Predicting Antimicrobial Resistance (AMR) using DNABERT

This project implements a machine learning model to predict Antimicrobial Resistance (AMR) from genomic sequences.
It utilizes the powerful DNABERT pre-trained language model from Hugging Face as a feature extractor, followed by a custom, dense classifier head for binary prediction.

---

1. Project Goal

The primary objective is to develop a reliable and reusable Python ML model capable of classifying Cefoxitin resistance in the Staphylococcus aureus pbp4 gene based solely on its raw genomic sequence data.

---

2. Model Architecture

The core of the solution is a custom PyTorch class that wraps the feature extraction and classification steps.

• Input: Tokenized DNA sequences (6-mers).
• Feature Extraction: The tokenized sequences are passed through the pre-trained DNABERT model (via Hugging Face Transformers).
• Embedding: The embedding of the [CLS] token is extracted, representing the aggregated features of the entire sequence.
• Classification Head: This embedding is fed into a custom, 4-layer fully connected (linear) network featuring non-linear activations (e.g., ReLU) and a final binary output layer (e.g., using a sigmoid activation for probability output).
  Dropout is utilized for regularization.

---

3. Setup and Installation

Prerequisites

• Python (3.8+)
• PyTorch (recommended backend)
• Git