search

fteufel / alphafold-peptide-receptors

Identifying peptide-receptor interactions using AlphaFold-Multimer
https://pubs.acs.org/doi/10.1021/acs.jcim.3c00378
19 stars 6 forks source link

Alphafold-multimer and human_receptors.csv consulting #2

Open Kangfengyuuuu opened 5 months ago

Kangfengyuuuu commented 5 months ago

Could you please provide me with this paper? Because my school does not have permission to read this article. In addition, the human_receptors.csv file contains Topology. Is this necessary? Is it okay if human_receptors.fasta is provided instead? Finally, did you generate the MSA yourself first and then use AFM for prediction? Didn't you directly ask AFM to perform MSA search?Very much looking forward to your reply, thank you!

fteufel commented 5 months ago

Hi,

I'm not sure I'm allowed to just post the paper here. But the findings of the final paper are pretty much the same as our preprint https://www.biorxiv.org/content/10.1101/2022.10.28.514036v1.full

The topology is only used after AlphaFold is done - we discard hits that are predicted to bind on the wrong side of the membrane.

We get the MSAs directly from AF. What we did is split the AF script into MSA generation + prediction, so that we can generate MSAs ahead of time, and reuse MSAs when we have predictions with the same receptor multiple times in different pairings.

Kangfengyuuuu commented 5 months ago

Thank you very much for replying to me. I'm sorry to bother you again, but I still have some questions that I want to inquiry. Your article mentioned that there was no template information used and there was no relaxation. Will this prediction still be accurate? Or what kind of situation is suitable for using only MSA? Also, may I ask how much computing resources are needed and how long will it take for you to do this job? Finally, I have been studying Alphafold-multimer for predicting protein-protein interactions recently and would like to refer to your work in this article. If I want to batch predict pairwise interactions between two protein families (predicting complexes), should I put all the protein sequences of these two protein families together in the input fasta file, or split them into two files? I have already installed Alphafold2 on my computer, and when I run run_alphafold_msaonly.py, I always get errors. So would it be convenient for you to provide your running code? Looking forward to receiving your reply again. Your answer can really help me a lot. Thank you very much！

At 2024-04-22 15:46:26, "felix" @.***> wrote:

Hi,

I'm not sure I'm allowed to just post the paper here. But the findings of the final paper are pretty much the same as our preprint https://www.biorxiv.org/content/10.1101/2022.10.28.514036v1.full

The topology is only used after AlphaFold is done - we discard hits that are predicted to bind on the wrong side of the membrane.

We get the MSAs directly from AF. What we did is split the AF script into MSA generation + prediction, so that we can generate MSAs ahead of time, and reuse MSAs when we have predictions with the same receptor multiple times in different pairings.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

fteufel commented 5 months ago

I normally estimate around 10min per complex using a bit older GPUs.

This codebase was really written around doing few-vs-many screening in AF2.2, using 1 file with a few proteins of interest (=peptides) and a larger file with candidate interactors (=receptors). I'm sure there's other solutions out there for running many-to-many AF predictions.

The code in this repo should work reasonably well with new .fasta and .csv files (that strictly follow the existing format) as inputs, I've been using it recently.