Docked molecule ideas from the public

[TomkUCL]

Here we will accept molecule design suggestions from the public (provided a score below -5.2 kcal/mol) to see if any improved Vina binding scores can be found at the C-terminus-B pocket.

A three-part tutorial series on molecular docking and interactive visualisation can be found on YouTube for newcomers to molecular modelling below.

Video link: https://www.youtube.com/playlist?list=PLIjjFfNrbXnw_N7WRYnFmgt0VBWLUvTf2

Simply run your docking calculation as per these tutorials, then if you obtain a Vina docking score more negative than -5.2 kcal/mol, then show us your interaction maps (just like those below) with any comments. If it looks chemically feasible, we can try to make it in the lab and test it!

[TomkUCL]

Here are the original fragment interactions from the protein data bank uploaded by Joe Newman et al. PDB ID: 5RM9. New interactions no found in this pdb structure will be italicised for clarity. Fragment crystal structure interactions: Halogen bond (Val495) Pi-Alkyl (LEU590, PRO 593) Pi-Cation (ARG 502) Hydrogen bond (ASN503) Pi-Alkyl (PHE472)

An AutoDock Vina score of -5.2 kcal/mol was found for the ammonium form of this molecule using the docking method followed in video 2 in the YouTube videos, so any score more negative than this is an improvement! Fragment 5RM9 (ammonium form) docked: Halogen bond (Val495) Pi-Alkyl (LEU590, PRO593, MET474) Pi-Cation (ARG 502) Hydrogen bond (ASN503) Carbon Hydrogen bond (PRO593)

[TomkUCL]

Here is a compound that I am currently attempting to synthesise in our lab at UCL. This is the lowest energy conformer with a Vina score of -6.7 kcal/mol (an improvement of -1.5 kcal/mol compared to the 5RM9 fragment). Molecule 1 docked: Carbon-Hydrogen bond (PRO593) Halogen bond (GLU591) Hydrogen bond (ARG502, TRP506) Pi-Cation (Arg502) Pi-Alkyl (LEU590)

We can see that the 4-fluoropyridine amide forms important Pi-cation, Pi-alkyl, carbon hydrogen bonding,and conventional hydrogen bonding interactions with ARG502, LEU590,PRO593,and ARG502, respectively, as was the case with the original fragment. A halogen bond is now formed with GLU591 within the pocket. The N-oxide appears to form a conventional hydrogen bond outside of the pocket with TRP506, however the amide bond is torsionally strained in this conformation.

[Ignatious1999]

Have you managed to run calculations for the free binding energy for this molecule to assess weather it will bind or not

[TomkUCL]

@Ignatious1999 thanks for your comment. Synthetic chemistry is more my niche rather than computational, so would you mind elaborating?

An improvement in the Vina calculated affinity compared to that of the co-crystallised fragment would suggest so, would it not? Are you referring to MD simulations?

[Ignatious1999]

I have never used vina in docking, i usually use Maestro from schrodinger and from what i have learnt docking score only gives the favorable low energy conformation position of a ligand on the receptor site. Unless the calculated affinity you are referring to is an equivalent of free energy binding in maestro.

[Ignatious1999]

Yes i am referring to MD simulations.

[h4RkhW8t53e]

Hi, I used smina v 1.2.3 (a version of vina) to dock some structures from ChEMBL v31 to 5rm9. Sending some examples with ChEMBL numbers, smiles, Vina score and Ligand Efficiency (Vina score / non-H atoms): Also attached are the docked structures with the native fragment as reference. Can look for more if you find this useful.

CHEMBL4097558 C/C(=C\c1ccccc1C#N)c1ccccc1 -6.50 -0.38 CHEMBL4973507 Cn1nc(C2CC2)nc1-c1csc2c1CCCC2 -6.30 -0.35 CHEMBL1457698 NC(=O)CSc1nnc(-c2ccccn2)n1N -6.00 -0.35

[h4RkhW8t53e]

[TomkUCL]

@Ignatious1999 No I have not run any MD simulations - I think you're right, the Vina score is equivalent to the free-binding energy in Maestro which is why the scores are negative, but the scoring function that is used for the calculation is different which is why Vina and GLIDE will give different binding free-energy scores for the same molecule on the same protein.

