In lab experiment, binding affinity is pretty strong, but pae_interaction is only around 28

[Huilin-Li]

So, I have an example protein A that is very similar to my protein B I aim to study. The protein A has a binder A-binder, and I use A-binder as a scaffold to design my binder B-binder to my protein B.

As I learned here, at the last step, I will use af2_initial_guess/predict.py to assess designed binders and pick binders whose pae_interaction<10 .

However, pae_interaction~=28 between protein A and A-binder. I feel confused about it a lot, because its binding affinity is very strong in lab experiment.

I worked very hard to think of how to improve pae_interaction below 10. Unfortunately, over 95% pae_interaction is around 26~28. Should I still work hard to look for pae_interaction<10 ?

I don't understand these two scenarios: 1) why pae_interaction is bad, but binding affinity is very good in lab experiment? 2) Is pae_interaction<10 very important? Is there any other criteria I should to consider when I assess my designed binders?

Thanks a lot!!!

[PeiyuJianghku]

Dear Huilin-Li,

Whatever I ran, I got the results from 26-28 of pae. Do you mind adding a wechat? We can talk about it together.

[ChengkuiZhao]

So, I have an example protein A that is very similar to my protein B I aim to study. The protein A has a binder A-binder, and I use A-binder as a scaffold to design my binder B-binder to my protein B.

As I learned here, at the last step, I will use af2_initial_guess/predict.py to assess designed binders and pick binders whose pae_interaction<10 .

However, pae_interaction~=28 between protein A and A-binder. I feel confused about it a lot, because its binding affinity is very strong in lab experiment.

I worked very hard to think of how to improve pae_interaction below 10. Unfortunately, over 95% pae_interaction is around 26~28. Should I still work hard to look for pae_interaction<10 ?

I don't understand these two scenarios: 1) why pae_interaction is bad, but binding affinity is very good in lab experiment? 2) Is pae_interaction<10 very important? Is there any other criteria I should to consider when I assess my designed binders?

Thanks a lot!!!

Could you give me your email? I can solve the pae interaction problem for you.

[PeiyuJianghku]

Hi,

Here is my email and my wechat is 17863113955. Please add me and I am strongly willing to discuss with you about the pae thing.

---

From: Chengkui Zhao @.> Sent: Thursday, May 30, 2024 11:04 PM To: RosettaCommons/RFdiffusion @.> Cc: Peiyu Jiang @.>; Comment @.> Subject: Re: [RosettaCommons/RFdiffusion] In lab experiment, binding affinity is pretty strong, but pae_interaction is only around 28 (Issue #232)

So, I have an example protein A that is very similar to my protein B I aim to study. The protein A has a binder A-binder, and I use A-binder as a scaffold to design my binder B-binder to my protein B.

As I learned herehttps://github.com/nrbennet/dl_binder_design, at the last step, I will use af2_initial_guess/predict.py to assess designed binders and pick binders whose pae_interaction<10 .

However, pae_interaction~=28 between protein A and A-binder. I feel confused about it a lot, because its binding affinity is very strong in lab experiment.

I worked very hard to think of how to improve pae_interaction below 10. Unfortunately, over 95% pae_interaction is around 26~28. Should I still work hard to look for pae_interaction<10 ?

I don't understand these two scenarios: 1) why pae_interaction is bad, but binding affinity is very good in lab experiment? 2) Is pae_interaction<10 very important? Is there any other criteria I should to consider when I assess my designed binders?

Thanks a lot!!!

Could you give me your email? I can solve the pae interaction problem for you.

— Reply to this email directly, view it on GitHubhttps://github.com/RosettaCommons/RFdiffusion/issues/232#issuecomment-2141290702, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BIJFHXDDGROBXPJY62H2DGTZFAHHTAVCNFSM6AAAAABG2IYFKSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNBRGI4TANZQGI. You are receiving this because you commented.Message ID: @.***>

[harshagrawal13]

Could you please post an update here in the chat as well? Am doing something similar and would love to know how this turns out.

[Huilin-Li]

Dear all,

Thanks a lot for your reply, and I am sorry for my late reply. Could we discuss our questions publicly rather than privately?

To this question, I think the main question is I wrongly used af_binder_design to assess the binder in lab experiments. There might be something wrong when I use af_initial_guess/predict to re-predict the binder in lab and score it.

[harshagrawal13]

I found in one of the threads that they use the Alphafold Multimer LSI score rather than PAE score for filtering. Have you tried that? https://github.com/RosettaCommons/RFdiffusion/issues/211#issuecomment-2125399851

[Huilin-Li]

Alphafold Multimer LSI score

Thanks! No, I am sorry I didn't try it, but I will try it as soon as possible. Because I can get pae < 10 when I largely improve the number of designed binders (around 10k). So, I followed my pipeline to design binders for a while.

Thanks! This is a good idea. I will have a look on it!

[ChengkuiZhao]

Alphafold Multimer LSI score

Thanks! No, I am sorry I didn't try it, but I will try it as soon as possible. Because I can get pae < 10 when I largely improve the number of designed binders (around 10k). So, I followed my pipeline to design binders for a while.

Thanks! This is a good idea. I will have a look on it!

