ampidentifier 1.0.1

A Python toolkit for Antimicrobial Peptide (AMP) prediction using ensemble machine learning

AMPidentifier

A Python toolkit for Antimicrobial Peptide (AMP) prediction and physicochemical assessment

////////////////////////////////////////////////////////////////////////
//                                                                    //
//      _    __  __ ____  _     _            _   _  __ _              //
//     / \  |  \/  |  _ \(_) __| | ___ _ __ | |_(_)/ _(_) ___ _ __    //
//    / _ \ | |\/| | |_) | |/ _` |/ _ \ '_ \| __| | |_| |/ _ \ '__|   //
//   / ___ \| |  | |  __/| | (_| |  __/ | | | |_| |  _| |  __/ |      //
//  /_/   \_\_|  |_|_|   |_|\__,_|\___|_| |_|\__|_|_| |_|\___|_|      //
//                                                                    //
////////////////////////////////////////////////////////////////////////

---

About

AMPidentifier is an open-source, modular Python toolkit for predicting Antimicrobial Peptides (AMPs) from amino acid sequences. It combines three pre-trained Machine Learning models (Random Forest, SVM, Gradient Boosting) with an ensemble voting system, and computes dozens of physicochemical descriptors via modlamp.

Users can run predictions with the built-in models, combine them in ensemble mode, or integrate external .pkl models for side-by-side comparison.

---

Related Projects

Project	Description	Link
AMPidentifier CLI	Full command-line version with training scripts, benchmarking, and extended documentation	github.com/madsondeluna/AMPidentifier
AMPidentifier Web Server	Browser-based interface for AMP prediction (no installation required)	github.com/madsondeluna/AMPidentifierServerBETA

---

Installation

pip install ampidentifier

We recommend using a virtual environment:

python3 -m venv venv
source venv/bin/activate   # macOS/Linux
# venv\Scripts\activate    # Windows
pip install ampidentifier

Available on PyPI: https://pypi.org/project/ampidentifier/

---

Quick Start

# Single model (Random Forest, default)
ampidentifier --input my_sequences.fasta --output_dir ./results

# Ensemble voting (recommended)
ampidentifier --input my_sequences.fasta --output_dir ./results --ensemble

# Compare SVM with an external model
ampidentifier --input my_sequences.fasta --output_dir ./results --model svm --external_models /path/to/my_model.pkl

---

Usage Examples

The examples below use this sample FASTA file (test_peptides.fasta) — it contains known AMPs and non-AMP peptides for demonstration:

>Magainin-2|Xenopus_laevis|Cationic_amphipathic_helix
GIGKFLHSAKKFGKAFVGEIMNS
>LL-37|Homo_sapiens|Cathelicidin_family
LLGDFFRKSKEKIGKEFKRIVQRIKDFLRNLVPRTES
>Melittin|Apis_mellifera|Venom_peptide
GIGAVLKVLTTGLPALISWIKRKRQQ
>Insulin_Chain_B|Homo_sapiens|Peptide_hormone
FVNQHLCGSHLVEALYLVCGERGFFYTPKT
>Glucagon|Homo_sapiens|Peptide_hormone
HSQGTFTSDYSKYLDSRRAQDFVQWLMNT
>Vasoactive_intestinal_peptide|Homo_sapiens|Neuropeptide
HSDAVFTDNYTRLRKQMAVKKYLNSILN

---

Usage Example — Google Colab / Jupyter Notebook

Click the badge to open directly in Colab:

Or run the cells below manually in any Colab notebook:

# Cell 1: Install
!pip install ampidentifier

# Cell 2: Create the example FASTA file
fasta_content = """>Magainin-2|Xenopus_laevis|Cationic_amphipathic_helix
GIGKFLHSAKKFGKAFVGEIMNS
>LL-37|Homo_sapiens|Cathelicidin_family
LLGDFFRKSKEKIGKEFKRIVQRIKDFLRNLVPRTES
>Melittin|Apis_mellifera|Venom_peptide
GIGAVLKVLTTGLPALISWIKRKRQQ
>Insulin_Chain_B|Homo_sapiens|Peptide_hormone
FVNQHLCGSHLVEALYLVCGERGFFYTPKT
>Glucagon|Homo_sapiens|Peptide_hormone
HSQGTFTSDYSKYLDSRRAQDFVQWLMNT
>Vasoactive_intestinal_peptide|Homo_sapiens|Neuropeptide
HSDAVFTDNYTRLRKQMAVKKYLNSILN
"""

with open("test_peptides.fasta", "w") as f:
    f.write(fasta_content)

print("FASTA file created with 6 sequences (3 known AMPs + 3 non-AMPs)")

# Cell 3: Run with default model (Random Forest)
# Import the pipeline function directly from the package
import os
from amp_identifier.core import run_prediction_pipeline

os.makedirs("./results_rf", exist_ok=True)

run_prediction_pipeline(
    input_file="test_peptides.fasta",
    output_dir="./results_rf",
    internal_model_type="rf",   # Random Forest: best single-model AUC-ROC (0.9503)
    use_ensemble=False,
    external_model_paths=[],
)

