RosettaCommons
How to redesign interface residues to enable binding to a homolog?
I have a question in protein design. I have a complex consisting of chains A and B. Now, I also have a homolog of chain A, called A'. I want to design a new protein, B', such that A' and B' can bind to each other. Most of the amino acids in B' should remain the same as in B; only the interface residues of B should be changed as needed to enable binding with A'. Essentially, I can only use the structure of B to design a B' that is similar to B, modifying only the interface residues to achieve binding with A'. Is there a method for doing this? I would appreciate any suggestions or advice.
If I'm interpreting your situation correctly, you'll likely need to make the A'+B starting complex somehow. If A and A' are structurally similar enough, it may be sufficient to simply align your crystal structure of A' onto the A in your A+B complex, then save the coordinates of A' and the coordinates of B into a new file. Otherwise you can try naive prediction of the the A'+B complex, either through AI tools, or by classical docking. (One benefit of classical docking like RosettaDock is that you can add restraints/constraints to ensure that the A'+B complex is sufficiently similar to the A+B complex.)
Once you have the starting A'+B complex, with the proper contig map you can diffuse the interface regions of B in the context of the fixed regions of B and the (fixed) A' coordinates.
One limitation this approach will impose is that there will be no ability of RFdiffusion to adjust the fixed portions of B, as there would be with a full de novo design of B'. But you can potentially accommodate for that with an iterative approach. Do your B' design in the context of a fixed orientation, then re-predict the A'+B' complex (with structure prediction AI tools and/or classical docking programs). You now have an update structure of a A'+B' complex, which you can feed back into RFdiffusion to re-build the interface residues (either by full diffusion regeneration, or by partial diffusion purturbation). Repeat however many times you want, or until you get a B' design which fulfills all your design requirements.
(Note that in the iterative process, you want to generate multiple possibilities at each stage -- At each stage (particularly after structure & complex prediction), throw out any designs which don't meet your criteria, or where the structure of the A'+B' complex varies too significantly from the parent A+B complex.)
Thank you very much for your reply; it is extremely helpful for my PhD project! I still have a few detailed questions I'd like to clarify with you:
Since I don't have structural data for A', I used AlphaFold2 to predict its structure, which turned out to be very similar to the structure of A (RMSD = 3.52). Do I still need to run FastRelax on this predicted structure of A', or can I use it directly?
I'm a bit unclear about how to redesign specific residues in protein B from the A'+B complex. Should I use the "Partial Diffusion" or "Binder Design" feature of RFdiffusion for this purpose? Moreover, how should the contigmap.contigs be set accordingly? I'm concerned that if I use the "Partial Diffusion" approach, the redesigned protein B' might lose its binding capability with A'.
Regarding your previous statement: "Note that in the iterative process, you want to generate multiple possibilities at each stage," I’m unsure about the meaning of the term "iterative process." My understanding is that once I've redesigned the interface residues of B in the A'+B complex using RFdiffusion, I then predict the structure using AlphaFold and select the best candidates based on metrics like pae_interaction. I assumed this would be a single-step selection process. Could you clarify where the iterative aspect comes into play here?
Thank you again for your clarification. I look forward to your reply.
If I'm interpreting your situation correctly, you'll likely need to make the A'+B starting complex somehow. If A and A' are structurally similar enough, it may be sufficient to simply align your crystal structure of A' onto the A in your A+B complex, then save the coordinates of A' and the coordinates of B into a new file. Otherwise you can try naive prediction of the the A'+B complex, either through AI tools, or by classical docking. (One benefit of classical docking like RosettaDock is that you can add restraints/constraints to ensure that the A'+B complex is sufficiently similar to the A+B complex.)
Once you have the starting A'+B complex, with the proper contig map you can diffuse the interface regions of B in the context of the fixed regions of B and the (fixed) A' coordinates.
One limitation this approach will impose is that there will be no ability of RFdiffusion to adjust the fixed portions of B, as there would be with a full de novo design of B'. But you can potentially accommodate for that with an iterative approach. Do your B' design in the context of a fixed orientation, then re-predict the A'+B' complex (with structure prediction AI tools and/or classical docking programs). You now have an update structure of a A'+B' complex, which you can feed back into RFdiffusion to re-build the interface residues (either by full diffusion regeneration, or by partial diffusion purturbation). Repeat however many times you want, or until you get a B' design which fulfills all your design requirements.
(Note that in the iterative process, you want to generate multiple possibilities at each stage -- At each stage (particularly after structure & complex prediction), throw out any designs which don't meet your criteria, or where the structure of the A'+B' complex varies too significantly from the parent A+B complex.)
Thank you very much for your reply; it is extremely helpful for my PhD project! I still have a few detailed questions I'd like to clarify with you:
Since I don't have structural data for A', I used AlphaFold2 to predict its structure, which turned out to be very similar to the structure of A (RMSD = 3.52). Do I still need to run FastRelax on this predicted structure of A', or can I use it directly?
I'm a bit unclear about how to redesign specific residues in protein B from the A'+B complex. Should I use the "Partial Diffusion" or "Binder Design" feature of RFdiffusion for this purpose? Moreover, how should the contigmap.contigs be set accordingly? I'm concerned that if I use the "Partial Diffusion" approach, the redesigned protein B' might lose its binding capability with A'.
Regarding your previous statement: "Note that in the iterative process, you want to generate multiple possibilities at each stage," I’m unsure about the meaning of the term "iterative process." My understanding is that once I've redesigned the interface residues of B in the A'+B complex using RFdiffusion, I then predict the structure using AlphaFold and select the best candidates based on metrics like pae_interaction. I assumed this would be a single-step selection process. Could you clarify where the iterative aspect comes into play here?
Thank you again for your clarification. I look forward to your reply.