ANTIOXIPRED

A method for predicting antioxidant potential in peptides

Introduction

AntioxiPred is a computational framework designed to predict the antioxidant potential of peptide sequences with high accuracy using machine learning approaches. The method integrates a similarity-based search using the Basic Local Alignment Search Tool (BLAST) with a CatBoost(Categorical Boosting) classifier. The predictive model is trained using composition-based features, including Pseudo Amino Acid Composition (PAAC) and one-hot encoded sequence profiles, enabling the capture of both global compositional characteristics and positional sequence information.

Requirements

scikit-learn=1.6.1

Pandas

Numpy

Joblib

You can set up the environment using either requirements.txt (for pip users) or environment.yml (for Conda users).

Using requirements.txt

pip install -r requirements.txt

Using environment.yml

conda env create -f env.yml

No additional package/tool is required for model = 1 (default model)

For the hybrid prediction mode (Model 2) in AntioxiPred requires the NCBI BLAST+ software.

BLAST executables are platform-specific. Therefore, users must download the appropriate version of NCBI BLAST+ from the official NCBI website:

https://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/

Minimum USAGE

To know about the available option for the standlone, type the following command:

antioxipred -h

To run the example, type the following command:

antioxipred -f AOPP.test.2023.fasta -o output

Here, -f argument is to enter the input file in Fasta format and -o argument is for giving the path to the output directory. By default, the package uses model (-m) = 1 which employs only ML algorithm (Categorical Boosting) to classify the peptide sequences, which generates a prediction file "classification_ml_(datetime).csv" in the specified output directory. If model (-m) = 2 is selected, then the hybrid model is employed (ML + BLAST) to classify the peptide sequences, which generates a prediction file "classification_hybrid(datetime).csv" in the specified output directory.

Full Usage

usage: antioxipred [-h] --file FILE --output OUTPUT [--model MODEL] [--threshold THRESHOLD]
Please provide following arguments for successful run
required arguments:
  --file FILE, -f FILE                   Path to fasta file
  --output OUTPUT, -o OUTPUT             Path to output

optional arguments:

  --model MODEL, -m MODEL                Model selection: 1 for ML only, 2 for ML + BLAST (By default model = 1)
  --threshold THRESHOLD, -t THRESHOLD    Threshold for classification (can be any value between 0-1 for model = 1 (by default = 0.5) and 0-2 for model = 2 (by default = 0.52))

For help:
  -h, --help            show this help message and exit

Standalone Minimum Usage

Run the program using:

python3 antioxipred.py -f AOPP.test.2023.fasta -o result

---

Arguments Description

Input File (-f)

Allows users to provide input peptide sequences in FASTA format.

Output File (-o)

The program saves the prediction results in the specified output folder.

Model (-m)

Users can choose which model to run:

• model = 1 → Runs only the Machine Learning model (CatBoost classifier)
• model = 2 → Runs the Hybrid model (Machine Learning + BLAST)

Default: model = 1

Threshold (-t)

Users can provide a threshold for classification.

• For model = 1 → threshold range: 0 – 1 (default = 0.50)
• For model = 2 → threshold range: 0 – 2 (default = 0.50)

---

ANTIOXIPRED Package Files

The package contains the following files:

File	Description
INSTALLATION	Installation instructions
README.md	Documentation and usage instructions
catboost_model_server.pkl	Pickled CatBoost prediction model
antioxipred.py	Main Python script used to run the prediction
pfeature_comp.py	Python script used to extract the feature PAAC
AOPP.test.2023.fasta	Example FASTA file containing peptide sequences
blast_db/	Database used for BLAST similarity search