# Cell 4: Run with ensemble mode (recommended)
# Combines RF + SVM + GB via majority voting for maximum robustness
import os
from amp_identifier.core import run_prediction_pipeline

os.makedirs("./results_ensemble", exist_ok=True)

run_prediction_pipeline(
    input_file="test_peptides.fasta",
    output_dir="./results_ensemble",
    internal_model_type="rf",   # ignored when use_ensemble=True
    use_ensemble=True,          # activates majority vote across all three models
    external_model_paths=[],
)

# Cell 5: Inspect results
# Runs ensemble first if output does not exist yet, then displays results
import os
import pandas as pd
from amp_identifier.core import run_prediction_pipeline

report_path   = "./results_ensemble/prediction_comparison_report.csv"
features_path = "./results_ensemble/physicochemical_features.csv"

if not os.path.exists(report_path):
    os.makedirs("./results_ensemble", exist_ok=True)
    run_prediction_pipeline(
        input_file="test_peptides.fasta",
        output_dir="./results_ensemble",
        internal_model_type="rf",
        use_ensemble=True,
        external_model_paths=[],
    )

report = pd.read_csv(report_path)
print("=== Ensemble Prediction Report ===")
print(report.to_string(index=False))

features = pd.read_csv(features_path)
print(f"\n=== Physicochemical Features ===")
print(f"Shape: {features.shape[0]} sequences x {features.shape[1]} descriptors")
print(features[['ID', 'Length', 'Charge', 'HydrophRatio']].to_string(index=False))

# Cell 6: Compare all three internal models individually
import os
import pandas as pd
from amp_identifier.core import run_prediction_pipeline

for model in ["rf", "svm", "gb"]:
    os.makedirs(f"./results_{model}", exist_ok=True)

    run_prediction_pipeline(
        input_file="test_peptides.fasta",
        output_dir=f"./results_{model}",
        internal_model_type=model,
        use_ensemble=False,
        external_model_paths=[],
    )

    report = pd.read_csv(f"./results_{model}/prediction_comparison_report.csv")
    pred_col = [c for c in report.columns if c.startswith("pred_")][0]
    amp_count = int(report[pred_col].sum())
    print(f"[{model.upper()}] Predicted AMPs: {amp_count}/6")

---

Arguments

Argument	Description	Required	Default
-i, --input	Path to the input FASTA file	Yes	-
-o, --output_dir	Path to the output directory	Yes	-
-m, --model	Internal model to use: rf, svm, gb	No	rf
--ensemble	Enable majority-vote ensemble across all internal models	No	Flag
-e, --external_models	One or more paths to external .pkl models for comparison (comma-separated)	No	-

---

Key Features

• Three pre-trained ML models: Random Forest, Gradient Boosting, SVM
• Ensemble voting: Majority vote across all models for improved robustness
• External model support: Load custom .pkl models for comparison
• Physicochemical descriptors: Compute and export an extensive set of sequence features via modlamp
• Fully open-source and modular: Each component can be used independently

---

Pre-Trained Model Performance

Best values per metric in bold.

Metric	Random Forest (RF)	SVM	Gradient Boosting (GB)
Accuracy	0.8845	0.8740	0.8585
Precision	0.8910	0.8880	0.8665
Recall	0.8762	0.8558	0.8475
F1-Score	0.8836	0.8716	0.8569
MCC	0.7692	0.7484	0.7172
AUC-ROC	0.9503	0.9356	0.9289

Recommended: use --ensemble for most robust predictions (Accuracy: 87.47%, Sensitivity: 85.96%, Specificity: 88.98%).

---

Outputs

File	Description
physicochemical_features.csv	Computed physicochemical descriptors for each input sequence
prediction_comparison_report.csv	AMP/non-AMP predictions with confidence scores per model and consensus

---

Project Structure

amp_identifier/
├── __init__.py
├── core.py               # Main prediction workflow
├── data_io.py            # FASTA input reader
├── feature_extraction.py # Physicochemical descriptor computation
├── prediction.py         # Model loading and inference
└── reporting.py          # CSV report generation

---

Contributors

Name	Role	Affiliation
Madson A. de Luna-Aragão, MSc	Lead developer; architecture; ML; docs	UFMG
Rafael L. da Silva, BSc	Collaborator; preprocessing; pipeline testing	UFPE
Ana M. Benko-Iseppon, PhD	Advisor; study design; biological validation	UFPE
João Pacífico, PhD	Co-Advisor; computational review; evaluation	UPE
Carlos A. dos Santos-Silva, PhD	Co-Advisor; pipeline testing; review	CESMAC

---

Funding & Acknowledgments

• Officially registered under UFPE - Universidade Federal de Pernambuco, Brazil
• Supported by FACEPE - Fundação de Amparo à Pesquisa do Estado de Pernambuco
• INPI Registration: BR 51 2025 005859-4

---

How to Cite

Luna-Aragão, M. A., da Silva, R. L., Pacífico, J., Santos-Silva, C. A. & Benko-Iseppon, A. M.
(2025). AMPidentifier: A Python toolkit for predicting antimicrobial peptides using ensemble
machine learning and physicochemical descriptors.
https://github.com/madsondeluna/AMPidentifier

---

License

This project is licensed under the terms specified in the repository. All rights reserved. © Madson A. de Luna Aragão et al., 2025.