Issues with Docking Scores in Libinvent Workflow

[natarajbalakrishnan]

Hello Reinvent team,
I ran the libinvent workflow with a set of scaffold SMILES provided as examples in the package and used the docking scoring component with Maize, utilizing Vina as the docking tool. When I analyzed the output molecules, I was surprised to find that in some cases, the docking SMILES were submitted with scaffolds as input, and in these instances, it appears that the docking failed and reported a zero score. I wonder why this is happening. Could you help me fix this issue in the Maize code?

> 10:54:10 <INFO> Score: 0.19 Agent NLL: 23.51 Valid:  85% Step: 1
>  | Agent Prior Target Score SMILES SMILES_state Scaffold docking with MAIZE docking with MAIZE (raw)
>  | 33.5323 32.1863 -26.9879 0.0406126 C[C:0](C)=[C:0](N)Nc1ccc([CH2:1][c:1]2cccc(NC(CC(=O)Nc3ccncc3)c3cccnc3)c2)cc1 1 O=C(CC(Nc1ccc[c:1]([CH2:1]c2ccccc2)c1)c1cccnc1)Nc1ccncc1 0.0406126 -4.6000
>  | 17.0623 17.8585 -17.8585 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(N)=O|*c1cc2c(N)ncnc2cc1 0 -- 0.0000000 0.0000
>  | 51.3347 52.4465 -52.4465 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(O)(C)C|c1(*)cc(C=C2c3c(cccc3)C3(c4ccccc4NC=N3)C2CN(C(OCc2ccccc2)=O)C)ccc1 0 -- 0.0000000 0.0000
>  | 21.6295 24.0515 -24.0515 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(O)=O|C(*)C(N=C1CCCCC1)c1ccccc1 0 -- 0.0000000 0.0000
>  | 49.4242 51.6185 -51.6185 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(O)=O|C(N=C(N)N)(C(NC(=O)C1N(C(=NCC(NN=Cc2cccc(*)c2)=NN=C(N)N)=O)CC(N=C(N)N)CC1)=O)N 0 -- 0.0000000 0.0000
>  | 29.9237 29.8384 -29.8384 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(O)=O|C1C(*)(O)C(=O)N(C#N)C1 0 -- 0.0000000 0.0000
>  | 12.0568 14.4818 -14.4818 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(O)=O|c1(OC)c(OC)c(OC)cc(*)c1 0 -- 0.0000000 0.0000
>  | 25.8595 25.2904 -25.2904 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(OC)=O|c1(C(=O)N2CCC(*)CC2)ccccc1N 0 -- 0.0000000 0.0000
>  | 9.6022 10.8005 -10.8005 0.0000000 c1(NC(N)=[*])ccc(C#[*])cc1*C(OCC)=O|c1cccc(*)c1 0 -- 0.0000000 0.0000
>  | 27.6745 31.7478 -18.0104 0.1073233 [CH2:0]=[C:0](N)Nc1ccc([CH2:1][c:1]2ccc(N=C(N=C3C(N)=Nc4ccccc43)N3CCCC3)cc2)cc1 1 C1=Nc2ccccc2C1=NC(=Nc1cc[c:1]([CH2:1]c2ccccc2)cc1)N1CCCC1 0.1073233 -6.3000
> 10:58:48 <INFO> Score: 0.25 Agent NLL: 22.99 Valid:  95% Step: 2
>  | Agent Prior Target Score SMILES SMILES_state Scaffold docking with MAIZE docking with MAIZE (raw)
>  | 5.1354 5.9366 15.0835 0.1642197 N#[C:0][CH2:0]c1ccc(N[C:1](N)=[NH:1])cc1 1 c1ccccc1 0.1642197 -7.1000
>  | 6.9654 9.3003 12.8210 0.1728219 [NH2:0][CH2:0]c1ccc(N[C:1](N)=[NH:1])cc1 1 c1ccccc1 0.1728219 -7.2000
>  | 56.7865 47.1197 -47.1028 0.0001318 CC(C)N=C(N)N=C(N)c1ccc(C(CCNC(N)=NC#N)N(CN=C(N)N)C(=O)c2ccc(NC(=N)N)[c:0]([CH2:0]c3ccc(N[C:1](N)=[NH:1])cc3)c2)cc1 1 O=C(NCc1ccccc1)c1ccc[c:0]([CH2:0]c2ccccc2)c1 0.0001318 4.8000
