K-mer–Driven Genome-Informed Machine Intelligence for Detecting Bioengineered Pathogens

This repository contains the full implementation of the research project
“K-mer–Driven Genome-Informed Machine Intelligence for Detecting Bioengineered Pathogens.”
It includes preprocessing scripts, trained machine-learning models, notebooks, FASTA test files, and a FastAPI-based web interface for inference.

The system detects whether a plasmid is natural or engineered and predicts its host genus using an interpretable ML pipeline built on 4-mer frequency vectors.

---

Overview

Features

• BLAST-based genus inference for synthetic plasmids
• Automated preprocessing and k-mer feature vectorization
• Binary (Natural vs Engineered) classification
• Genus attribution across eight bacterial taxa
• High-performance Random Forest classifier
• t-SNE visualizations of learned genomic structure
• Reproducible code and saved models
• Deployable FastAPI-based web application

Model Performance

Task	Accuracy	ROC–AUC
Binary (Natural vs Engineered)	98.98%	0.9994
Genus Attribution (8-class)	90.34%	0.9895

---

Project Structure

ML_Bioengineered_Detection/
│
├── models/
│   ├── binary/
│   │   ├── binary_model.joblib
│   │   ├── binary_vectorizer.joblib
│   │   ├── binary_simple_model.joblib
│   │   └── binary_simple_vectorizer.joblib
│   │
│   └── genus/
│       ├── genus_full_model.joblib
│       ├── genus_full_vectorizer.joblib
│       ├── genus_simple_model.joblib
│       └── genus_simple_vectorizer.joblib
│
├── notebooks/
│   ├── ML_ENDV2.ipynb
│   └── ML_ENDV3.ipynb
│
├── test_fasta_files/
│   ├── AB282595.1.fasta
│   ├── AP040173.1.fasta
│   └── DL143694.1.fasta
│
|
├── web_app/
│   ├── test_fasta_files/
│   |   ├── AB282595.1.fasta
│   |   ├── AP040173.1.fasta
│   |   └── DL143694.1.fasta
│   │
│   └── app.py
|
├── requirements.txt
├── README.md
└── LICENSE

---

Installation

Clone the repository

git clone https://github.com/RithvikReddy0-0/ML_Bioengineered_Detection.git
cd ML_Bioengineered_Detection

---

Install dependencies

Ensure you have Python installed, then install the required packages

pip install -r requirements.txt

---

Usage

Running the Web Application

The project includes a web interface built with FastAPI/Uvicorn.

uvicorn app:app --reload

• App Interface: http://127.0.0.1:8000
• API Documentation: http://127.0.0.1:8000/docs

---

CLI Testing

You can test specific FASTA files using the command line:

python app.py --file test_fasta_files/AB282595.1.fasta

Adjust the path to point to any FASTA file you want to analyze.

---

Prediction Pipeline

The end‑to‑end prediction pipeline follows these steps, using standard k‑mer profiling approaches for genomic analysis.

1. FASTA upload

   • The user uploads a nucleotide sequence in FASTA format via the web UI or API.

2. Sequence preprocessing

   • Convert sequence to uppercase.
   • Filter to valid IUPAC nucleotide characters.
   • Optionally truncate sequences longer than 10,000 bp.

3. 4‑mer feature extraction

   • Slide a window of length 4 along the sequence.
   • Compute normalized 4‑mer frequencies, yielding a 256‑dimensional feature vector.

4. Model inference

   • The 4‑mer vector is passed into two classifiers:
     • binary_model.joblib: predicts Natural vs Engineered origin.
     • genus_full_model.joblib: predicts Genus attribution.

5. Outputs

   • Origin prediction (Natural / Engineered) with probability or confidence score.
   • Genus prediction with associated probability or confidence.

Both “simple” and “full” variants of the models and vectorizers are provided in the models/ directory for experimentation and ablation studies.

---

Notebooks

The notebooks/ directory contains Jupyter notebooks that reproduce the data processing and model development workflow.

• ML_ENDV2.ipynb

  • Data collection from public nucleotide repositories.
  • BLAST‑based genus inference and label curation.
  • Preprocessing and k‑mer feature generation.

• ML_ENDV3.ipynb

  • Model training and hyperparameter tuning using scikit‑learn.
  • Evaluation (ROC curves, confusion matrices, and other metrics).
  • Figure generation for the associated research manuscript.

---

Test FASTA Files

Example sequences are included in test_fasta_files/:

• AB282595.1.fasta
• AP040173.1.fasta
• DL143694.1.fasta

You can quickly verify the pipeline with:

python app.py --file test_fasta_files/AB282595.1.fasta

Each file corresponds to a real genomic record sourced from public nucleotide databases such as NCBI.

---

Citation

If you use this code or models in your research, please cite:

@article{bioengineered_detection_2025,
title = {K-mer–Driven Genome-Informed Machine Intelligence for Detecting Bioengineered Pathogens},
author = {Reddy, Mukkara Rithvik and Barshilia, Vasudha and Bhattacharya, Debanjali},
year = {2025}
}

---

License

This project is distributed under the MIT License; see the LICENSE file for details.

The MIT License is a permissive open‑source license that allows reuse, modification, and distribution of the software, provided that the original copyright and license notice are included in copies or substantial portions of the software.

---

Acknowledgements

• NCBI Nucleotide Database for access to curated genomic sequences used in this work.
• PLSDB plasmid repository for plasmid sequence resources.
• Google Colab for providing GPU‑enabled compute resources for training and experiments.
• Scikit‑learn, NumPy, Matplotlib, and other open‑source libraries that underpin the modeling and analysis pipeline.