Regardless of the docking software used, a forcefield free-energy binding score is a direct result of both the scoring function used and of the properties of the molecule. AutoDock Vina uses a hybrid scoring function, the details of which can be found in this paper; https://doi.org/10.1002%2Fjcc.21334

[TomkUCL]

Thanks, @h4RkhW8t53e are those scores ranges or absolute values, e.g. -6.50 to -0.38, for example?

[h4RkhW8t53e]

-6.50 is the Vina/Smina score, -0.38 is the Vina score divided by the number of non-H atoms. Same for the other structures.

[TomkUCL]

Great thanks! @h4RkhW8t53e Nice to see an improvement in the scores of ~1 kcal/mol for similar size fragments. Would you be able to include interaction maps like those shown above so we can explain these scores?

[h4RkhW8t53e]

It seems that Discovery Studio Visualiser does not have Mac version, have to get and install the Linux one ... I attach a zip file of the docked structures in mol2 format. Hope this helps. 5rm9.zip

[holeung]

I think this is a challenging pocket. It is shallow and exposed. I could not come up with very high scoring hits. Here are the best I have found so far.

mol2 O=C1NC(CN(CC2)CCC2c([nH]c2ccc3)nc2c3F)=Nc2c1ccc(C(F)(F)F)c2 Pubchem 166211415 Vina score: -8.3

mol3 O=C([C@@H]1Oc(cccc2)c2OC1)N(CC1)CCN1C(c(cc1)cc2c1nc(C(F)(F)F)[nH]2)=O PubChem-17603525 Vina score: -8.1

mol2-3.sdf.zip

[TomkUCL]

Excellent work @holeung May I ask where this molecule came from? @kipUNC is this useful? My Vina docking scores range from -7.8 kcal/mol (binding mode pictured below) to -6.8 kcal/mol for this molecule, whilst Ho leung is getting slightly higher. I think this is slightly higher than the Vina scores for the de novo generative molecules.

[holeung]

The molecules came from screening Pubchem. I am not confident that it is possible to make much better molecules than these due to the pocket.

Are there reactive side chains nearby for designing a covalent inhibitor?

[TomkUCL]

@holeung Yes the binding affinity scores are generally quite low (usually scoring about -6 to -8 kcal/mol in Vina). Your point on covalent inhibition for this pocket has been raised before by Peter Kenny and I think should be seriously considered for this pocket given the mobility of the RecA2 domain during ATP hydrolysis.

Looking at the fragment cocrystal structure the following residues look like the most promising residues for covalent warhead targeting (ref: https://doi.org/10.1016/bs.armc.2020.10.001) :

• MET466 (oxaziridine or epoxide warhead).
• GLU583 (Isoxazolium or 2H-Azirine warhead).
• SER581 (sulfonyl fluoride or fluorosulfate warhead).
• THR580 (epoxy ketone, boronate, or chloroalkyl β-lactone warhead).
• SER460 (lactam, lactone, carbamate, α-amino nitrile, phenol ester, fluorophosphonate, fluorosulfate warhead).
• CYS463

Part of the challenge of introducing covalent warheads here is that these are non-catalytic residues.

[h4RkhW8t53e]

Several more molecules from screens of ChEMBL and ZINC with scores in the range around -7 kcal/mol. Good luck with the project!

ID (CHEMBL/ZINC), smiles, Vina score

ZINC-40533905 O=C(CCc1nc(-c2cccnc2)no1)N[C@@H]1CCCCNC1=O -7.20

CHEMBL1517294 Cc1ccc(-c2ccc(S(=O)(=O)NCc3ccco3)cc2)cc1 -7.60
and similar ZINC-10430983 O=C(Nc1ccccn1)c1nn(-c2ccc(F)cc2)c2c1CCC2 -7.50 CHEMBL3314997 Cc1c(Br)cccc1C(=O)Nc1ccccn1 -7.10 CHEMBL1540986 O=C(NCc1ccccn1)c1ccc(-c2ccccc2)cc1 -7.50

ZINC-89669014 Cc1cccc(C(=O)C[C@@]2(O)C(=O)Nc3ccccc32)c1 -7.80 and similar ZINC-79112380 O=C(C[C@@]1(O)C(=O)Nc2ccccc21)c1cccc(F)c1 -7.80 ZINC-2495741 Nc1cccc(C(=O)C[C@@]2(O)C(=O)Nc3ccccc32)c1 -7.30 ZINC-170620016 O=C(C[C@@]1(O)C(=O)Nc2cc(Br)ccc21)c1ccccc1 -7.30

CHEMBL330037 Cc1cccc(Nc2nccc(-c3c[nH]c4ccccc34)n2)c1 -7.70

5rm9.zip

[Sharmeen25j]

Hi @TomkUCL,

We performed an extensive virtual screening of a molecular library of over 73k+ molecules against the target in question, SARS-CoV-2 helicase (PDB ID: 5rm9). The ligands of interest were downloaded from the Maybridge and ZINC-15 databases, the FDA library (containing FDA-approved drugs), the world library (containing drugs approved by other agencies in the world), and the investigational subset (containing drugs currently in clinical trials).

