key: cord-0710678-o0f2vde2
authors: Abbasi, Maryam; Sadeghi-aliabadi, Hojjat
title: An In-silico Screening Strategy to the Prediction of New Inhibitors of COVID-19 M(pro) Protein
date: 2021
journal: Iran J Pharm Res
DOI: 10.22037/ijpr.2021.114997.15146
sha: 1b33a2eccec4bbbb79b8c7808921001172430f75
doc_id: 710678
cord_uid: o0f2vde2

The coronavirus disease-2019 (COVID-19) was first recognized in Wuhan, China, and quickly spread worldwide. Between all proposed research guidelines, inhibition of the main protease (M(pro)) protein of the virus will be one of the main strategies for COVID-19 treatment. The present work was aimed to perform a computational study on FDA-approved drugs, similar to piperine scaffold, to find possible M(pro) inhibitors. Firstly, virtual screening studies were performed on a library of FDA-approved drugs (43 medicinal compounds, similar to piperine scaffold). Among imported 43 drugs to virtual screening, 34 compounds were extracted. Four top-ranked drugs in terms of the highest interactions and the lowest binding energy were selected for the IFD study. Among these selections, lasofoxifene showed the lowest IFD score (-691.743 kcal mol(-1)). The stability of lasofoxifene in the COVID-19 M(pro) protein active site was confirmed with 100 ns MD simulation. Lasofoxifene binding free energy was obtained -107.09 and -173.97 kcal mol(-1), using Prime MM-GBSA and g_mmpbsa methods, respectively. The identified lasofoxifene by the presented computational approaches could be a suitable lead for inhibiting M(pro) protein and COVID-19 treatment.

Coronavirus disease 2019 (COVID- 19) is an emerging developed human infectious from coronavirus family, outbreak in Wuhan, China, and has been outspreading quickly in China and other countries since December 2019 (1) . The coronavirus family comprises two high pathogenic forms; SARS-CoV (Severe Acute Respiratory Syndrome in 2002) and MERS-CoV (Middle East Respiratory Syndrome in 2012), but the COVID-19 has a significant difference from them due to the mutation processes (2) . According to WHO reports, COVID-19 has infected more than 88 million and close to 2 million deaths worldwide (10 January 2021).

The main awful manifestations of COVID-19 are fever, cough, severe acute respiratory syndrome, and the lack of therapeutic protocols (3) .

The coronavirus family includes the single-strand, positive-sense RNA genomes having 6-12 open reading frames (ORFs). The first and largest ORF contains genetic codes for two polyproteins termed ppla and pplab, which are autoproteolytically cleaved into 15 or 16 nonstructural proteins (nsp) (4) . Among these nsps, a chymotrypsin-fold proteinase, the main protease (M pro ), plays a prominent role in viral gene expression and replication (5) .

The coronavirus M pro protein is exceptionally conserved among the coronavirus family. M pro dimer consists of 3 structural domains; domain I (residues 8-101), domain II (residues 102-184) have an antiparallel β-barrels, and domain III (residues 201-303) that is linked to domain II using a long loop region (residues 185-200). COVID-19 M pro has a cysteinehistidine catalytic dyad, and its binding site is placed in a gap between domains I and II. Unlike other viral proteases, M pro utilizes cysteine amino acid instead of serine for nucleophilic attacks in the active site (6) ; thus, cysteine-histidine catalytic dyad inhibition can stop virus function.

Due to increasing concerns over the fast spread of COVID-19 globally, the rapid identification of drug candidates is essential. Hence, drug repurposing seems to be very impressive to create potent drugs to battle coronavirus in a short time (5) . In addition, high throughput screening of all approved drugs on M pro protein is a costly and timeconsuming process; thus, high throughput virtual screening method could be an excellent alternative to save time and money.

Recently, some dietary molecules from edible herbs and vegetables such as piperine, apigenin, curcumin, quercetin, and genistein, previously known as anti-viral, were evaluated as anti-COVID-19 agents (7) . Their binding energy values in the COVID-19 M pro protein active site were calculated using the molecular docking method. Among those, chromene scaffold demonstrated the lowest binding energy in EGCG (-6.99 kcal/mol), myricetin (-5 .38 kcal/mol), and quercetin (-5.29 kcal/ mol) compounds. Curcumin was another compound that showed low binding energy in COVID-19 M pro active site (-6.04 kcal/mol). Piperine was another compound with a -5.16 kcal/mol binding energy in the M pro protein active site. The virtual studies based on quercetin and curcumin scaffolds have been performed with COVID-19 M pro protein (8, 9) . Some FDA-approved drugs similar to the piperine scaffold ( Figure 1 ) were extracted as a drug library in this study. Then virtual screening was performed on the drug library to find potential drugs as COVID-19 M pro inhibitor. Induced fit docking and MM-GBSA calculation were performed to consider ligand and receptor flexibility and computing the binding free energy. Also, a molecular dynamic simulation was employed to confirm the stability of the best-chosen compound as anti-COVID-19 in the dynamic environment.

