DPO-LLPS

This repository contains the implementation code for our research paper, [DPO-LLPS: Biologically-informed hierarchical transfer learning Strategy for Designing Phase Separation–Driving Proteins]

Introduction

The framework combines hierarchical transfer learning and generative modeling for LLPS protein design. Localization fine-tuning encodes compartment-specific “chemical grammar,” while DPO captures LLPS-driving “molecular grammar.” The model generates novel proteins with targeted localization, tunable phase behavior, and validated condensate stability, enabling programmable and mechanistically interpretable LLPS design.

Installation

Create and activate conda environment

conda env create -f env.yml
conda activate DPO-LLPS

Usage

Model Checkpoints and data

Model checkpoints and complete datasets are available on Zenodo.

Train

To train a model with the default configuration (e.g., DPO-LLPS), simply run:

python train_llps_dpo_multigpu.py

Inference

To perform inference with the default configuration (e.g., DPO-LLPS), run:

python LLPS-DPO-inference.py

Data process

The processing procedures for all datasets can be found in the data_process folder.

Citation

If you find the models useful in your research, please cite our paper.

Contact

If you have any question, please feel free to email us (yangyangzhang@zju.edu.cn).