Authors: Manitejus Kotikalapudi
AutoDeep is a tree-boosting software that further stratifies miRDeep2 (Mackowiak, Friedländer) outputs into Candidate, Confident, and Potential False-Positive Labels
Linux system with GCC, Conda Environment generated from AutoDeep/AutoDeep.yml file
Download the AutoDeep Directory. Within the Directory, and with the AutoDeep conda environment activated, type
pip install -e .Core Logic of AutoDeep package. Formats miRDeep2 output CSV file, performs feature extraction, and then Classifies pre-miRNA loci with XGBoost
- Directory in which miRDeep2 was run
The input directory should have the following file structure:
├── directory in which miRDeep2 was run
│ ├── result_<[0-9]>.csv
│ ├── pdfs_<[0-9]>
│ │ ├── <loci_name_[0-9]>.pdf
A directory named "AutoDeepRun" which contains
- XGBoost label prediction CSV
Trains AutoDeep's underlying XGBoost model with user data. Must be run within AutoDeepRun directory
- CSV file with loci names in first column and class names in second
- Training Log CSV
- -n Omits original training data from model training (i.e only uses your inputs)
- -t, --targets_path Path to targets file
- -n, --no_db_data Flag that omits original dataset from training
- -r, --tuning_rounds Number of tuning rounds: Default <10>
- -o, --output Name of output training_log file
- -nw, --no_weights Flag that omits saving the model weights (recommended for testing)
- -hp, --hyperparameters Path to hyperparameter configuration file in case of manual tuning
- --help Show this message and exit.
Visualizes XGBoost model via tree structure and gain metrics.
N/A
- Folder containing relevant figures as png
- --no_tree Do not output tree plots
- -o, --output TEXT Output directory for tree plots
- --help Show this message and exit.