How lormebioai/dcipatho works
Project type and purpose
• DCiPatho is a Python-based machine-learning application (an ML pipeline) for classifying bacterial sequences as pathogenic or non-pathogenic.
• It provides two main workflows:
• Training & evaluation on k-mer frequency features (via DCiPatho_main.py + DCiPatho_trainer.py + DCiPatho_network.py).
• Inference/prediction on raw FASTA reads (via DCiPatho_predict.py + utility modules).
File structure
Top-level files:
• DCiPatho_config.py – central Config class for both training and prediction settings.
• DCiPatho_main.py – entry point for training/evaluation.
• DCiPatho_trainer.py – orchestrates training loops, early stopping, model checkpointing.
• DCiPatho_network.py – defines the Deep Cross, ResNet, and DeepNet submodules; composes full PyTorch model.
• DCiPatho_predict.py – entry point for inference: combines FASTA, computes k-mer freqs, loads model, generates predictions.
• BaseModel/basemodel.py – abstract base class for models (shared logic).
• Utils/ – helper functions for k-mer calculation (cal_utils.py), file combination (combine_fna.py), training callbacks (pytorchtools.py), general tools (tool.py).
• requirements.txt – external dependencies (numpy, torch, pandas, scikit-learn).
• temp_data/ – sample precomputed data for quick demos.
README for additional insights
• Clearly describes two workflows (training vs. prediction), all configurable hyperparameters, data sources, and evaluation metrics.