After the Virtual Screening, considering the top 100 drugs based on their docking scores, we manually screened them for all their properties, like ADME properties, physicochemical properties, etc. and found that the protein docked with the ligand 'zinc000024666930_uff_e=933.58' gave the best docking score of -11.58 kcal/mol at the C-terminus-B pocket of the protein.

Here are some of the details about the successful compound obtained:

Ligand: zinc000024666930_uff_e=933.58 iDock score: -11.58 kcal/mol SMILES: CN(C1CCN(CC1)C(=O)C1=CC2=C(C=CC=C2)C=C1)C(=O)C1=C(C=C2C=CC=CC2=C1)C(=O)C(C1=CC=CC2=C1C=CC=C2)P(O)(O)O

The interaction map of the complex is as follows:

To further validate this, we confirmed the stability of the molecule by performing molecular dynamics simulation of the top 10 drugs and found that the same ligand was the most stable among them as seen in the RMSD graph which gave results in the range of 0.3-0.6 nm.

A detailed link for the protocol can be found here: https://docs.google.com/document/d/1jUnN_blt8NKTbQyELLFSWn0W46aN3zmcviyRrmnF9K8/edit#heading=h.qczz2sn6af5s

We'll be happy to solve any queries if needed.

Hoping to see this molecule taken forward for synthesis and testing.

[h4RkhW8t53e]

A few molecules from a screen of PubChem with scores in the range around -8.5 kcal/mol ...

PubChem ID, smiles, Vina score 1329558 Cc1cc(C)cc(N2C(=O)[C@H]3C4c5ccccc5C(c5ccccc54)[C@@H]3C2=O)c1 -8.40 and similar 1070619 CC12c3ccccc3C(c3ccccc31)[C@@H]1C(=O)N(c3cccc(F)c3)C(=O)[C@H]12 -8.60
1101300 Cc1cccc(N2C(=O)[C@H]3C4c5ccccc5C(c5ccccc54)[C@@H]3C2=O)c1 -8.50

88335464 O=S(=O)(OSnc1ccccc1)c1ccccc1 -8.60

6330251 c1ccc(CSiCc2ccccc2)cc1 -8.40 5rm9.zip

[TomkUCL]

@Sharmeen25j thanks very much for your input! I have just re-docked your suggestion in Vina since the iDock score (-11.58 kcal/mol) is much better than anything we have had. I was only able to produce a Vina score of -6.7 kcal/mol. Maybe it is just the difference in scoring functions (I am not familiar with iDock), although this discrepancy seems large for just that. Maybe it is worth @kipUNC checking this molecule with GLIDE.

[holeung]

With Sharmeen's compounds, most of the interactions are non-specific hydrophobic interactions. Is this compound even soluble?

[Sharmeen25j]

@TomkUCL ..Thanks for your valuable input. We'll screen the library in Vina and get back to you with the updated score.

[mattodd]

As mentioned in the meeting in #37 I think it's important, if we're using different software (and maybe even if we're not) for everyone to dock a "control" - a molecule for which we have a good number and for which we're pretty confident. Then all the novel, suggested structures can be compared to that by the suggester. I'd suggest we use one of the N-oxides that have been studied fairly extensively, such as one of these three. @TomkUCL @kipUNC would that be reasonable? If so, do we have a diagram (revealing the pose) and a number (score) for one of those compounds that people can refer to, to check their own methods and results? Pinging @AndyXGH who was asking about this too.

[holeung]

These flat molecules that sit on the protein surface (not a pocket) and make few specific interactions usually don't bind.

[TomkUCL]

@AndyXGH what are your grid coordinates? We need to check that these are the same for reproducibility purposes. Also, it would be helpful to see the hydrophobic surface of the protein and 3-letter residue codes in your Discovery Studio visualisations so we know which residues these are. Let me know if you need help, but you can see how to do this in the original videos I uploaded earlier.