trials on drug repositioning for COVID-19 treatment

Instant JChem was used for structure database management, search, and prediction

The Chemistry 645 Development Kit (CDK): An open-source Java library for chemo-and bioinformatics

Rdkit: A software suite for cheminformatics, computational chemistry, and predictive 648 modeling

alvaDesc: A Tool to Calculate and Analyze Molecular Descriptors and Fingerprints

DataWarrior: an open-source program for chemistry 652 aware data visualization and analysis

656 21. Schuffenhauer, A., Varin, T. Rule-based classification of chemical structures by scaffold. Molecular 657 informatics

Journal of chemical 659 information and computer sciences

Drug discovery applications for KNIME: 661 an open source data mining platform. Current topics in medicinal chemistry

MONA 2: a light cheminformatics platform for interactive compound library 663 processing

Structural 665 analysis of sortase A inhibitors

The properties of known drugs. 1. Molecular frameworks

Structure−activity landscape index: identifying and quantifying activity 669 cliffs

Gephi: an open-source software for exploring and 671 manipulating networks

ChemMine tools: an online service for analyzing and clustering 673 small molecules

Discovering H-bonding rules in crystals with inductive logic programming

Epigenetic drug discovery: systematic assessment of chemical space. Future 678 medicinal chemistry

Methods of measuring the concentration of wealth. Publications of the American 680 statistical association

Inequality and visibility of wealth in experimental 685 social networks

A Gini coefficient based evaluation on the reliability of travel time 688 forecasting

A novel application of Gini 691 coefficient for the quantitative measurement of bacterial aggregation

Using the Gini coefficient to measure the chemical diversity of 693 small-molecule libraries

Scaffold diversity of exemplified medicinal chemistry space

Structural diversity of organic chemistry. A scaffold analysis of the CAS Registry. The Journal of organic 698 chemistry

The use of spirocyclic scaffolds in drug discovery. Bioorganic & 700 medicinal chemistry letters

To be drug or prodrug: 702 structure-property exploratory approach regarding oral bioavailability

Integrated cheminformatics 705 approaches toward epigenetic drug discovery

Molecular properties 708 that influence the oral bioavailability of drug candidates

The severe acute respiratory syndrome (SARS) coronavirus NTPase/helicase belongs to a distinct 711 class of 5′ to 3′ viral helicases

Identification and characterization 713 of severe acute respiratory syndrome coronavirus replicase proteins

Crystal structure of SARS-CoV-2 main protease provides a basis for design of improved 717 α-ketoamide inhibitors

719 Structure of M pro from SARS-CoV-2 and discovery of its inhibitors

The papain-like protease 721 of severe acute respiratory syndrome coronavirus has deubiquitinating activity

Drug-like properties and the causes of poor solubility and poor permeability

Molecular de-novo design through deep 726 reinforcement learning

Extended-connectivity fingerprints

Gephi: an open source software for exploring and manipulating 730 networks

Polypharmacology in HIV inhibition: can a drug with simultaneous 732 action against two relevant targets be an alternative to combination therapy?

Exploring Polypharmacology in Drug Discovery and Repurposing Using 736 the CANDO Platform

Screening of an FDA-approved 739 compound library identifies four small-molecule inhibitors of Middle East respiratory syndrome 740 coronavirus replication in cell culture

Raltegravir in HIV-1 infection: safety and efficacy in treatment-naive patients. Clinical 742 medicine reviews in therapeutics

Targeting SARS-CoV-2: a systematic drug repurposing approach to identify promising 745 inhibitors against 3C-like proteinase and 2′-O-ribose methyltransferase

Inflammation: A bridge 748 between diabetes and COVID-19, and possible management with sitagliptin

DataWarrior: An evaluation of the open-source 751 drug discovery tool. Expert opinion on drug discovery

Identification of 754 potential Mpro inhibitors for the treatment of COVID-19 by using systematic virtual screening 755 approach

Structure and inhibition of the SARS-CoV-2 main protease reveal strategy for 759 developing dual inhibitors against Mpro and cathepsin L

Recent Review on Indole: A 762 Privileged Structure Scaffold

Polypharmacology: drug discovery for the future

Emergence of genomic diversity and recurrent mutations in SARS