The FDA-approved drugs (43 medicinal compounds) with at least 40% similarity with piperine scaffold were extracted from Drug Bank (www. drug bank. ac). The 3D structure of all compounds was drawn in Marvin Sketch v5.7, ChemAxon (10) , and kept in a PDB format.

Forty-three drugs were subjected to molecular docking studies. The molecular docking process was steered using the AutoDock4.2 software package (11) . The PDB file of M pro protein was obtained from the Protein Data Bank (PDB ID 6lu7) (6) . Protein preparation was started with the elimination of all water molecules, ligands, and ions. After affixing polar hydrogens, the partial atomic charge was computed by the Kollman method, Figure 1 . The chemical structure of piperine. and the pdbqt file was kept (10) . The pdbqt file of ligands was created by calculating the Gasteiger-Marsili charge (12) . A grid box with the dimension of 50 × 50 × 50 Å (x, y, and z) was expanded on the protein binding pocket with a 0.375 nm spacing for each dimension; then, grid maps were made by Autogrid 4.2.

All docking parameters were unchanged except for the number of Lamarckian job, the initial population, and the maximum number of energy evaluations, respectively 50, 150, and 25 × 10 5 . Finally, the docking process was performed by AutoDock 4.2. Docking results were analyzed in terms of conformation and binding energy in the active site of the M pro protein. The best drugs with the lowest binding energy and the highest interaction in the binding pocket were chosen. The drugprotein interactions were imagined using AutoDockTools 1.5.6 and PyMOL Tcl (13) .

Protein structure preparation The atomic coordinates of M pro protein were obtained from the RCSB protein data bank (PDB ID 6lu7) and were prepared using Schrodinger's protein preparation wizard module (14, 15) . The M pro protein was provided by removing crystallographic water molecules, adding all hydrogen atoms to protein, allocating bond orders, generating disulfide. To process the maximum degree of hydrogen bond interactions, 180° spins of the terminal angle of Asn and Gln amino acids were allocated. Finally, a restrained minimization was carried out using optimized potentials for liquid simulations (OPLS-2005) force field to optimize the geometry and minimize the energy of the M pro protein (16) . After energy conversion, the minimization was done, and the RMSD score got a maximum cut-off of 0.420 Å.

The 3D structure of compounds was optimized using the LigPrep part of the Schrodinger suite 2015 (17) . The most probable ionization state was generated at the cellular pH value of 7.4 using the Epik tool (18) . Finally, the compound with the lowest energy conformer was optimized using the OPLS-2005 force field (16) .

Induced fit docking (IFD) is a powerful and precise method to compute ligand and receptor flexibility (19) . To understand the proteinligand interactions of the docked compound, we performed the induced fit docking in the Schrodinger software suite (20) . The IFD protocol was carried out in three successive steps. Firstly, optimized ligand and rigid protein were introduced to the IFD module. The docking box was specified to include all residues within the radius 30 Å from the ligand center. A van der Waals scaling of 0.7 and 0.5 was utilized for non-polar atoms of the protein and ligand, respectively. The Glide SP mode did the initial docking, and 20 ligand postures were maintained for protein structural modifications. Secondly, each of the ligand poses was subjected to side-chain and backbone refinement. All residues with at least one atom within 5 Å of each corresponding ligand pose were inserted into a conformational search and minimization using the Prime program (15) . The refined complexes were arranged by prime energy. Finally, the new receptorligand conformations within 30 kcal/mol of minimum energy were introduced to Glide redocking. The more desirable the binding affinity identified with, the more negative the Glide Score. The visualization of the best pose of protein-ligand complexes was done using PyMOL Tcl.

