Protein/Enzyme Design with RFdiffusion and ProteinMPNN

Proteins are designed using RFdiffusion based on given constraints. Sequences are generated using ProteinMPNN and validated using AlphaFold2. Have fun!

New updated repository: https://github.com/lauramarie99/ProtDesign

Overview

• diffuse.py: Diffusion (RFdiffusion or RFdiffusion all-atom)
• validate.py: Validation (ProteinMPNN + AF2)
• configs: Folder containing example config files for diffusion and validation
• run_cluster.py: Slurm script generation and job submission to cluster
• RFdiffusion_dockerfile: Dockerfile for RFdiffusion step
• Validation_GPU_dockerfile: Dockerfile for validation step

Setup

ColabDesign

• Clone this repository
• cd into ColabDesign repository
• Get AF models

mkdir params
aria2c -q -x 16 https://storage.googleapis.com/alphafold/alphafold_params_2022-12-06.tar
tar -xf alphafold_params_2022-12-06.tar -C params
touch params/done.txt

• Build container using Validation_GPU_dockerfile

RFdiffusion

• Clone RFdiffusion repository (https://github.com/lauramarie99/RFdiffusion)
• cd into RFdiffusion repository
• Get RF models

mkdir models && cd models
wget http://files.ipd.uw.edu/pub/RFdiffusion/6f5902ac237024bdd0c176cb93063dc4/Base_ckpt.pt
wget http://files.ipd.uw.edu/pub/RFdiffusion/5532d2e1f3a4738decd58b19d633b3c3/ActiveSite_ckpt.pt

• Build docker/singularity container based on provided dockerfile RFdiffusion_dockerfile
• Adapt path to RFdiffusion in diffuse.py

RFdiffusion all-atom

• Clone RFdiffusion all-atom repository (https://github.com/lauramarie99/rf_diffusion_all_atom)
• cd into RFdiffusion all-atom repository
• Download the model weights

wget http://files.ipd.uw.edu/pub/RF-All-Atom/weights/RFDiffusionAA_paper_weights.pt

• Download the singularity container

wget http://files.ipd.uw.edu/pub/RF-All-Atom/containers/rf_se3_diffusion.sif

• Initialize git submodules

git submodule init
git submodule update

• Adapt path to RFdiffusion all-atom in diffuse.py

Get started

For the diffusion and validation (ProteinMPNN + AF) steps, only one single config file is used. Example config files can be found in the folder configs.

• type: base (RFdiffusion) or all-atom (RFdiffusion all-atom)
• ckpt_override_path: Override RFdiffusion model path (e.g. Active_Site model)
• contigs: Contig string (Specify always a range, e.g. 16-16 instead of 16!)
• enzyme_design: Set true if you want to use an external potential
• guide_potentials: External potential to use (only used if enzyme_design = true)
• guide_scale: Scale factor for guide potential (only used if enzyme_design = true)
• substrate: Substrate name (only used if enzyme_design = true OR type = all-atom)
• iterations: Number of RFdiffusion steps
• name: Experiment name
• noise scale: RFdiffusion noise scale
• num_designs: Number of designs to generate with RFdiffusion
• path: Directory where to store results
• pdb: Input structure (The structure where the fixed residues are taken from)
• num_recycles: Number of AF2 recycles
• num_seqs: Number of ProteinMPNN sequences to generate
• rm_aa: Avoid using specific aa, e.g. cysteines
• use_multimer: Use AF multimer?

Run diffusion

python3.9 diffuse.py --config config.yml

Run validation

python3.8 validate.py --config config.yml

Large scale studies

For generation of many config files based on a general config file, the script create_configs.py in the folder configs can be used. An example general config file is experiment1.yml.

To automatically generate slurm scripts and submit the jobs, the script run_cluster.py can be used. You need to modify the paths for your purposes.

ColabDesign Contributors:

• Sergey Ovchinnikov @sokrypton
• Shihao Feng @JeffSHF
• Justas Dauparas @dauparas
• Weikun.Wu @guyujun (from Levinthal.bio)
• Christopher Frank @chris-kafka