How praneeth728/megad works
MegaD is a Python tool for training deep neural networks on metagenomic taxonomic profiles and predicting disease labels. The main workflow accepts an abundance table and metadata, processes features, trains and evaluates a classifier, saves its parameters, and can predict labels for an unknown profile. Documented options include taxonomic-level selection, thresholding, normalization, randomized grid search, and ROC reporting. The source excerpts show the processing and model implementation; prediction and some reporting details are documented in the README but not fully sampled.
Main entry points: DNN.py Training Workflow · DataProcessing.py Feature Processing · neuralnet.py Neural Network
Components 11
Input Processing
- Abundance Table
- Sample Metadata
- Feature Processing
Scripts/DataProcessing.py - Unknown Profile
Model Training
- Training Workflow
Scripts/DNN.py - Neural Network
Scripts/neuralnet.py
Evaluation and Prediction
- Model Evaluation
Scripts/DNN.py - Saved Model
- Label Prediction
Scripts/DNN.py - ROC Curve
Scripts/DNN.py
Other components
- Researcher
Connections 14
- Researcher → Abundance Table: provides
README.md - Researcher → Sample Metadata: provides
README.md - Training Workflow → Feature Processing: processes inputs
Scripts/DNN.py - Abundance Table → Feature Processing: reads
Scripts/DataProcessing.py - Sample Metadata → Feature Processing: reads
Scripts/DataProcessing.py - Feature Processing → Training Workflow: returns features
Scripts/DNN.py - Training Workflow → Neural Network: constructs and trains
Scripts/DNN.py - Training Workflow → Model Evaluation: computes accuracy
Scripts/DNN.py - Training Workflow → Saved Model: saves parameters
Scripts/DNN.py - Researcher → Unknown Profile: provides
README.md - Label Prediction → Neural Network: runs inference
Scripts/DNN.py - Label Prediction → Unknown Profile: classifies
README.md - Saved Model → Label Prediction: loads for prediction
- Training Workflow → ROC Curve: generates
README.md