The best obtained ligand-M pro complex from IFD was further investigated with MM-GBSA procedure in the prime segment of Schrödinger suite 2015 (15) . MM-GBSA technique is introduced as a valuable and effective method to computing the binding free energy (ΔG bind ) between ligand and receptor with more precision. In this method, molecular mechanical (MM) energies are considered with a continuum solvent generalized Born (GB) model for polar solvation and a solventaccessible surface area (SASA) for non-polar solvation (21) . The binding free energy of the docked ligands is calculated according to the following equations (22, 23) :

Induced fit docking (IFD) is a powerful and precise method to compute liga flexibility (19) . To understand the protein-ligand interactions of the docked performed the induced fit docking in the Schrodinger software suite (20) . Th .

.

6 between ligand and receptor with more precision. In this method, molecular mechanical (MM) energies are considered with a continuum solvent generalized Born (GB) model for polar solvation and a solvent-accessible surface area (SASA) for non-polar solvation (21) . The binding free energy of the docked ligands is calculated according to the following equations (22, 23) :

. Equation 2 .

.

.

introduced as a valuable and effective method to computing the binding free energy (ΔGbind)

between ligand and receptor with more precision. In this method, molecular mechanical (MM) energies are considered with a continuum solvent generalized Born (GB) model for polar solvation and a solvent-accessible surface area (SASA) for non-polar solvation (21) . The binding free energy of the docked ligands is calculated according to the following equations .

.

introduced as a valuable and effective method to computing the binding free energy (ΔGbind)

between ligand and receptor with more precision. In this method, molecular mechanical (MM) energies are considered with a continuum solvent generalized Born (GB) model for polar solvation and a solvent-accessible surface area (SASA) for non-polar solvation (21) . The binding free energy of the docked ligands is calculated according to the following equations (22, 23) :

. Equation 2 .

.

.

Where ∆E MM is the difference of the gas phase of MM energy between the M pro -ligand complex and the aggregate of the energies of the free ligand and M pro protein. In the end, the outcomes were ordered based on the calculated ∆G binding Values.

The finest ordered candidate from docking outcomes was further investigated for appraising their thermodynamic properties and the binding stability in the M pro binding pocket by molecular dynamic (MD) simulation studies. The molecular dynamic simulation was carried out by the GROMACS-2019.3 package (24, 25). The Amber99.sb force field was employed in MD simulations (26) , and all drug topology parameters were ready by the AnteChamber Python Parser InterfacE (ACPYPE) (27) . The pKa values were calculated to determine which residue adopts non-standard ionization states, using PROPKA 3.1 webserver (28) . A TIP3P

water model was chosen (29) , and the drug-protein complex was waterlogged in a dodecahedron box. To neutralize the system, Na + ions were substituted with solvent water molecules. The complex energy minimization was carried out, and two stages of the process started MD simulation: 1) 500 ps simulation in the NVT ensemble (constant number of particles, volume, and temperature); 2) 1 ns simulation in the NPT ensemble (constant number of particles, pressure, and temperature). In the end, the MD simulation was run at 300 K temperature for 100 ns. The Particle Mesh Ewald (PME) method and the linear constraint (LINCS) algorithm were employed to calculate long-range electrostatic interactions and covalent bond constraints, respectively. Moreover, structure imagining was done by VMD 1.8.6 (30) and PyMOL Tcl.

Virtual screening study on FDA approved drugs (24, 25) . The Amber99.sb force field was employed in (26) , and all drug topology parameters were ready by the AnteChamber erfacE (ACPYPE) (27) . The pKa values were calculated to determine which n-standard ionization states, using PROPKA 3.1 webserver (28) . A TIP3P as chosen (29) , and the drug-protein complex was waterlogged in a

x. To neutralize the system, Na + ions were substituted with solvent water omplex energy minimization was carried out, and two stages of the process lation: 1) 500 ps simulation in the NVT ensemble (constant number of , and temperature); 2) 1 ns simulation in the NPT ensemble (constant number sure, and temperature). In the end, the MD simulation was run at 300 K Values.

