Hello,
I hope this email finds you well.
I am writing to inquire about a few questions regarding the tools provided by Zhang Lab, specifically the TM-Score, TM-Align, and 3DRobot. I have encountered some observations and would greatly appreciate your assistance in clarifying these matters.
Q1: Discrepancy between TM-Align and TM-Score results
While utilizing both TM-Align and TM-Score to assess the structural similarity between an AlphaFold2 model and the true structure, I noticed that TM-Align yields slightly lower TM scores compared to the TM-Score server. I am curious to understand the reasons behind this disparity.
Q2: RMSD calculation and its correlation to structural proximity
Upon utilizing TM-Align and evaluating the Root Mean Square Deviation (RMSD), I observed low RMSD scores indicating "Good" alignment, despite the structure not being particularly close to the true structure. I would appreciate an explanation regarding how RMSD is calculated and why it may not always reflect structural proximity accurately.
Q3: Discrepancy in the number of decoys generated by 3DRobot
When submitting a job using 3DRobot and specifying a request for 1000 decoys, I received an output with only 308 decoys. I am interested in understanding the factors contributing to this discrepancy and why the requested number of decoys was not fully generated.
Thank you for your time and attention to these inquiries. I greatly value the expertise of Zhang Lab and look forward to your insights.
Best regards,
MNA
Inquiry regarding TM-Score, TM-Align, and 3DRobot
Moderator: robpearc
Re: Inquiry regarding TM-Score, TM-Align, and 3DRobot
Dear user,
Q1: TM-align will align two structures based on their real structure similarity, while TM-score will align them based on the residue order in those pdb files.
If your case is the model and true structure that has the same residue index, then always use TM-score program to assess the model quality.
TM-score will align two structures based on the residue order, so you can see the final alignment is based on this order
model residues <-> native residues
1<->1
2<->2
3<->3
4<->4
...
100<->100
while TM-align will re-align based on structure and ignore the residue index, so you may find the final alignment like this
model residues <-> native residues
1<->-
2<->1
3<->2
4<->3
...
100<->99
-<->100
It is a complete residue index mismatch.
That is why TM-score and any TM-align have different results.
Q2: RMSD will be largely affected by some long flexible tail of the structure. For example, if you have two structures, the core region is the same, but they have a loop tail that is far away from each other, then using RMSD to assess, the RMSD will be very large. In contrast, TM-score will give different weight to those core regions and bad modeling (aligned) regions. So please try to use TM-score, which is more suitable than RMSD.
Q3: Please let us know your 3DRobot job ID, and we will see the case.
Best
IT Team
Q1: TM-align will align two structures based on their real structure similarity, while TM-score will align them based on the residue order in those pdb files.
If your case is the model and true structure that has the same residue index, then always use TM-score program to assess the model quality.
TM-score will align two structures based on the residue order, so you can see the final alignment is based on this order
model residues <-> native residues
1<->1
2<->2
3<->3
4<->4
...
100<->100
while TM-align will re-align based on structure and ignore the residue index, so you may find the final alignment like this
model residues <-> native residues
1<->-
2<->1
3<->2
4<->3
...
100<->99
-<->100
It is a complete residue index mismatch.
That is why TM-score and any TM-align have different results.
Q2: RMSD will be largely affected by some long flexible tail of the structure. For example, if you have two structures, the core region is the same, but they have a loop tail that is far away from each other, then using RMSD to assess, the RMSD will be very large. In contrast, TM-score will give different weight to those core regions and bad modeling (aligned) regions. So please try to use TM-score, which is more suitable than RMSD.
Q3: Please let us know your 3DRobot job ID, and we will see the case.
Best
IT Team
Re: Inquiry regarding TM-Score, TM-Align, and 3DRobot
Thank you very much for the helpful answers.
Here is the 3DRobot Job ID: D001291
Here is the 3DRobot Job ID: D001291
Re: Inquiry regarding TM-Score, TM-Align, and 3DRobot
Dear user,
Since you set the threshold as 5A, so I guess there are no that much different decoys could be generated by this very high structure similarity threshold, we tested your target with 20A as threshold, the job id is https://zhanggroup.org/3DRobot/output/D001293/, so you can see we get 1000 decoys.
Best
IT Team
Since you set the threshold as 5A, so I guess there are no that much different decoys could be generated by this very high structure similarity threshold, we tested your target with 20A as threshold, the job id is https://zhanggroup.org/3DRobot/output/D001293/, so you can see we get 1000 decoys.
Best
IT Team