Hello, Li. Thanks for your reply. I have a question about sequence design in the lab. For the experiment design part, the RF author indicate "MSG-sequence-...", I don't quiet understand what the "MSG" is. If you don't follow the experiment design by RF diffusion, could you please tell me how you do the lab experiment for the computational designed sequence?

The stable lab experiment validation is really important for me. Many thanks and wait for your reply!

[Huilin-Li]

Alphafold Multimer LSI score

Thanks! No, I am sorry I didn't try it, but I will try it as soon as possible. Because I can get pae < 10 when I largely improve the number of designed binders (around 10k). So, I followed my pipeline to design binders for a while. Thanks! This is a good idea. I will have a look on it!

Hello, Li. Thanks for your reply. I have a question about sequence design in the lab. For the experiment design part, the RF author indicate "MSG-sequence-...", I don't quiet understand what the "MSG" is. If you don't follow the experiment design by RF diffusion, could you please tell me how you do the lab experiment for the computational designed sequence?

The stable lab experiment validation is really important for me. Many thanks and wait for your reply!

Hey.

I am not sure whether I understand your question correctly. However, I would like to tell you my workflow. I have a target protein target.pdb, and I aim to create a binder.pdb for it. I use scaffold-based binder design to design binder.pdb, and then my designed binders (~5) will be given to my colleagues who will go to test these binders in the lab experiment (wet-experiment).

[ChengkuiZhao]

Alphafold Multimer LSI score

Thanks! No, I am sorry I didn't try it, but I will try it as soon as possible. Because I can get pae < 10 when I largely improve the number of designed binders (around 10k). So, I followed my pipeline to design binders for a while. Thanks! This is a good idea. I will have a look on it!

Hello, Li. Thanks for your reply. I have a question about sequence design in the lab. For the experiment design part, the RF author indicate "MSG-sequence-...", I don't quiet understand what the "MSG" is. If you don't follow the experiment design by RF diffusion, could you please tell me how you do the lab experiment for the computational designed sequence? The stable lab experiment validation is really important for me. Many thanks and wait for your reply!

Hey.

I am not sure whether I understand your question correctly. However, I would like to tell you my workflow. I have a target protein target.pdb, and I aim to create a binder.pdb for it. I use scaffold-based binder design to design binder.pdb, and then my designed binders (~5) will be given to my colleagues who will go to test these binders in the lab experiment (wet-experiment).

Yeah, Thanks Li. Actually my question is about wet-lab experiment. I followed almost the same workflow as you to design binders (~5), and gave them to my colleagues. While the expression of our protein in wet lab is quiet low. So, could you give me some information about how to express the protein in wet lab? Such as :

1. which cellline you use?
2. what is your expression vector?
3. what's the promoter in the vecter?
4. what is the signal peptide you use? Many thanks for your reply!

[ZhongyanLI0108]

So, I have an example protein A that is very similar to my protein B I aim to study. The protein A has a binder A-binder, and I use A-binder as a scaffold to design my binder B-binder to my protein B. As I learned here, at the last step, I will use af2_initial_guess/predict.py to assess designed binders and pick binders whose pae_interaction<10 . However, pae_interaction~=28 between protein A and A-binder. I feel confused about it a lot, because its binding affinity is very strong in lab experiment. I worked very hard to think of how to improve pae_interaction below 10. Unfortunately, over 95% pae_interaction is around 26~28. Should I still work hard to look for pae_interaction<10 ? I don't understand these two scenarios: 1) why pae_interaction is bad, but binding affinity is very good in lab experiment? 2) Is pae_interaction<10 very important? Is there any other criteria I should to consider when I assess my designed binders? Thanks a lot!!!

Could you give me your email? I can solve the pae interaction problem for you.

I met the same problem. Could I discuss solutions with you? Here is my email: 219059027@link.cuhk.edu.cn. Many thanks!

[Huilin-Li]

Sorry guys, sorry for my late reply. Because I gave up to design the binder in my case. I would like to share more details for you guys. Hope they might be helpful in your cases.

1. I am worried that I might not explain my question clearly. It is not that my designed binder had good wet-lab results. What I was confused was that when I tested the interaction score of a pubislished protein structure (with two sub-proteins) using RFdiffusion, the interaction score is around 26~27. However, this is a pubilished protein with well-tested results in wet-lab experiments. To this confusion, it might be because RFdiffusion used a initial guess structure prediction. (I didn't dig into here.)
2. To my own case, 1) my target protein is also predicted which might have some negative influence when RFdiffusion designed its binders. 2) the number of my deisgns were still too small.

[ZhongyanLI0108]

Sorry guys, sorry for my late reply. Because I gave up to design the binder in my case. I would like to share more details for you guys. Hope they might be helpful in your cases.

1. I am worried that I might not explain my question clearly. It is not that my designed binder had good wet-lab results. What I was confused was that when I tested the interaction score of a pubislished protein structure (with two sub-proteins) using RFdiffusion, the interaction score is around 26~27. However, this is a pubilished protein with well-tested results in wet-lab experiments. To this confusion, it might be because RFdiffusion used a initial guess structure prediction. (I didn't dig into here.)
2. To my own case, 1) my target protein is also predicted which might have some negative influence when RFdiffusion designed its binders. 2) the number of my deisgns were still too small.

Thanks for your kind reply!