>  | 25.5939 23.6034 8.9778 0.2545407 NC(N)=Nc1n[c:0]([CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)cc(-c2ccccc2N)n1 1 c1ccc([CH2:0][c:0]2cc(-c3ccccc3)ncn2)cc1 0.2545407 -8.0000
>  | 15.1799 17.9614 24.3369 0.3304554 CCOC(=O)N1CCN(C(=O)c2ccc[c:0]([CH2:0]c3ccc(N[C:1](N)=[NH:1])cc3)c2)CC1 1 O=C(c1ccc[c:0]([CH2:0]c2ccccc2)c1)N1CCNCC1 0.3304554 -8.6000
>  | 24.1443 24.2348 0.2245 0.1910884 CC(C)(C)c1c[c:0]([CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)ccc1OC(=O)Nc1ccccc1 1 O=C(Nc1ccccc1)Oc1cc[c:0]([CH2:0]c2ccccc2)cc1 0.1910884 -7.4000
>  | 14.8015 15.2772 61.3276 0.5984749 NC1CCCN(C(=O)c2ccc[c:0]([CH2:0]c3ccc(N[C:1](N)=[NH:1])cc3)c2)C1 1 O=C(c1ccc[c:0]([CH2:0]c2ccccc2)c1)N1CCCCC1 0.5984749 -10.4000
>  | 33.4354 37.0291 -36.9509 0.0006115 COc1ccc(C[n+]2cc(C(=O)N(C)C)cc(-c3cc[c:0]([CH2:0]c4ccc(N[C:1](N)=[NH:1])cc4)cc3)c2)cc1OC 1 c1ccc([CH2:0][c:0]2ccc(-c3ccc[n+](Cc4ccccc4)c3)cc2)cc1 0.0006115 2.3000
>  | 18.4086 19.3133 -19.3133 0.0000004 CCON=C(C#N)c1ccc(-c2cc[c:0]([CH2:0]c3ccc(N[C:1](N)=[NH:1])cc3)cc2)cc1 1 c1ccc([CH2:0][c:0]2ccc(-c3ccccc3)cc2)cc1 0.0000004 14.6000
>  | 13.6629 12.9614 4.0910 0.1332216 CC(C)(C)OC(=O)N=C(N)c1cc[c:0]([CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)cc1 1 c1cc[c:0]([CH2:0]c2ccccc2)cc1 0.1332216 -6.7000
> 11:00:29 <INFO> Score: 0.29 Agent NLL: 16.22 Valid:  99% Step: 3
>  | Agent Prior Target Score SMILES SMILES_state Scaffold docking with MAIZE docking with MAIZE (raw)
>  | 14.2955 16.1739 8.2854 0.1910884 N#C[C:0]([CH2:0]c1ccc(N[C:1](N)=[NH:1])cc1)(c1ccccc1)c1ccccc1 1 c1ccc([CH2:0][CH:0](c2ccccc2)c2ccccc2)cc1 0.1910884 -7.4000
>  | 20.9876 22.2736 23.5720 0.3581687 N[C:1](Nc1ccc([CH2:0][CH:0]2C(C(=O)O)C2c2ccccc2)cc1)=[NH:1] 1 c1ccc([CH2:0][CH:0]2CC2c2ccccc2)cc1 0.3581687 -8.8000
>  | 18.4087 22.5084 4.4732 0.2107936 N=C(N)c1ccc(C(=O)[NH:0][CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)cc1 1 O=C([NH:0][CH2:0]c1ccccc1)c1ccccc1 0.2107936 -7.6000
>  | 26.2269 27.7428 37.2396 0.5076747 N[C:1](Nc1ccc([CH2:0][c:0]2[nH]ccc2-c2ccc(C(F)(F)F)cn2)cc1)=[NH:1] 1 c1ccc([CH2:0][c:0]2[nH]ccc2-c2ccccn2)cc1 0.5076747 -9.8000
>  | 8.4125 10.0970 35.7486 0.3581687 Cc1cccc[c:0]1[CH2:0]c1ccc(N[C:1](N)=[NH:1])cc1 1 c1cc[c:0]([CH2:0]c2ccccc2)cc1 0.3581687 -8.8000
>  | 16.0995 19.3292 41.7256 0.4769905 COc1c(C#N)ccc[c:0]1[CH2:0]c1ccc(N[C:1](N)=[NH:1])cc1 1 c1cc[c:0]([CH2:0]c2ccccc2)cc1 0.4769905 -9.6000
>  | 15.1363 15.2161 30.6295 0.3581687 CN(C)CC(O)c1ccc[c:0]([CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)c1 1 c1cc[c:0]([CH2:0]c2ccccc2)cc1 0.3581687 -8.8000
>  | 6.1996 7.5897 41.9288 0.3868632 Nc1cc[c:0]([CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)cc1 1 c1cc[c:0]([CH2:0]c2ccccc2)cc1 0.3868632 -9.0000
>  | 9.7395 16.3593 16.2219 0.2545407 NS(=O)(=O)c1cc[c:0]([CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)cc1 1 c1cc[c:0]([CH2:0]c2ccccc2)cc1 0.2545407 -8.0000
>  | 28.6089 30.3740 -30.3571 0.0001318 Cc1cc(C#Cc2ccc(N)cc2)ccc1NS(=O)(=O)c1cc[c:0]([CH2:0]c2ccc(N[C:1](N)=[NH:1])cc2)cc1 1 O=S(=O)(Nc1ccc(C#Cc2ccccc2)cc1)c1cc[c:0]([CH2:0]c2ccccc2)cc1 0.0001318 4.800

