Neeraj0101/AMP-Predict

AMPPred — Antimicrobial Peptide Prediction Tool

AMPPred is a Python package with a BERT-based deep learning model for predicting antimicrobial peptides (AMPs) from nucleotide sequences. It supports raw nucleotide sequences as well as coding sequences extracted via gene prediction, enabling high-accuracy AMP classification.

Model Overview

• Fine-tuned BERT architecture based on the GENA-LM framework for sequence classification
• Trained on curated AMP datasets from APS (UNMC) and dbAMP (CUHK), with validated negative samples from NCBI
• Achieves 99.05% accuracy on test data
• Supports positional embeddings, sparse attention, and sequence length up to 256 tokens

Features

• Predict whether nucleotide sequences encode antimicrobial peptides
• Optional gene prediction using Pyrodigal to extract coding regions
• Handles FASTA nucleotide inputs with batch processing
• Outputs CSV reports with prediction confidence and optional FASTA of predicted AMPs
• GPU acceleration supported

Usage Example

from amppred.utility.helper import load_model, get_predictions, read_fasta

model, tokenizer = load_model('Neeraj0101/AMP-Predict')
sequence_df = read_fasta('sequences.fasta')
predicted_df, _ = get_predictions(model, tokenizer, sequence_df)

Or via CLI:

amp-build --model Neeraj0101/AMP-Predict
amp-run --input sequences.fasta --get_amp

Citation

If you use this model, please cite:

@misc{amppred2025,
  title={AMPPred: Antimicrobial Peptide Prediction Tool},
  author={Singh, Neeraj Kumar},
  year={2025},
  url={https://github.com/neeraj1014/Antimicrobial-Peptide-Predictor}
}

---