Initial Model Version — AMR Classifier Using FCGR
This is the first version of our deep learning model developed to classify antimicrobial resistance (AMR) in Escherichia coli using genomic data. The model exclusively uses Frequency Chaos Game Representation (FCGR) matrices derived from whole genome sequences as input.
Key details:
Objective: Binary classification — Resistant vs. Non-resistant
Input: Grayscale FCGR matrices representing k-mer frequency patterns
Architecture: Convolutional Neural Network (CNN), optimized for image-like genomic features
Evaluation: Monitored via loss curves across epochs, accuracy, precision, recall, and AUC
Data Source: Preprocessed genome sequences stored in CSV format, each associated with a resistance label
This version serves as a foundational implementation to validate the predictive power of FCGR-based representations in AMR classification tasks.
Full Changelog: https://github.com/vmlcode/AMR-VSA25/commits/v1.0.0-research