0

[halx]

Hi,

many thanks for your interest and welcome to the community!

Can you tell us please what version you are using?

Many thanks,
Hannes.

[natarajbalakrishnan]

Hi,

I am using Reinvent v4.3.3 and Maize Release 0.8.5 .

Regards,
Nataraj

[halx]

We have just released 4.5 so this may have been fixed since then.

[natarajbalakrishnan]

Thanks, I just checked changelog and I found the following fix and you are mentioning this fix right ?

[4.4.21] 2024-07-10
Changed
Removed labels from Libinvent generated molecules before passing to scoring components: maize, icolos, dockstream and ROCS
Write out clean SMILES without labels to CSV and tensorboard in staged learning and sampling run modes
Simplified update code for Libinvent RL

[halx]

Possibly, yes. You will need to try it out and let us know if it doesn't.

[natarajbalakrishnan]

Sure, thanks for your prompt response.

[natarajbalakrishnan]

Hi Halx,

We just tested the Reinvent output after upgrading to version 4.5. The LibInvent output CSV file still contains molecules with the * symbol in the SMILES column. We couldn't try the Maize integration due to certain issues with the upgraded Reinvent 4.5. However, our preliminary findings indicate that the * in the generated SMILES will be causing docking failures when using LibInvent. I have attached the CSV file for your verification. Could you please suggest a workaround for this issue?

Attachment:
staged_learning_1_4.5.csv

Regards,
Nataraj

[halx]

This means that REINVENT generated invalid fragments and a molecule cannot be constructed. We still leave in the invalid SMILES and you would have to check the SMILES_state column.

Internally, REINVENT should only pass valid (and non-duplicate) SMILES to scoring components. If that is not the case you will need to let me know.