ncRNA Classification

Code for the paper ncRNA Classification with Graph Convolutional Networks by E. Rossi et. al.

Citation

To cite our work, please reference our KDD@DLG 2019 paper

@inproceedings{rossi2019ncrna,
  title={ncRNA Classification with Graph Convolutional Networks},
  author={Rossi, Emanuele and Monti, Federico and Bronstein, Michael and Li{\`o}, Pietro},
  booktitle={Proceedings of the 1st International Workshop on Deep Learning on Graphs: Methods and Applications (DLG@KDD)},
  year={2019}
}

Requirements

Our code has been tested with python=3.7, but it may be possible for it to work with lower versions as well (eg. python 3.4-3.6)*[]:

To install all the required packages it is sufficient to run:

pip install -r requirements.txt

The setting has been tested for macOS Mojave version 10.14.5, but we expect this setup to work with other unix systems.

Data

• train.fasta: Training dataset. Contains 5670 ncRNA sequences belonging to 13 different classes
• val.fasta: Validation dataset. Contains 650 ncRNA sequences belonging to 13 different classes
• test_13_classes.fasta: Full test dataset. Contains 2600 ncRNA sequences belonging to 13 different classes
• test_12_classes.fasta: Reduced test dataset. Contains 2400 ncRNA sequences belonging to 12 different classes.
• foldings.pkl: A dictionary containing precomputed mapping from sequence to corresponding most likely folding (graph) for all sequences from the above datasets.

The data has been taken from 'nRC: non-coding RNA Classifier based on structural features'. The files train.fasta and val.fasta have been obtained by randomly splitting the their original training dataset into two.

Usage

Training

It is possible to train the model with the commands:

cd src/
python training/train_model.py

The hyperparameters of the model can be passed as arguments to the script. It is possible to see which hyperparameters can be passed by running (from inside src/):

python training/train_model.py --help

Evaluation

After training, the model can be evaluated with the command (from inside src/):

python evaluation/evaluate_model.py

By default, the model is evaluated on the test_13_classes.fasta test dataset, but it is possible to change this by passing another test dataset as a script argument.