genoarmory 0.1.0

A DNA sequence Adversial attack and defense benchmark

A comprehensive toolkit for DNA sequence Adversarial Attack and Defense Benchmark.

Installation

You can install GenoArmory using pip:

pip install genoarmory

Quick Start

# Initialize model
from GenoArmory import GenoArmory
import json
# You need to initialize GenoArmory with a model and tokenizer.
# For visualization, you don't need a real model/tokenizer, so you can use None if the method doesn't use them.
gen = GenoArmory(model=None, tokenizer=None)
params_file = '/projects/p32013/DNABERT-meta/scripts/PGD/pgd_dnabert.json'

# Visulization
gen.visualization(
    folder_path='/projects/p32013/DNABERT-meta/BERT-Attack/results/meta/test',
    output_pdf_path='/projects/p32013/DNABERT-meta/BERT-Attack/results/meta/test'
)

# Attack
if params_file:
  try:
      with open(params_file, "r") as f:
          kwargs = json.load(f)
  except json.JSONDecodeError as e:
      raise ValueError(f"Invalid JSON in params file '{params_file}': {e}")
  except FileNotFoundError:
      raise FileNotFoundError(f"Params file '{params_file}' not found.")

gen.attack(
    attack_method='pgd',
    model_path='magicslabnu/GERM',
    **kwargs
)

Command Line Usage

GenoArmory can also be used from the command line:

# Attack
python GenoArmory.py --model_path magicslabnu/GERM attack --method pgd --params_file /projects/p32013/DNABERT-meta/scripts/PGD/pgd_dnabert.json

# Defense
python GenoArmory.py --model_path magicslabnu/GERM defense --method at --params_file /projects/p32013/DNABERT-meta/scripts/AT/at_pgd_dnabert.json

# Visualization
python GenoArmory.py --model_path magicslabnu/GERM visualize --folder_path /projects/p32013/DNABERT-meta/BERT-Attack/results/meta/test --save_path /projects/p32013/DNABERT-meta/BERT-Attack/results/meta/test/frequency.pdf


# Read MetaData
python GenoArmory.py --model_path magicslabnu/GERM read --type attack --method TextFooler --model_name dnabert

Features

• Multiple attack methods:

  • BERT-Attack
  • TextFooler
  • PGD
  • FIMBA

• Defense methods:

  • ADFAR
  • FreeLB
  • Traditional Adversarial Training

• Visualization tools

• Artifact management

• Batch processing

• Command-line interface

Documentation

For detailed documentation, visit [docs](We will release soon).

License

This project is licensed under the MIT License.

Citation

If you have any question regarding our paper or codes, please feel free to start an issue.

If you use GenoArmory in your work, please kindly cite our paper: