Seeking Advice on Difficult MR Case: TFZ 6–8 for All Search Models, R-free ~0.50
I am working on a protein containing two domains: an N-terminal FHA domain and a C-terminal RING domain. I have collected a native dataset that extends to 2.45 Å resolution in space group P2₁2₁2 with unit-cell parameters approximately a = 46.9 Å, b = 91.9 Å, c = 165.0 Å. I have been struggling to obtain a convincing molecular replacement solution despite trying multiple approaches. What I have tried so far: Full-length AlphaFold model as a search model. Splitting the AlphaFold model into individual FHA and RING domains. Sequential/domain-wise MR searches. Aggressive model trimming. Poly-alanine versions of the models. Various combinations of domain placement strategies. However, the results remain largely the same: TFZ values are consistently in the range of 6–8. LLG values are not particularly convincing. Refinement of the best solutions leads to R-free values around 0.50 or higher. Electron-density maps are not readily interpretable. No solution has shown clear evidence of being correct after refinement and map inspection. At this stage I am trying to determine whether the issue is: Failure of MR due to domain rearrangement between the FHA and RING domains. Incorrect number of molecules in the asymmetric unit. Partial placement of domains that is being missed. Crystal pathologies such as pseudo-symmetry or translational NCS. Some other issue that I may be overlooking. I would appreciate suggestions on: Additional diagnostics to assess whether any of these TFZ 6–8 solutions might actually be correct. Strategies for multi-domain MR in cases where AlphaFold-based models are unsuccessful. Whether further model editing is likely to help. Other MR pipelines that may be worth trying before moving to experimental phasing. Any advice would be greatly appreciated.
I suppose you have tried to trim and partition the alphafold model according to low confidence regions using the phenix.process_predicted_model tool. It also adjust atomic B-factors of the partitioned alphafold model which should yield better results in the subsequent MR search.
Rob
________________________________
From: [email protected]
Thank you for the suggestion. I did use phenix.process_predicted_model on the AlphaFold model before MR. I tested both the processed full-length model and the resulting domain-partitioned models. Unfortunately, the results were very similar to those obtained with manually trimmed/domain-separated models. TFZ values remain in the 6–8 range and subsequent refinement stalls with R-free around 0.50. Therefore, I suspect there may be a larger issue than simply search-model preparation.
I think the first thing I'd try is Simbad: https://www.ccp4.ac.uk/html/simbad.html
If it doesn't find that my data comes from a contaminant:
1. Double-check space group attribution and dataset quality as there could be a missing screw axis
2. Crop the search model to secondary structure only, transform it to polySer if I'm desperate
3. Use Balbes
4. Use Arcimboldo (mostly beta-sheet)
5. Try a cropped version of the RING finger domain where only the secondary structure elements are present (hoping that the helix will give some contrast)
Switch from Phaser to MolRep for some of the search jobs above. You might still get weak scores from the MR functions, check the maps - and specially packing from Molrep - to judge the proposed solutions.
---------------------------------------------------
Carlos KIKUTI, PhD
Structural Motility Team
UMR144 - CNRS - Institut Curie
12 rue Lhomond - 75005 Paris, France
[email protected] mailto:[email protected]
Le 01/06/2026 10:34, « [email protected] mailto:[email protected] »
Hi, if it is P-21-21-2, then the axes might be a = 91.9 Å, b= 165.0 Å, c = 46.9 Å, (the shortest axis as c axis) Best, Guenter
I am working on a protein containing two domains: an N-terminal FHA domain and a C-terminal RING domain. I have collected a native dataset that extends to 2.45 Å resolution in space group P2₁2₁2 with unit-cell parameters approximately a = 46.9 Å, b = 91.9 Å, c = 165.0 Å.
Its always possible you crystallized a contaminant rather than your protein of interest. Try searching the PDB with unit cell and space group to find potential contaminant structures. And when in doubt try MR with GAPDH that sucker loves to hang around protein preps and loves to crystalize Becky Sent from my Galaxy -------- Original message -------- From: [email protected] Date: 6/1/26 3:51 AM (GMT-05:00) To: [email protected] Subject: [phenixbb] Seeking Advice on Difficult MR Case: TFZ 6–8 for All Search Models, R-free ~0.50 I am working on a protein containing two domains: an N-terminal FHA domain and a C-terminal RING domain. I have collected a native dataset that extends to 2.45 Å resolution in space group P2₁2₁2 with unit-cell parameters approximately a = 46.9 Å, b = 91.9 Å, c = 165.0 Å. I have been struggling to obtain a convincing molecular replacement solution despite trying multiple approaches. What I have tried so far: Full-length AlphaFold model as a search model. Splitting the AlphaFold model into individual FHA and RING domains. Sequential/domain-wise MR searches. Aggressive model trimming. Poly-alanine versions of the models. Various combinations of domain placement strategies. However, the results remain largely the same: TFZ values are consistently in the range of 6–8. LLG values are not particularly convincing. Refinement of the best solutions leads to R-free values around 0.50 or higher. Electron-density maps are not readily interpretable. No solution has shown clear evidence of being correct after refinement and map inspection. At this stage I am trying to determine whether the issue is: Failure of MR due to domain rearrangement between the FHA and RING domains. Incorrect number of molecules in the asymmetric unit. Partial placement of domains that is being missed. Crystal pathologies such as pseudo-symmetry or translational NCS. Some other issue that I may be overlooking. I would appreciate suggestions on: Additional diagnostics to assess whether any of these TFZ 6–8 solutions might actually be correct. Strategies for multi-domain MR in cases where AlphaFold-based models are unsuccessful. Whether further model editing is likely to help. Other MR pipelines that may be worth trying before moving to experimental phasing. Any advice would be greatly appreciated. _______________________________________________ phenixbb mailing list -- [email protected] To unsubscribe send an email to [email protected] Unsubscribe: phenixbb-leave@%(host_name)s
... you can try ContaMiner for this:
https://strube.cbrc.kaust.edu.sa/contaminer/submit/
Clemens
________________________________
Von: Rebekah E Bjork
On 1 Jun 2026, at 08:51, [email protected] wrote:
I have collected a native dataset that extends to 2.45 Å resolution in space group P2₁2₁2 with unit-cell parameters approximately a = 46.9 Å, b = 91.9 Å, c = 165.0 Å.
Just to check - you have been trying MR in all 8 space groups compatible with P222 point group - i.e. P222, P222(1), P22(1)2, P2(1)22, P2(1)2(1)2, P2(1)22(1), P 22(1)2(1) and P2(1)2(1)2(1)? It's not uncommon for assignment of screw axes based on systematic absences to turn out to be incorrect. Huw
Hi,
Have you tried to process in P1? That is my first guess when things don't go as expected. Also, you don't have to feed the MR search with the full protein complex model. You can search individual components and check if they sit well in the electron density. If none of this work, I would be more inclined to believe that you might have crystallized a contaminant or a fragment of your protein of interest.
Best wishes
______________________________________________________
Rafael Marques da Silva
PhD Student – Structural Biology
University of Leicester
Mestre em Física Biomolecular
Universidade de São Paulo
Bacharel em Ciências Biológicas
Universidade Federal de São Carlos
phone: +44 07861 273773
"A sorte acompanha uma mente bem treinada"
________________________________________________
________________________________
De: Huw Jenkins via phenixbb
On 1 Jun 2026, at 08:51, [email protected] wrote:
I have collected a native dataset that extends to 2.45 Å resolution in space group P2₁2₁2 with unit-cell parameters approximately a = 46.9 Å, b = 91.9 Å, c = 165.0 Å.
Just to check - you have been trying MR in all 8 space groups compatible with P222 point group - i.e. P222, P222(1), P22(1)2, P2(1)22, P2(1)2(1)2, P2(1)22(1), P 22(1)2(1) and P2(1)2(1)2(1)? It's not uncommon for assignment of screw axes based on systematic absences to turn out to be incorrect. Huw _______________________________________________ phenixbb mailing list -- [email protected] To unsubscribe send an email to [email protected] Unsubscribe: phenixbb-leave@%(host_name)s
Thank you for the suggestion. I have already explored this possibility using the Phaser option that automatically tests all compatible space groups within the same point group. In my case, the searches were carried out across all orthorhombic space groups compatible with the P222 point group. Unfortunately, the results were very similar across the different space-group assignments, with TFZ values generally remaining in the 6–8 range and no solution leading to convincing maps or improved refinement statistics
participants (8)
-
Clemens Grimm -
Guenter Fritz -
Huw Jenkins -
Kikuti Carlos -
Rafael Marques -
Rebekah E Bjork -
Robert Oeffner -
syedsuhail7654＠gmail.com