The finest ordered candidate from docking outcomes was further investigated for appraising their thermodynamic properties and the binding stability in the M pro binding pocket by molecular dynamic (MD) simulation studies. The molecular dynamic simulation was carried out by the GROMACS-2019.3 package (24, 25). The Amber99.sb force field was employed in MD simulations (26) , and all drug topology parameters were ready by the AnteChamber Python Parser InterfacE (ACPYPE) (27) . The pKa values were calculated to determine which residue adopts non-standard ionization states, using PROPKA 3.1 webserver (28) . A TIP3P water model was chosen (29) , and the drug-protein complex was waterlogged in a dodecahedron box. To neutralize the system, Na + ions were substituted with solvent water molecules. The complex energy minimization was carried out, and two stages of the process started MD simulation: 1) 500 ps simulation in the NVT ensemble (constant number of particles, volume, and temperature); 2) 1 ns simulation in the NPT ensemble (constant number of particles, pressure, and temperature). In the end, the MD simulation was run at 300 K temperature for 100 ns. The Particle Mesh Ewald (PME) method and the linear constraint (LINCS) algorithm were employed to calculate long-range electrostatic interactions and covalent bond constraints, respectively. Moreover, structure imagining was done by VMD 1.8.6 (30) and PyMOL Tcl.

The validation docking was carried out and interactions between N3 (N-[(5methylisoxazol-3-yl)carbonyl]alanyl-L-valyl-N-1-((1R,2Z)-4-benzyloxy)-4-oxo-1-{[(3R)-2-oxopyrrolidin-3-yl]methyl}but-2-enyl)-Lleucinamide, crystallography ligand) and M pro protein was compared with the crystallographic poses in PDB: 6lu7. The binding pocket of N3 was investigated as the following: the catalytic His163 and Cys145 residues in the bottom of the packet formed two hydrogen bonds, and hydrogen bonds were seen between Gly143 and Glu166 amino acids in the edges pocket with N3. The main hydrophobic amino acids that sat around N3 were His41, Phe140, Lue141, Asn142, Ser144, His164, Met165, His172, and Gln189, as seen in Figure 2 (6) . The root-mean-square deviation (RMSD), between the valid conformations and its crystallography conformation, was obtained below 2 Å.

To predict new M pro inhibitors, a structurebased virtual screening (VS) study was carried out over created library from the Drug Bank web server. Docking runs were done for all 43 drugs similar to the piperine scaffold, and its results were organized in terms of binding energy and fundamental catalytic interactions. Thirty-four of them were sat in the M pro active site, but between them, four drugs (curcumin, lasofoxifene, alvimopan, and donepezil) with the lowest binding energy were selected for further investigation. Among these four drugs, curcumin structurally showed the most similarity to the piperine scaffold. In lasofoxifene, alvimopan, and donepezil structures, saturated rings are piperidine or pyrrolidine rings, the same as piperidine ring in piperine structure.

Curcumin, extracted from a plant called "turmeric," with the lowest binding energy, acts as a lipoxygenase inhibitor and prevents tumor invasion by irreversibly binding CD13/ aminopeptidase. Turmeric is commonly consumed as a color food, and its root is also used in a few medicinal products to treat pain and inflammation, such as osteoarthritis. The medicinal properties of turmeric have been known for thousands of years, but curcumin from natural dietary or synthetic sources gained first approval by the Food and Drug Administration (FDA) in 2013.

Lasofoxifene is a selective estrogen receptor modulator (SERM) with a non-steroidal structure that selectively binds to estrogen receptors α and β. The European Commission approved it in March 2009. Lasofoxifene could be used for postmenopausal osteoporosis treatment to decrease the risk of both vertebral and nonvertebral fractures.

Alvimopan competitively binds to the μ-opioid receptor as a selective antagonist. FDA licensed alvimopan in 2008 to speed up the time to upper and lower gastrointestinal recovery following surgeries.

Finally, donepezil increases the accessibility of acetylcholine at the synapses by binding reversibly to acetylcholinesterase and inhibiting the hydrolysis of acetylcholine. Donepezil gained approval by FDA in 2004 and could be used to treat confusion (dementia) related to Alzheimer's disease.

Although molecular docking simulation using Autodock software is highly favored in presenting the ligand poses within the protein active site, the protein is considered rigid in docking calculations. In this study, the IFD method was carried out to account for both ligand and receptor flexibility.

The validation of the IFD model was carried out before the docking simulation. The IFD-generated N3 model (cyan, Figure 3 ) and the native structure of N3 (yellow, Figure 3 ) in N3/M pro complex (6lu7) were superimposed, and RMSD of 2.34 Å was obtained for entire heavy atoms (excepting the hydrogen atoms). Therefore, the IFD module could be predicting the binding interactions between inhibitors and M pro protein.

