Resources:

• README.md: this file.
• result.csv: The performance for different classifers and features. Measured in F1, Precision, Recall, Specificity, Accuracy, and AUC.

Source codes:

• utils.py: provides functions of (1) extracting 'closeness centrality' for graphs: get_closeness_centrality, (2) evaluating classification performance: evaluate, and (3) converting graphs of continuous weights to ajacency graphs: to_adjacency_G
• PAN.ipynb: creates PAN graphs and returns their graph properties
• LIONESS.ipynb: gets graph properties for LIONESS graphs returned by pypanda (https://github.com/davidvi/pypanda)
• PPI.ipynb: creates PPI graphs and returns their graph properties
• classification.ipynb: provides two classifers, SVM and LogisticRegression, with gene expression and graph propeties as features.
• PAN.html, LIONESS.html, PPI.html, and classification.html: html files for PAN.ipynb, LIONESS.ipynb, PPI.ipynb, and classification.ipynb, respectively.

Folder input_data/:

• UK207.csv: UK207 dataset. Other two datasets will be provided upon request.
• hgncToKEGG.csv, hgncToDO.csv, hgncToHPO.csv: 3 ontologies.
• COMBINED.DEFAULT_NETWORKS.BP_COMBINING.txt: gene interaction, from GeneMania (http://genemania.org/data/current/Rattus_norvegicus.COMBINED/)
• ENST-ENSG-HGNC-GO-kegg-map.csv: map ensembl_gene_id in COMBINED.DEFAULT_NETWORKS.BP_COMBINING.txt to hgnc_symbol in gene expression datasets.

Step-by-step running:

0. Installing Python libaries needed

• Install pypanda, https://github.com/davidvi/pypanda. This creates LIONESS graphs.
• Install sklearn: pip install scikit-learn
• Install networkx: pip install networkx

1. PAN:

Running

PAN.ipynb

This returns

• Top genes selected by VarianceThreshold, saved in output_data/top_genes.txt
• Gene expression for the top genes and its label, saved in output_data/gene_condition.csv.
• Label, saved in output_data/y.txt
• Non-graph data, or gene expression, save in output_data/non_graph.csv
• Gene expression, as the transpose of output_data/non_graph.csv, saved in output_data/gene_expression.txt. This is the input to LIONESS.
• 10fold_idx/: ten-fold splits to be used for all models
• PAN_KEGG graph (saved in output_data/pan_KEGG_graph.csv) and the feature for this graph (output_data/pan_KEGG_graph_feature.csv) for the dataset.
• PAN_HPO graph (saved in output_data/pan_HPO_graph.csv) and the feature for this graph (output_data/pan_HPO_graph_feature.csv) for the dataset.
• PAN_DO graph (saved in output_data/pan_DO_graph.csv) and the feature for this graph (output_data/pan_DO_graph_feature.csv) for the dataset.

2. LIONESS:

• (Please install pypanda, https://github.com/davidvi/pypanda)
• Running

./pypanda  -e output_data/gene_expression.txt -o output_data/gene_expression_panda.txt -q output_data/lioness.txt

The output output_data/lioness.txt is the input to LIONESS.ipynb. Running

LIONESS.ipynb

This returns the feature for LIONESS graph, saved in output_data/lioness_graph_feature.csv

3. PPI:

running

PPI.ipynb

This returns the feature for PPI graph, saved in output_data/ppi_graph_feature.csv

4. Classification:

Now all the graphs and their features are already extracted. They are the input to classification.ipynb. Running

classification.ipynb

This returns the performance for all the features in two classifers, saved in result.csv. A visualization on the average of ten-fold running is saved at figs/SVM.png and figs/LR.png, for the two classfiers.