TR2-D2 For Multi-Objective Therapeutic Peptide Design 🧫

This part of the code is for finetuning a peptide MDM to optimize multiple therapeutic properties, including binding affinity to a protein target, solubility, non-hemolysis, non-fouling, and cell membrane permeability, with TR2-D2.

The codebase is partially built upon PepTune (Tang et.al, 2024), MDLM (Sahoo et.al, 2023), and MDNS (Zhu et.al, 2025).

Environment Installation

conda env create -f environment.yml

conda activate tr2d2-pep

Model Pretrained Weights Download

Follow the steps below to download the model weights requried for this experiment, which is originally from PepTune.

Download the PepTune pre-trained MDLM and place in /TR2-D2/peptides/pretrained/: https://drive.google.com/file/d/1oXGDpKLNF0KX0ZdOcl1NZj5Czk2lSFUn/view?usp=sharing
Download the pre-trained binding affinity Transformer model and place in /TR2-D2/tr2d2-pep/scoring/functions/classifiers/: https://drive.google.com/file/d/128shlEP_-rYAxPgZRCk_n0HBWVbOYSva/view?usp=sharing

Finetune with TR2-D2

After downloading the pretrained checkpoints, follow the steps below to run fine-tuning:

Fill in the base_path in scoring/scoring_functions.py and diffusion.py.
Fill in HOME_LOC to the base path where TR2-D2 is located and ENV_PATH to the directory where your environment is downloaded in finetune.sh.
Create a path tr2d2-pep/results where the fine-tuning curves and generation results will be saved and tr2d2-pep/checkpoints for checkpoint saving. Also, create tr2d2-pep/logs where the training logs will be saved.
To specify a target protein, set --prot_seq <insert amino acid sequence> and --prot_name <insert protein name>. Default protein is Transferrin receptor (TfR).

Run fine-tuning using nohup with the following commands:

chmod +x finetune.sh

nohup ./finetune.sh > finetune.log 2>&1 &

Evaluation will run automatically after the specified number of fine-tuning epochs --num_epochs is finished. To summarize metrics, fill in path and prot_name in metrics.py and run:

python metrics.py