Trained Deep Learning Models for Antimicrobial Resistance Prediction from MALDI-TOF Mass Spectrometry This archive contains trained neural network models for predicting antibiotic resistance from MALDI-TOF mass spectrometry data across three clinically important bacterial pathogens. Models Included (~6-7 GB) (Just Download rar files only) E. coli-Ceftriaxone Models: Baseline model (MSDeepAMR architecture) Attention-enhanced model (Squeeze-and-Excitation blocks) Hyperparameter-optimized model 5-model ensemble (seeds: 42, 123, 456, 789, 1024) K. pneumoniae-Ceftriaxone Models: Baseline model - Attention model (paper parameters) Species-specific optimized models (dropout 0.35) 5-model ensemble S. aureus-Oxacillin Models: Baseline model Species-specific optimized models (LR 3×10⁻⁴) 5-model ensemble Model Architecture Type: 1D Convolutional Neural Network Input: 6,000 m/z bins (2000-20000 Da, 3 Da bin width) Framework: TensorFlow 2.19 / Keras Format: HDF5 (.h5 files) Total Parameters: ~25.2M per model Performance Results Species Antibiotic AUROC Paper Target Achievement E. coli Ceftriaxone 0.901 0.87 103.6% ✅ K. pneumoniae Ceftriaxone 0.808 0.82 98.5% S. aureus Oxacillin 0.907 0.93 97.5% Usage Download models from this archive (~6-7 GB) Clone repository: git clone https://github.com/Muhammad-Lukman/MSDeepAMR_Recreation_Enhancement_project.git Extract models to models/ folder in repository root Load models in Python: from tensorflow import keras model = keras.models.load_model('models/ecoli/5_ensemble/model_seed42.h5') File Structure models/ ├── ecoli/ │ ├── 1_baseline/final_model_ecoli_ceftriaxone.h5 │ ├── 2_attention/final_attention_model.h5 │ ├── 3_optimized/final_optimized_model.h5 │ └── 5_ensemble/ │ ├── model_seed42.h5 │ ├── model_seed123.h5 │ ├── model_seed456.h5 │ ├── model_seed789.h5 │ └── model_seed1024.h5 ├── kpneumoniae/ │ ├── 1_baseline/ │ ├── 2_attention_paper_params/ │ └── 3b_optimized/ │ ├── single_best_model.h5 │ └── ensemble/ (5 models) └── saureus/ ├── baseline/ └── ensemble/ (5 models) Data Availability Preprocessed training data is available in the GitHub repository via Git LFS: Repository: MSDeepAMR_Recreation_Enhancement_project Path: data/processed/ Access: git lfs pull after cloning Raw MALDI-TOF data from DRIAMS database: Source: https://doi.org/10.5061/dryad.bzkh1899q Database: DRIAMS-A (Weis et al., 2022) Citation If you use these models, please cite: This dataset: Weis, Caroline; Cuénod, Aline; Rieck, Bastian et al. (2025). DRIAMS: Database of Resistance Information on Antimicrobials and MALDI-TOF Mass Spectra [Dataset]. Dryad. https://doi.org/10.5061/dryad.bzkh1899q GitHub repository: @software{muhammadlukman2025msdeepamr, author = {Lukman, Muhammad}, title = {Recreation and Enhancement of MSDeepAMR: A Deep Learning Approach for Antimicrobial Resistance Prediction from MALDI-TOF Mass Spectrometry Data}, year = 2025, url = {https://github.com/Muhammad-Lukman/MSDeepAMR_Recreation_Enhancement_project} } Original MSDeepAMR paper: @article{lopez2024msdeepamr, author = {López-Cortés, X. A. and Manríquez-Troncoso, J. M. and Hernández-García, R. and Peralta, D.}, title = {MSDeepAMR: antimicrobial resistance prediction based on deep neural networks and transfer learning}, journal = {Frontiers in Microbiology}, volume = {15}, pages = {1361795}, year = {2024}, doi = {10.3389/fmicb.2024.1361795} } Related Resources GitHub Repository: MSDeepAMR_Recreation_Enhancement_project Original Paper: MSDeepAMR: antimicrobial resistance prediction based on deep neural networks and transfer learning DRIAMS Database: DRIAMS: Database of Resistance Information on Antimicrobials and MALDI-TOF Mass Spectra Technical Requirements Python 3.8+ TensorFlow 2.19.0 16GB RAM recommended GPU recommended for inference (CPU compatible) Contact Email: dr.mlukmanuaf@gmail.com Issues: MSDeepAMR_Recreation_Enhancement_project/issues