DeepSS2GO

Introduction

DeepSS2GO is a deep learning predictor th incorporates secondary structure features along with primary sequence and homology information. The algorithm expertly harnesses the speed of sequence-based analysis and the accuracy of structure-based analysis, streamlining primary sequences and elegantly sidestepping the time-consuming challenges of tertiary structure analysis. The results show that the prediction performance surpasses state-of-the-art algorithms. It has the ability to predict key functions by effectively utilizing secondary structure information, rather than broadly predicting general Gene Ontology terms. Additionally, DeepSS2GO predicts five times faster than advanced algorithms, making it highly applicable to massive sequencing data.

Quick taste on --

For Colab Pro users, it is recommended to use DeepSS2GO_v2_colab_pro.ipynb. Just upload the fasta file in the first step, and you can run all cells with one click.
For Colab free users, it is recommended to use DeepSS2GO_v2_colab_free.ipynb. It needs to predict the corresponding secondary structure (8-class) of the primary amino acid sequence on other websites and upload both together.
To be noticed, quick and fully use requires Colab Pro for higher RAM. Details can be found here:

# Colab Pro, GPU=L4, RAM=53GB, GPURAM=22.5GB, Disk=201GB
https://colab.research.google.com/github/orca233/DeepSS2GO_v2_colab/blob/main/DeepSS2GO_v2_colab_pro.ipynb

# Colab Free, GPU=T4, RAM=12.7GB, GPURAM=15GB, Disk=78GB
https://colab.research.google.com/github/orca233/DeepSS2GO_v2_colab/blob/main/DeepSS2GO_v2_colab_free.ipynb

# Details
https://github.com/orca233/DeepSS2GO_v2_colab/

An update free colab version can be find here, with NetSurfP3.0 to predict ss8 from aa. Take a try this one:

https://colab.research.google.com/github/orca233/DeepSS2GO_v2_colab/blob/main/DeepSS2GO_v2_colab_free_NetSurfP3.ipynb

Dependency

DeepSS2GO is developed under Linux environment with:

# Name                    Version                  
blast                     2.5.0                        
click                     8.1.3          
diamond                   2.1.7          
fair-esm                  2.0.0          
matplotlib                3.7.1        
numpy                     1.24.3       
pandas                    1.3.5        
pip                       23.1.2       
python                    3.8.16         
scikit-learn              1.2.2        
scipy                     1.10.1       
seaborn                   0.12.2       
torch                     1.8.0+cu111  
torchaudio                0.8.0        
torchsummary              1.5.1        
torchvision               0.9.0+cu111  
tqdm                      4.65.0       
transformers              4.29.2

Usage

Step 1. Download and setup pre-trained models

Download and setup the following pre-trained models:

For aa to ss8 (step 2):

ESM1b_t33_650M_UR50S

Prot_T5_XL_UniRef50

SPOT1DLM_checkpoints

Simply run, predict BPO/CCO/MFO in a batch step 3.1:

s3_AlphaBeta_bpccmf

For higher precision purpose, predict BPO/CCO/MFO separately step 3.2:

s3_AlphaBeta_TrainALL00_TestALL00_bp_aaK16F32768_ss8K32F32768

s3_AlphaBeta_TrainALL00_TestALL00_cc_aaK16F32768_ss8K48F16384

s3_AlphaBeta_TrainALL00_TestALL00_mf_aaK16F32768_ss8K32F32768

Download links:

# ESM-1b 
# Save to: /home/USERNAME/.cache/torch/hub/checkpoints/esm1b_t33_650M_UR50S-contact-regression.pt & esm1b_t33_650M_UR50S.pt
https://dl.fbaipublicfiles.com/fair-esm/models/esm1b_t33_650M_UR50S.pt

# ProtTrans
# Save to: /home/USERNAME/.../Prot_T5_XL_UniRef50/
https://huggingface.co/Rostlab/prot_t5_xl_uniref50/tree/main
# Modify path_Prot_T5_XL_UniRef50 in step0_DataPreprocessingSetting.py according to previous path
path_Prot_T5_XL_UniRef50 = /home/USERNAME/.../Prot_T5_XL_UniRef50/

