chun-he-316 / DeepAlgPro

DeepAlgPro: an interpretable deep neural network model for predicting allergenic proteins

Introduction

Allergies have become an emerging public health problem worldwide.It is critical to evaluate potential allergens, especially today when the number of modified proteins in food, therapeutic drugs, and biopharmaceuticals is increasing rapidly. Here, we proposed a software, called DeepAlgPro, that combined a convolutional neural network (CNN) with Multi-Headed Self-Attention (MHSA) and was suitable for large-scale prediction of allergens.

Requirements

• Platform requirement
  We trained the model under linux OS, but it can also be run under windows, mac OS. Your operating system must be supported by the deep learning framework and related libraries you used to use this model. For example, our model was implemented in Pytorch 1.12.1, you must check its OS compatibility list here to ensure that your OS (e.g., Ubuntu, Windows, macOS) is supported.
• Device requirement
  This model was trained on NVIDIA GeForce RTX 3090. When using it, it is supported to run under both GPU and CPU. When the GPU is not available(torch.cuda.is_available()=False), the model will run using the CPU.
• Packages requirement
  
  • python 3.9
    
  • Bio==1.5.3
    
  • numpy==1.23.4
    
  • pandas==1.5.0
    
  • scikit_learn==1.2.1
    
  • torch==1.12.1+cu116
    
  • torchmetrics==0.9.3
    

Installation

1. Download DeepAlgPro

git clone https://github.com/chun-he-316/DeepAlgPro.git

2. Install required packages
   

pip3 install torch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116
pip install -r requirements.txt

Train and Test the model

usage: python main.py [-h] [-i INPUTS] [--epochs N] [--lr LR] [-b N] [--mode {train,test}]

Optional arguments

  -h, --help                     show this help message and exit
  -i INPUTS, --inputs INPUTS
  --epochs N                     number of total epochs to run
  --lr LR, --learning-rate LR    learning rate
  -b N, --batch-size N
  --mode {train,test}

Example

python main.py -i data/all.train.fasta --epochs 120 --lr 0.0001 -b 72 --mode train

Output files

The training process generates a number of files. The types of files are listed below.

• train.log file: Record loss values for each batch.
• .everyepoch.valid.txt files: Record the validation results after each training epoch.
• .pt file: The model obtained by training.
• valid.log: Results of 10-fold cross-validation.

Use DeepAlgPro to predict allergens

usage: python predict.py [-h] [-i INPUTS] [-b N] [-o OUTPUT]

Optional arguments

  -h, --help                    show this help message and exit
  -i INPUTS, --inputs INPUTS    input file
  -b N, --batch-size N
  -o OUTPUT, --output OUTPUT    output file

Input files

The input file specified by -i is a protein sequence file; each sequence has a unique id and starts with >. The input protein sequence number must be divisible by the batch size.

Example

python predict.py -i data/all.test.fasta -o allergen.predict.txt

Output files

The default result file is allergenic_predict.txt, a file with tabs as spacers. You can also specify the output file with -o. The first column in the output file is the id of the input protein, the second column is the score between 0 and 1 predicted by the model, and the third column value is the predicted result, allergenicity or non-allergenicity.For example,

        protein scores  predict result
protein_1   0.9983819723129272      allergenicity
protein_2   0.999177873134613       allergenicity
protein_3   0.000125454544823       non-allergenicity
protein_4   0.9991099238395691      allergenicity

Citation

He C, Ye X, Yang Y, Hu L, Si Y, Zhao X, Chen L, Fang Q, Wei Y, Wu F, Ye G. DeepAlgPro: an interpretable deep neural network model for predicting allergenic proteins. Brief Bioinform. 2023 Jul 20;24(4):bbad246. doi: 10.1093/bib/bbad246.