TMB_pipeline

This repository contains code for running a generative pipeline for transmembrane β-barrel (TMB) design. It includes scripts demonstrating how to use TMB_RFD2 for backbone generation and TMB_MPNN for sequence design, as well as notebooks describing the curation and filtering of the distillation set used for training.

---

Installation and setup

1. Clone the repository

git clone https://github.com/baker-laboratory/TMB_pipeline.git
cd TMB_pipeline

---

2. Set environment variables and create directories

The example commands in this repository use the following environment variables.

mkdir -p weights environments

export TMB_PIPELINE_DIR="$PWD"
export WEIGHTS_DIR="$TMB_PIPELINE_DIR/weights"
export ENV_DIR="$TMB_PIPELINE_DIR/environments"

You may change the directory names if desired, but make sure the exported paths match your chosen locations.

---

3. Download model weights

Navigate to the weights directory and download the checkpoints.

cd "$WEIGHTS_DIR"

# TMB_RFD2 checkpoint for standard backbone generation
curl -O https://files.ipd.uw.edu/pub/TMB_pipeline/TMB_RFD2/TMB_RFD2.pt

# TMB_RFD2 checkpoint for motif scaffolding
curl -O https://files.ipd.uw.edu/pub/TMB_pipeline/TMB_RFD2/TMB_RFD2_motif_scaffolding.pt

# TMB_MPNN checkpoint for standard sequence design
curl -O https://files.ipd.uw.edu/pub/TMB_pipeline/TMB_MPNN/TMB_MPNN.pth

# TMB_MPNN checkpoint for tuning hydrophobic thickness
curl -O https://files.ipd.uw.edu/pub/TMB_pipeline/TMB_MPNN/TMB_MPNN_tuning_hydrophobic_thickness.pth

Set checkpoint paths.

export TMB_RFD2_CKPT_PATH="$WEIGHTS_DIR/TMB_RFD2.pt"
export TMB_RFD2_MOTIF_SCAFFOLDING_CKPT_PATH="$WEIGHTS_DIR/TMB_RFD2_motif_scaffolding.pt"
export TMB_MPNN_CKPT_PATH="$WEIGHTS_DIR/TMB_MPNN.pth"
export TMB_MPNN_COND_CKPT_PATH="$WEIGHTS_DIR/TMB_MPNN_tuning_hydrophobic_thickness.pth"

---

4. Set up the environment: Singularity / Apptainer

Navigate to the environments directory and download the containers.

cd "$ENV_DIR"

curl -O https://files.ipd.uw.edu/pub/TMB_pipeline/sif_file/modulus_latest.sif
curl -O https://files.ipd.uw.edu/pub/TMB_pipeline/sif_file/TMB_RFD2.sif
curl -O https://files.ipd.uw.edu/pub/TMB_pipeline/sif_file/TMB_MPNN.sif

Set container paths.

export MODULUS_APPTAINER_PATH="$ENV_DIR/modulus_latest.sif"
export TMB_RFD2_APPTAINER_PATH="$ENV_DIR/TMB_RFD2.sif"
export TMB_MPNN_APPTAINER_PATH="$ENV_DIR/TMB_MPNN.sif"

The container variables are used as follows:

# General utilities / examples
apptainer exec "$MODULUS_APPTAINER_PATH" python --version

# TMB_RFD2 backbone generation
apptainer exec "$TMB_RFD2_APPTAINER_PATH" python --version

# TMB_MPNN sequence design
apptainer exec "$TMB_MPNN_APPTAINER_PATH" python --version

---

5. Verify setup

Check that all paths are set correctly.

echo "$TMB_PIPELINE_DIR"
echo "$WEIGHTS_DIR"
echo "$ENV_DIR"

echo "$TMB_RFD2_CKPT_PATH"
echo "$TMB_RFD2_MOTIF_SCAFFOLDING_CKPT_PATH"
echo "$TMB_MPNN_CKPT_PATH"
echo "$TMB_MPNN_COND_CKPT_PATH"

echo "$MODULUS_APPTAINER_PATH"
echo "$TMB_RFD2_APPTAINER_PATH"
echo "$TMB_MPNN_APPTAINER_PATH"

Check that the files exist.

ls "$TMB_RFD2_CKPT_PATH"
ls "$TMB_RFD2_MOTIF_SCAFFOLDING_CKPT_PATH"
ls "$TMB_MPNN_CKPT_PATH"
ls "$TMB_MPNN_COND_CKPT_PATH"

ls "$MODULUS_APPTAINER_PATH"
ls "$TMB_RFD2_APPTAINER_PATH"
ls "$TMB_MPNN_APPTAINER_PATH"

6. Re-export paths in a new terminal session

Environment variables are not permanent unless added to your shell startup file. If you open a new terminal session, rerun:

cd /path/to/TMB_pipeline

export TMB_PIPELINE_DIR="$PWD"
export WEIGHTS_DIR="$TMB_PIPELINE_DIR/weights"
export ENV_DIR="$TMB_PIPELINE_DIR/environments"

export TMB_RFD2_CKPT_PATH="$WEIGHTS_DIR/TMB_RFD2.pt"
export TMB_RFD2_MOTIF_SCAFFOLDING_CKPT_PATH="$WEIGHTS_DIR/TMB_RFD2_motif_scaffolding.pt"
export TMB_MPNN_CKPT_PATH="$WEIGHTS_DIR/TMB_MPNN.pth"
export TMB_MPNN_COND_CKPT_PATH="$WEIGHTS_DIR/TMB_MPNN_tuning_hydrophobic_thickness.pth"

export MODULUS_APPTAINER_PATH="$ENV_DIR/modulus_latest.sif"
export TMB_RFD2_APPTAINER_PATH="$ENV_DIR/TMB_RFD2.sif"
export TMB_MPNN_APPTAINER_PATH="$ENV_DIR/TMB_MPNN.sif"

---

---

Distillation set notebook

The notebook at Distillation_set_curation/distillation_set.ipynb loads the distillation-set silent file from:

• Distillation_set_curation/distillation_set.silent by default, or
• the path provided by the DISTILLATION_SILENT_PATH environment variable.

The silent file is large, so it is intentionally excluded from git. Keep a local copy in the path above, or point the environment variable at your own copy before running the notebook.

This notebook also expects a PyRosetta-enabled environment. It adds TMB_RFD2/TMB_RFD2_code to sys.path automatically so it can reuse the repository's shared silent-file reader.

---

Notes

All assets are hosted at:

https://files.ipd.uw.edu/pub/TMB_pipeline/

This pipeline assumes access to:

• a SLURM cluster
• GPU resources for full-scale runs
• Singularity or Apptainer