IFD was performed between M pro protein and the four obtained drugs with the lowest binding energy by Autodock. The IFD results are reported in Table 1 . The GlideScores in the curcumin-M pro and lasofoxifene-M pro complexes were obtained lower than alvimopan and donepezil complexes. It suggests that the binding affinity of curcumin and lasofoxifene complexes is also lower than alvimopan and donepezil complexes, the same as the calculated binding affinity by Autodock. The IFD generated interactions between curcumin and lasofoxifene with M pro protein are shown in Figure 4 . IFD results show that curcumin interacted with Cys145, Gly143, Glu166, Ser144, Leu141, and Arg188 of M pro protein via hydrogen bonds lasofoxifene formed hydrogen bonds only with Cys145, Gly143, and Glu166 amino acids. As seen in Figure 4B , Glu166, with a negative charge, sat close to pyrrolidinium ion (the distance: 1.58 Å), and an ion bond could be formed between negative and positive charges. The ion bond formation could be a reason for the lasofoxifene GlidScore value that is close to the obtained value of curcumin. The docking affinity of lasofoxifene (IFD score = -691.743) was better than curcumin, alvimopan, and donepezil with IFD scores of -684.229, -672.998, and -674.228, respectively.

Combining more energy terms such as surface accessibility area and solvation energy with a suitable force field can make more satisfactory accuracy for the ligand binding energy computing (21) . Thus for each four selected compounds, the Prime MM-GBSA method was done on the state with the bottommost obtained GlideScore from IFD studies. The calculated ΔG bind of the compounds and the contribution of main energy components (coulomb, covalent, hydrogen bonding, lipophilic binding, the generalized born electrostatic solvation, and van der Waals) were reported in Table 2 . The same as IFD results, lasofoxifene showed the lowest ΔG bind . This suggests that lasofoxifene is the most stable ligand in the M pro protein binding pocket. In Table 2 , the free energy components revealed that the lipophilic and van der Waals interaction energies (ΔG Lipo and ΔG vdW ) have the most significant contribution in the ligands binding energy. The binding energy contribution of the main amino acids in the active site is shown in Table 3 . In M pro protein active site, the contribution of lipophilic interactions is more than hydrogen bond interactions. The contribution of His163 and Glu166 was more than other amino acid residues in the binding pocket.

According to Table 1 , lasofoxifene showed the lowest binding energy among all 34 extracted drugs with two main hydrophilic interactions (hydrogen bonds with Cys145 and His163). MD simulation of lasofoxifene was performed to ensure its stability in the binding pocket of M pro protein. Recently curcumin was introduced as M pro inhibitor (31); therefore, it was chosen as a reference and compared its interaction modes with lasofoxifene.

After 100 ns simulations, the MD trajectories were analyzed. RMSD was calculated to determine the conformational stability of M pro protein in all simulation times. As illustrated in Figure 5a , the RMSD profile of backbone atoms in M pro -lasofoxifene and M pro -curcumin complexes showed small variations about 0.35 nm and 0.25 nm, respectively. By analyzing the RMSD plots of lasofoxifene and curcumin (Figure 5b) , it can be recognized that both drugs were almost entirely superimposed during the simulation. RMSD plots suggest that both complexes were stable during simulation time.

Rg was calculated to evaluate the compactness of the protein. Variations of protein flexibility were obtained by RMSF, as are demonstrated in Figure 6 . The oscillations of both complexes were superimposed ( Figure   6a ), except in residues 45-60 in the M prolasofoxifene complex. This indicates that the central regions of the protein, such as Cys145, His163, and His41, in both complexes were more stable during MD simulation. The Rg values of both complexes were identical, and their conjunction was kept in all simulation time, as represented in Figure  6b . Average values of RMSD, RMSF, and Rg were calculated at 0.225, 0.090, and 2.204, respectively, for M pro -lasofoxifene complex.

The average values of RMSD, RMSF, and Rg of M pro -curcumin complex were also obtained from 0.193, 0.089, and 2.213, respectively.

The binding free energy has also been recalculated for M pro -curcumin and M prolasofoxifene complexes using the g_mmpbsa program. The obtained average binding energy components are reported in Table 3 . The results showed that M pro -lasofoxifene binding energy was lower than M pro -curcumin binding energy checked by both programs (MMGBSA and g_mmpbsa). The contribution of each main amino acid in the binding energy has shown in Table 4 .

