We introduced PolypeptideDesigner (PPD) model, a conditional text generation model that utilizes per-residue secondary structure conditions to de novo design long polypeptide sequences up to 250 residues in length. Our PPD model combines diffusion-based models with a lightweight LSTM-attention neural network denoiser. PPD can produce more diverse sequences due to its innovative architecture and efficient denoiser, reflecting an enhanced understanding of protein structures.
Python 3.8.12
PyTorch 2.0.1
TensorFlow 2.11.0
The early stage and final stage parameters of PPD models are hosted on Google Drive.
@article{Liao2024PPD,
title={De Novo Designing of Large Polypeptide Using a Lightweight LSTM and Attention-based Diffusion Model with Per-residue Secondary Structure Condition},
author={Liao, Sisheng and Xu, Gang and Li, Jin and Ma, Jianpeng},
year={2024},
}