# SPOT1DLM_checkpoints.xz
# Unpack and save to: /home/fsong/work/py_proj/prot_algo/DeepSS2GO_v1/pub_data/SPOT1DLM_checkpoints
wget https://huggingface.co/orca233/DeepSS2GO/resolve/main/SPOT-LM-checkpoints.xz

# s3_AlphaBeta_bpccmf/
# s3_AlphaBeta_TrainALL00_TestALL00_bp_aaK16F32768_ss8K32F32768/
# s3_AlphaBeta_TrainALL00_TestALL00_cc_aaK16F32768_ss8K48F16384/
# s3_AlphaBeta_TrainALL00_TestALL00_mf_aaK16F32768_ss8K32F32768/ 
# Unpack and save to: ..../DeepSS2GO/PredictNew/s3_PredictNew_AlphaBeta/
wget https://huggingface.co/orca233/DeepSS2GO/resolve/main/s3_AlphaBeta_bpccmf.tar.gz?download=true
wget https://huggingface.co/orca233/DeepSS2GO/resolve/main/s3_AlphaBeta_TrainALL00_TestALL00_bp_aaK16F32768_ss8K32F32768.tar.gz?download=true
wget https://huggingface.co/orca233/DeepSS2GO/resolve/main/s3_AlphaBeta_TrainALL00_TestALL00_cc_aaK16F32768_ss8K48F16384.tar.gz?download=true
wget https://huggingface.co/orca233/DeepSS2GO/resolve/main/s3_AlphaBeta_TrainALL00_TestALL00_mf_aaK16F32768_ss8K32F32768.tar.gz?download=true

Step 2. Data Preprocessing

In this section, primary amino sequence (aa) will be converted to secondary structure (ss8) by modified SPOT-1D-LM algorithm Ref.

2.1. Save the awating test fasta file to: `/pub_data/data_new/new_aa.fa`

Input fasta file format example:

>slam1
MVIFYFCGKTFMPARNRWMLLLPLLASAAYAEETPREPDLRSRPEFRLHEAEVKPIDREKVPGQVREKGKVLQIDGETLLKNPELLSRAMYSAVVSNNIAGIRVILPIYLQQAQQDKMLALYAQGILAQADGRVKEAISHYRELIAAQPDAPAVRMRLAAALFENRQNEAAADQFDRLKAENLPPQLMEQVELYRKALRERDAWKVNGGFSVTREHNINQAPKRQQYGKWTFPKQVDGTAVNYRLGAEKKWSLKNGWYTTAGGDVSGRVYPGNKKFNDMTAGVSGGIGFADRRKDAGLAVFHERRTYGNDAYSYTNGARLYFNRWQTPKWQTLSSAEWGRLKNTRRARSDNTHLQISNSLVFYRNARQYWMGGLDFYRERNPADRGDNFNRYGLRFAWGQEWGGSGLSSLLRLGAAKRHYEKPGFFSGFKGERRRDKELNTSLSLWHRALHFKGITPRLTLSHRETRSNDVFNEYEKNRAFVEFNKTF
>slam2
MLYFRYGFLVVWCAAGVSAAYGADAPAILDDKALLQVQRSVSDKWAESDWKVENDAPRVVDGDFLLAHPKMLEHSLRDALNGNQADLIASLADLYAKLPDYDAVLYGRARALLAKLAGRPAEAVARYRELHGENAADERILLDLAAAEFDDFRLKSAERHFAEAAKLDLPAPVLENVGRFRKKTEGLTGWRFSGGISPAVNRNANNAAPQYCRQNGGRQICSVSRAERAAGLNYEIEAEKLTPLADNHYLLFRSNIGGTSYYFSKKSAYDDGFGRAYLGWQYKNARQTAGILPFYQVQLSGSDGFDAKTKRVNNRRLPPYMLAHGVGVQLSHTYRPNPGWQFSVALEHYRQRYREQDRAEYNNGRQDGFYVSSAKRLGESATVFGGWQFVRFVPKRETVGGAVNNAAYRRNGVYAGWAQEWRQLGGLNSRVSASYARRNYKGIAAFSTEAQRNREWNVSLALSHDKLSYKGIVPALNYRFGRTESNVPYAKRRNSEVFVSADWRF

In case your protein name contains dot '.', or sequence in multi-line, please run utils_modified_input_fasta.py to uniform input new_aa.fa as the example format.

