Repository - pssm-distil

登录注册

版本库

查看Git使用指南如操作版本库需要认证，请使用您的邮箱

ZIP下载

克隆网址复制

wangqin4377@qq.com 提交于 12 months前：Update README.md

DataLoader
a3ms
bert_pssms
configs
enhanced_pssms
img
labels
low_pssms
models
module
sequences
.gitignore
README.md
inference_our.py
inference_real.py
sample_msa_from_pssm.py

导出统计结果：各用户提交统计信息

# PSSM-Distil
The paper "PSSM-Distil: Protein Secondary Structure Prediction (PSSP) on Low-QualityPSSM by Knowledge Distillation with Contrastive Learning" of AAAI21 Conference

> Requirement
>
>
    pip install torch
    pip install glob
    pip install tqdm
    pip install numpy

> Instructions

python inference_real.py
    
Aboving command will predict secondary structure for a sequence in 'sequences' folder with a pssm in 'low_pssm' folder and print the accuracy.

python inference_our.py
    
Aboving command will predict secondary structure of sequence in 'sequences' folder with enhanced pssm which refined by PSSM-Distil and print the accuracy. 
Besides, this command will save an enhanced pssm file in 'enhanced_pssms' folder.

python sample_msa_from_pssm.py ./enhanced_pssms/4ynhA.npy
    
Aboving command will sample 2000 MSAs from enhanced PSSM and save in a3ms folder as '4ynhA_enhanced_pssms.a3m'.

> Visualization

Please upload original low-quality .a3m file and the enhanced one in 'a3ms' folder to the website: https://weblogo.berkeley.edu/logo.cgi respectively.
Then you will see such comparison images.

> BC40 dataset

https://drive.google.com/drive/folders/15vwRo_OjAkhhwfjDk6-YoKGf4JzZXIMC?usp=sharing

The dataset we released to examine the performance of PSSM-Distil

ihub@pcl.ac.cn 鹏城实验室人工智能研究中心