Several of the fragments occur in commercially available compounds. The Express-Pick collection is available as dry solids for about $20 per compound (excluding shipping, duty and taxes and for a minimum of about 100 compounds) which provides an alternative to custom synthesis of novel molecules for leads.
This thread relates to those molecules that contain C(=O)Nc1cnccc1 which an essential part of fragment x0678. We filtered the molecules using THINK’s standard property filters and then docked the 1122 remaining into Mpro-x1093 saving the conformer with the best score prior to side-chain relaxation. We found a further 1385 compounds in the ChemBridge CORE library which we have also docked.
We used the 1093 (5RF3) crystal structure with 3 centre pharmacophore docking targeting residues that have strong interactions in the non-covalent crystal complexes: (41),(44),(140),(142),(143),(144),(163),(166),(189).
At the present time, we have only submitted the first 100 molecules from Express-Pick (and after the 13 May which means that they were not included in our docking of submitted molecules). An SD file of the Docked molecules (ranked) is available. Some of these have contacts which would normally be removed during refinement (see comments on other thread). An SD file of My selection from this subset is also available.
To date I have failed to upload any SD files (SERVER ERROR 500).
We have also docked commercially available molecules containing fragments x0540, x0995 and x1093. Obviously, there are other companies that supply molecules for screening and we have some of those collections as well! However, the tedious step is submitting the molecules by cutting and pasting SMILES (it is also crucial to link these back to the molecule catalogue IDs).
1 Like
Hi @EKDavies,
Sorry I am just getting around to this, and was looking into some of the ChemBridge molecules for ordering purposes. Just wondering, have you run the docking and seen if it recapitulates the pose of the the initial amino-pyridine hits such as
all of which have structures ( x2563, x2569, x2646, x2649 )
Having found that the first one I tried wasn’t in the SD file I reran these from E-services on treweren.com (I’d need your email address to setup an account so you could do so). This facility gives a brief analysis for molecules which did not dock. I think I used a cutoff of -40 on the score so any molecules with less negative (poorer scores) were not included in previous SD files. I’m in the process of setting up “portal” (I hate that term) to access a MySQL database of docking results and activities but this wont give any analysis about molecules which failed to dock.
The links to the SD files should work. I’m happy to run these with relaxed side-chains if that would help. The following are just cut and pasted from the E-services results docking into 1093 (5RF7).
Job ID: w187: DAR-DIA-23aa0b97-20
Using: Covid-19 Protease inhibitor search
Result: Covid-19 search failed. Reason: Clash with protein for molecule :
This means that there was one or more bad contacts with the protein which significantly reduced the score and wasn’t relaxed during refinement whilst maintaining pharmacophore constrants.
Job ID: w188: DAR-DIA-23aa0b97-17
Using: Covid-19 Protease inhibitor search
Result: Completed. Found 1 hit (Score: -32.874)
Molecules: http://www.treweren.com/DeNovo2D/qgdd28ud5nyuxndi.sdf
Graphics: http://www.treweren.com/DeNovo2D/qgdd28ud5nyuxndi.htm
So I didnt include this molecule in the top hits.
Job ID: w189: TRY-UNI-714a760b-20
Using: Covid-19 Protease inhibitor search
Result: Completed. Found 1 hit (Score: -38.573)
Molecules: http://www.treweren.com/DeNovo2D/i3yk3w91481va5go.sdf
Graphics: http://www.treweren.com/DeNovo2D/i3yk3w91481va5go.htm
This scores a little better but not necessarily significantly better.
Job ID: w190: TRY-UNI-714a760b-18
Using: Covid-19 Protease inhibitor search
Result: Covid-19 search failed. Reason: No Suitable Conformers for molecule :
This is not the most helpful analysis message but it means that none of the conformers samples meet the pharmacophore constraints but this could be simply that the conformer sampling was too coarse. I’m in the process of beginning to look at the PDB structures 6M0K and 6LZE to understand if the starting crystal structures (and the associated 3D pharmacophores) make a significant difference.
Job ID: w191: TRY-UNI-714a760b-6
Using: Covid-19 Protease inhibitor search
Result: Covid-19 search failed. Reason: Refined score too high for molecule :
This molecule had a conformer which fitted at least one pharmacophore but it scored too high after adjusting torsions to be considered a hit. Relaxing side-chains might make a difference here.
Ok thanks for letting me know @EKDavies. I am not a docking specialist, and structures aren’t my strong suit, but I know others have had success in recapitulating the crystal structure poses of these initial hits using GOLD/other software. That seems like a sensible step to me before using the method to guide ordering of analogs, but again I am no expert here.
I will revisit the molecules that werent docked with a changed setup in the coming days.
BTW The virtual screening pharmacophore technology we use we originally developed at Chemical Design in the early 1990s (for the Chem-X software - the first 3D database searching software). The goal then (and now) was to find molecules which had a higher activity frequency (or hit rate) than random screening. This differs from the “occlusion” docking software which we have but dont recommend for virtual screening.
Your fragment crystal structures provide a fantastic resource to identify residues and pharmacophores which can be used in searches. Our virtual screening to date has been blind - without considering any of your activity data. We have been excited by the fact that more than 1 in 4 of our predicted hits bind. We have also been excited that our virtual screening query predicts the (published) activity of Famotidine and the inactivity of Lopinavir and Ritonavir.
In principle, we can tune our virtual screenining using your activity data to change the rankings of some residue interactions. However, this approach is usually used with some higher activity data when the objective is to find structurally diverse/different lead compounds.
Ultimately, virtual screening is a quick and dirty calculation to find leads for medicinal chemistry faster and cheaper that pure HTS IMHO more accurate/precise predictions would need solubility data/calculations. However, a cost-benefit analysis makes this difficult to justify!
1 Like
I had feared that it might be timeconsuming (and tedious) to track down exactly what was happening, so I gave priority to some other stuff! It proved really straight forward. I increased the granularity of the conformer searching and reduced the number of required strong residue interactions from 3 to 2.
Job ID: w187: DAR-DIA-23aa0b97-20
Using: Covid-19 Protease inhibitor search
Result: Completed. Found 1 hit (Score: -27.085)
Molecules: http://www.treweren.com/DeNovo2D/52atoay0vsuv1kf4.sdf
Job ID: w190: TRY-UNI-714a760b-18
Using: Covid-19 Protease inhibitor search
Result: Completed. Found 1 hit (Score: -28.358)
Molecules: http://www.treweren.com/DeNovo2D/zg3nt1f3bjwxmaf4.sdf
Job ID: w191: TRY-UNI-714a760b-6
Using: Covid-19 Protease inhibitor search
Result: Completed. Found 1 hit (Score: -25.288)
Molecules: http://www.treweren.com/DeNovo2D/7xkyb3wzbed7upaa.sdf
1 Like