SAINT-Angle

This repository contains the standalone version (command line interface/CLI tool) of the method proposed in our paper titled SAINT-Angle: self-attention augmented inception-inside-inception network and transfer learning improve protein backbone torsion angle prediction, published in the journal Bioinformatics Advances (Oxford Academic).

If any content in this repository is used for research or other purpose, we shall be obliged if you kindly cite our paper.

Guidelines

Getting Started

In order to run SAINT-Angle on your local machine, navigate to a workspace directory and run the following commands in a terminal.

git clone https://github.com/bayzidlab/SAINT-Angle.git
cd SAINT-Angle/saint_angle

Then, create a separate Conda environment for running SAINT-Angle with Python version 3.7. Install TensforFlow version 2.6.0, Keras version 2.6.0, and pandas version 1.3.5. We tested our method with aforementioned versions of TensorFlow and Keras. We recommend you install GPU version of TensorFlow for faster inference. Besides, run the following command in a terminal to install SentencePiece and transformers.

pip install -q SentencePiece transformers

Downloading Pretrained Model Weights

Pretrained model weights essential for running SAINT-Angle can be found here. Download these model weights files and place them in a folder named models inside SAINT-Angle/saint_angle directory. There are 8 (eight) model weights files with .h5 extension, each of which corresponding to one of the base models used in ensemble of SAINT-Angle. A short description of each base model is provided in the below table.

Model Name	Architecture	Features
model_1	Basic	Base
model_2	Basic	Base, Window10
model_3	ProtTrans	Base, ProtTrans
model_4	ProtTrans	Base, Window10, ProtTrans
model_5	ProtTrans	Base, Window20, ProtTrans
model_6	ProtTrans	Base, Window50, ProtTrans
model_7	Residual	Base, ProtTrans
model_8	Residual	Base, Window10, ProtTrans

For details on different architectures and features, we request you to go through our paper.

Preparing Input Features

Follow any of the following 2 (two) options to prepare input proteins list and corresponding feature files before running SAINT-Angle.

Option-1

List the name of input proteins in a file named proteins_list and place this file inside SAINT-Angle/saint_angle directory. Besides, place the .fasta, .pssm, .hhm, and .spotcon files of each input protein in a folder named inputs located inside SAINT-Angle/saint_angle directory. A .spotcon file should contain the contact map of an input protein in CASP format.

Option-2

If you already have a text file containing a list of names of input proteins (one name in each line) and a folder containing the .fasta, .pssm, .hhm, and .spotcon files of each input protein, you can simply provide the corresponding paths when running SAINT-Angle (discussed in next section).

Keep in mind that, some of the base models of SAINT-Angle require contact map (window) features of input proteins to predict their torsion φ and ψ angles. If you do not provide spotcon file for a particular input protein, then SAINT-Angle will use, in its ensemble, only the base models that do not require window features, in order to predict its torsion angles.

Running SAINT-Angle

In order to run SAINT-Angle for protein backbone φ and ψ angles prediction, run one of the following commands in a terminal from SAINT-Angle/saint_angle directory.

To see available command line parameters

python saint_angle.py -h OR python saint_angle.py --help

When input features prepared following Option-1

python saint_angle.py

When input features prepared following Option-2

python saint_angle.py --list PATH_TO_PROTEINS_LIST --inputs PATH_TO_INPUT_FILES_FOLDER

Understanding Output Files

SAINT-Angle will generate a folder named outputs inside SAINT-Angle/saint_angle directory (if not specified otherwise via command line argument) and place all the output files in this folder. For each input protein, you can expect either 4 (four) or 9 (nine) output .csv files inside the output folder. A short description of each output file format is provided in the below table.

For Protein without .spotcon File
Output File Format	Model Used for Prediction
PROTEIN_NAME.model_1.csv	model_1
PROTEIN_NAME.model_3.csv	model_3
PROTEIN_NAME.model_7.csv	model_7
PROTEIN_NAME.ensemble3.csv	ensemble of above 3 models
For Protein with .spotcon File
Output File Format	Model Used for Prediction
PROTEIN_NAME.model_1.csv	model_1
PROTEIN_NAME.model_2.csv	model_2
PROTEIN_NAME.model_3.csv	model_3
PROTEIN_NAME.model_4.csv	model_4
PROTEIN_NAME.model_5.csv	model_5
PROTEIN_NAME.model_6.csv	model_6
PROTEIN_NAME.model_7.csv	model_7
PROTEIN_NAME.model_8.csv	model_8
PROTEIN_NAME.ensemble8.csv	ensemble of above 8 models

The φ angle of first amino acid residue and the ψ angle of last amino acid residue for each input protein are masked with 360° angle.

Data Availability

Backbone torsion φ and ψ angles of proteins from a recent benchmark dataset, used to evaluate the performance of SAINT-Angle, can be found here.

Citation

A K M Mehedi Hasan, Ajmain Yasar Ahmed, Sazan Mahbub, M Saifur Rahman, and Md Shamsuzzoha Bayzid, SAINT-Angle: self-attention augmented inception-inside-inception network and transfer learning improve protein backbone torsion angle prediction, Bioinformatics Advances, Vol. 3(1): vbad042, https://doi.org/10.1093/bioadv/vbad042, 2023.

BibTeX

@article{hasan2023saint,
  title={SAINT-Angle: self-attention augmented inception-inside-inception network and transfer learning improve protein backbone torsion angle prediction},
  author={Hasan, AKM Mehedi and Ahmed, Ajmain Yasar and Mahbub, Sazan and Rahman, M Saifur and Bayzid, Md Shamsuzzoha},
  journal={Bioinformatics Advances},
  volume={3},
  number={1},
  pages={vbad042},
  year={2023},
  publisher={Oxford University Press}
}