Since inhibition of M pro protein has been known as a critical strategy for COVID-19 treatment, highly effective M pro inhibitors have been investigated with various in silico and traditional methods. In this study, virtual screening with different computational methods has been introduced to predict new M pro inhibitors. According to this method, virtual screening was first performed on 43 drugs (similar to piperine scaffold). Finally, 34 medicinal compounds were extracted with therapeutic effects against osteoporosis, Alzheimer's, cancer, and reduced heart rate. Among extracted drugs, curcumin, lasofoxifene, alvimopan, and donepezil demonstrated the lowest binding energy and the most interactions with the COVID-19 M pro protein active site. Among extracted drugs, lasofoxifene, as a third-generation SERM with high selective affinity for both ERa and ERb subtypes, was the best in terms of binding energy and the critical interactions with amino acids. An induced-fit docking method was performed to consider the flexibility of both ligand and receptor. The obtained score from IFD confirmed molecular docking results, and the IFD score of lasofoxifene was observed lower than others (-691.743 kcal mol-1). Lasofoxifene binding free energy, calculated by two programs including Prime MM-GBSA and g_mmpbsa, showed the lowest value; -107.09 and -173.97 kcal mol-1, respectively. And finally, MD simulation investigations on the finest docked pose of lasofoxifene-M pro and curcumin-M pro complexes were employed to reveal the overall stability of the two complexes. RMSD, RMSF, and Rg results obtained from MD simulation revealed that lasofoxifene and curcumin exhibited the best stability in the M pro active site. As a result, lasofoxifene or its close derivatives can be considered promising drugs for the treatment of COVID-19.

We would like to express our special thanks to the Hormozgan University of Medical Sciences for their financial support (Project No. 990193).

The authors declare no conflict of interest.

Maryam Abbasi: Participated in the study's design, carried out IFD study, performed molecular dynamic simulation, and drafted the manuscript. Hojjat Sadeghi-aliabadi: Participated in the study's design, was responsible for the study registration and gave final approval of the version to be published. Table 4 . The average binding energy components were obtained from the g_mmpbsa program.

Potential treatments for COVID-19; a narrative literature review

Simeprevir, potential candidate to repurpose for coronavirus infection: virtual screening and molecular docking study. Preprints (2020)

The substrate specificity of SARS coronavirus 3C-like proteinase

Systemic in silico screening in drug discovery for coronavirus disease (COVID-19) with an online interactive web server

Drug repurposing for coronavirus (COVID-19): in-silico screening of known drugs against coronavirus 3CL hydrolase and protease enzymes

Structure of Mpro from COVID-19 virus and discovery of its inhibitors

Identification of Dietary Molecules as Therapeutic Agents to Combat COVID-19 Using Molecular Docking Studies. Preprints (2020)

Perspectives on plant flavonoid quercetin-based drugs for novel SARS-CoV-2. Beni-Suef Univ

Discovery of new hydroxyethylamine analogs against 3CLpro protein target of SARS-CoV-2: Molecular docking, molecular dynamics simulation, and structure-activity relationship studies

Potential inhibitor of COVID-19 main protease (Mpro) from several medicinal plant compounds by molecular docking study. Preprints (2020)

AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility

Automated docking using a Lamarckian genetic algorithm and an empirical binding free energy function

Molecular dynamics simulation by GROMACS using GUI plugin for PyMOL

Schrödinger Suite 2015-2 Protein Preparation Wizard; Epik version 3.2, Schrödinger, LLC

Development and testing of the OPLS all-atom force field on conformational energetics and properties of organic liquids

Schrödinger Suite 2015-2, Protein Preparation Wizard, Epik version 3.2

Induced-fit docking studies of the active and inactive states of protein tyrosine kinases

Induced fit docking

Physics-based scoring of protein− ligand complexes: Enrichment of known inhibitors in large-scale virtual screening

Accurate prediction of the relative potencies of members of a series of kinase inhibitors using molecular docking and MM-GBSA scoring

Combined molecular mechanical and continuum solvent approach (MM-PBSA/GBSA) to predict ligand binding

The GROMACS development team GROMACS User Manual

GROMACS: High performance molecular simulations through multilevel parallelism from laptops to supercomputers

A second generation force field for the simulation of proteins, nucleic acids, and organic molecules

ACPYPE-Antechamber python parser interface

Improved treatment of ligands and coupling effects in empirical calculation and rationalization of p K a values

Comparison of Simple Potential Functions for Simulating Liquid Water

VMD: visual molecular dynamics

Virtual screening with AutoDock: theory and practice