deepakagrawal/hack4rare

Hack4Rare

Please download the pbta-gene-expression-kallisto.stranded.rds data from OpenPBTA to data/ directory. Instructions on how to download the dataset are here.

How to run the code.

1. Run src/create_raw_files.py file to generate node and edge_list files.
2. Run src/main.py to generate the node embedding. This code is a based on metapath2vec paper. pytorch_geometric has incorporated the methodology here. metapath2vec is an approach based on unweighted edges. Our proposed approach generalizes this to weighted networks.

• Please use the --use_weight flag to run weighted Deep Walk. Weighted deep walk is slow. Therefore, the ideal approach is to run the unweighted approach first and then fine tune the embedding by using the weighted approach.
• Please use the --help flag to see the command line arguments available.

3. To visualize the generated embedding please use projetor.tensorflow.org.

hack4rare's Issues

Create KNN Clusters using the generated gene embeddings

Generate KNN clusters using the generated gene/sample embedding files.

Build a recommendation system for patient treatment

Hyper parameter tuning for Metapath2vec model parameters

We need to tune the following hyperparameters in order to improve the embedding loss as well as training loss on sample nodes.

• embed-dim
• walk_length
• context_size
• walks_per_node
• num_negative_samples