PASNet

PASNet is a pathway-based sparse deep neural network. The PASNet model has the following contributions:

• Interpretable neural network on the biological pathway level

• Training the neural netowrk with high-dimension, low-sample size data

• Automatically optimizing the sparse neural network

• Better classification performance

Reference

@article{hao2018pasnet:,
  author = {Hao, Jie and Kim, Youngsoon and Kim, Tae-Kyung and Kang, Mingon},
  year = {2018},
  title = {PASNet: pathway-associated sparse deep neural network for prognosis prediction from high-throughput data},
  journal = {BMC Bioinformatics},
  doi = {10.1186/s12859-018-2500-z},
  volume = {19},
  month = {12},
  pages = {510},
  number = {1},
  url = {https://doi.org/10.1186/s12859-018-2500-z},
}

Get Started

Example Datasets

To get started, you need to download example datasets from URLs as below:

Train data

Validation data

Pathway Mask data

Empirical Search for Hyperparameters

Run_EmpiricalSearch.py: to find the optimal pair of hyperparmaters for PASNet before performing cross validation. PASNet is trained with the inputs from train.csv. Hyperparameters are optimized by emipirical search with validation.csv.

5-fold Cross Validation

Run.py: to train and evaluate the model performance based on 10 times 5-fold cross validation.