Table 1. Major functions for basic and advanced model-building, and result evaluation.
Intermediate functions used to prepare data are not shown, but are illustrated in the use cases. The third column shows outdated names for these functions from the software release that accompanied the paper describing the methods.
| Function name
(v1.1.4) |
Purpose |
|---|---|
| buildPredictor() | Turnkey function to build predictor |
| buildPredictor_sparseGenetic() | Turnkey function to build predictor from sparse
mutation data of arbitrary ranges (e.g. copy number variants) |
| makePSN_NamedMatrix() | Create PSN from a single data layer (matrix
representation) |
| createPSN_MultiData() | Create PSN from multiple data layers. May include
calls to makePSN_NamedMatrix() |
| plotPerf() | Plot ROC and PR curves, compute AUROC and AUPR |
| plotEMap() | Plot enrichment map in Cytoscape, annotating main
themes. Used in pathway-based feature design |
| plotIntegratedPatientNetwork() | Plot patient network by integrating predictive features
for all patient labels |
| splitTrainTest() | Randomly split patients into training and test samples |
| setupFeatureDB() | Collects all created features into a database in
preparation for feature selection |
| compileFeatures() | Feature selection process. Iterative scoring of
input networks based on network integration and regularized regression. Each scoring step is called a “query”. A pre-specified number of queries is run with different subsamples of the training set |
| runFeatureSelection() | Feature selection: Run network integration and
label propagation. Used in feature selection for unit increase in network weights. Also used for classification of test samples, where it returns similarity scores |
|
writeQueryFile()
runQuery() |
Feature selection: Prepare input for unit-level network
integration. Feature selection: Run unit-level network integration. Each run of this step allows unit-level increase of feature scores |
| compileFeatureScores() | Feature selection: Collect feature scores for unit-level
network integration and feature scoring steps |
| getPatientRankings() | Model evaluation: Process test patient rankings from
label propagation, to compute model performance measures |
| predictPatientLabels() | Model evaluation: Given patient rankings for each
label, assign predicted class label to test patients |