2.2. Convert primary sequence to secondary structure

Navigate to the directory: ..../DeepSS2GO/PredictNew/s1_DataPreprocessing_PredictNew/

Execute steps 1-8 in s1_DataPreprocessing_New/. The following final files will be generated in /pub_data/data_new/:

new_clean_aa.pkl
new_clean_aa.fa
new_clean_ss8.pkl
new_clean_ss8.fa

Step 3. Prediction

If you require general predictions, run only step 3.1. For higher precision, proceed with step 3.2 instead.

Step 3.1. Simply run, predict BPO/CCO/MFO in a batch:

Navigate to the directory and run: ..../DeepSS2GO/PredictNew/s3_PredictNew_AlphaBeta/s3_AlphaBeta_bpccmf/

# Modify (step6_cpData_Diamond4New.sh) with your own path
path_base="/home/USERNAME/work/py_proj/prot_algo/DeepSS2GO/"

bash step6_cpData_Diamond4New.sh  # Copy these four *pkl/fa files to the corresponding directories and run diamond
bash step7_PredictAlphaBeta_New.sh  # Set the threshold accordingly

Find Results in directory: /data/ as:

results_bp.csv
results_cc.csv
results_mf.csv

Step 3.2. For higher precision purpose, predict BPO/CCO/MFO, separately:

Take BPO as example: Navigate to the directory and perform the same steps as step 3.1: ..../DeepSS2GO/PredictNew/s3_PredictNew_AlphaBeta/s3_AlphaBeta_TrainALL00_TestALL00_bp_aaK16F32768_ss8K32F32768/

Same for CCO and MFO.

s3_AlphaBeta_TrainALL00_TestALL00_bp_aaK16F32768_ss8K32F32768/
s3_AlphaBeta_TrainALL00_TestALL00_cc_aaK16F32768_ss8K48F16384/
s3_AlphaBeta_TrainALL00_TestALL00_mf_aaK16F32768_ss8K32F32768/

Citation and Contact

Citation

The paper has been published by Briefings in Bioinformatics.

https://academic.oup.com/bib/article/25/3/bbae196/7663430

Please cite it as follows:

@article{song2024deepss2go,
  title={DeepSS2GO: protein function prediction from secondary structure},
  author={Song, Fu V and Su, Jiaqi and Huang, Sixing and Zhang, Neng and Li, Kaiyue and Ni, Ming and Liao, Maofu},
  journal={Briefings in Bioinformatics},
  volume={25},
  number={3},
  year={2024},
  publisher={Oxford University Press}
  doi = {10.1093/bib/bbae196},
  url = {https://doi.org/10.1093/bib/bbae196},
}

Contact:

Fu Song (songf@mail.sustech.edu.cn)

Ming Ni (niming@mgi-tech.com)

Maofu Liao (liaomf@sustech.edu.cn)

orca233 / DeepSS2GO

readme

DeepSS2GO

Introduction

Quick taste on --

Dependency

Usage

Step 1. Download and setup pre-trained models

Step 2. Data Preprocessing

2.1. Save the awating test fasta file to: `/pub_data/data_new/new_aa.fa`

2.2. Convert primary sequence to secondary structure

Step 3. Prediction

Step 3.1. Simply run, predict BPO/CCO/MFO in a batch:

Step 3.2. For higher precision purpose, predict BPO/CCO/MFO, separately:

Citation and Contact

Citation

Contact:

orca233 / DeepSS2GO

readme

DeepSS2GO

Introduction

Quick taste on --

Dependency

Usage

Step 1. Download and setup pre-trained models

Step 2. Data Preprocessing

2.1. Save the awating test fasta file to: /pub_data/data_new/new_aa.fa

2.2. Convert primary sequence to secondary structure

Step 3. Prediction

Step 3.1. Simply run, predict BPO/CCO/MFO in a batch:

Step 3.2. For higher precision purpose, predict BPO/CCO/MFO, separately:

Citation and Contact

Citation

Contact:

2.1. Save the awating test fasta file to: `/pub_data/data_new/new_aa.fa`