why the same task has different results for chai1 on the web server and on the local

[binbinbinv]

I fed an octameric helix bundle - the same protein sequence copied 8 times, and a small molecule into the chai1 model's web server and my own locally built version of chai1 - November 16, 2024, but the outputs of the two seem to be very different, and the results generated by the web server are even more what I expected, why is that?

Here are the different results:

1. From the website server:
   

2. From the local:
   

We can notice a big difference in the position of the small molecules.

Here are the input:

1. From the website server:
   

2. Fasta input from the local:

protein|Copy_1
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
protein|Copy_2
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
protein|Copy_3
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
protein|Copy_4
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
protein|Copy_5
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
protein|Copy_6
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
protein|Copy_7
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
protein|Copy_8
SAEELLRRSREYLKKVKEEQERKAKEFQELLKELSERSEELIRELEEKGAASEAELARMKQQHMTAYLEAQLTAWEIESKSKIALLELQQNQLNLELRHI
ligand|CH4T
O=C(NCCS(=O)(O)=O)CCC1=CC=C(C=C1)N(C2=CC=C(CCC(NCCS(=O)(O)=O)=O)C=C2)C3=CC=C(C=C3)C4=C5C(N=S=N5)=C(C6=NSN=C64)C7=CC=C(C=C7)N(C8=CC=C(CCC(NCCS(=O)(O)=O)=O)C=C8)C(C=C9)=CC=C9CCC(NCCS(=O)(O)=O)=O

[arogozhnikov]

@binbinbinv thanks for report, can you also post job_id for server task?

[binbinbinv]

@binbinbinv thanks for report, can you also post job_id for server task?

Sure, but I can not find the job id on the website, so I put the info on this screenshot

[arogozhnikov]

@binbinbinv

I assume that you ruled out potential simple reasons (i.e. installed chai-lab from scratch in clean virtualenv).

Here is my guess for you to test: what if you put ligand as first entry in fasta (currently it is last)?

[binbinbinv]

@binbinbinv

I assume that you ruled out potential simple reasons (i.e. installed chai-lab from scratch in clean virtualenv).

Here is my guess for you to test: what if you put ligand as first entry in fasta (currently it is last)?

@arogozhnikov
I tried the method you guessed — put the ligand as the first entry in the FASTA file. However, based on the generated results, only 2 out of 5 ligands ended up in reasonable positions (in the middle of the helical bundle), while the other 3 were still attached to the periphery. I'm curious, would installing it locally via "git clone" make a significant difference compared to using the web server?

[arogozhnikov]

I'm curious, would installing it locally via "git clone" make a significant difference compared to using the web server?

I don't expect any difference (unless you use MSAs), so guessing if that's just random variation. Having 3 runs on server vs 3 runs locally would be a more reliable comparison.

[arogozhnikov]

Independent suggestion: if you know for sure that ligand is inside, try encoding this info with restraints

[binbinbinv]

I'm curious, would installing it locally via "git clone" make a significant difference compared to using the web server?

I don't expect any difference (unless you use MSAs), so guessing if that's just random variation. Having 3 runs on server vs 3 runs locally would be a more reliable comparison.

@arogozhnikov
I tried several times on the web server. The results showed the ligand in different positions, which seem more like random occurrences. It was just by chance that the first set of five results were all in the middle position. Thanks for your helping.

[arogozhnikov]

cool, thx for checking

[Ruibin-Liu]

Independent suggestion: if you know for sure that ligand is inside, try encoding this info with restraints

Curious, can we adapt the current restraint example for protein-ligand?