Skip to content

Stable nested CV, univariate filtering, faster sparsification

Pre-release
Pre-release
Compare
Choose a tag to compare
@shraddhapai shraddhapai released this 26 Jun 21:25
· 1812 commits to master since this release
v1.0.23
aee3680

New version pre-release

  • Improved sparsification > improved scalability. More control over sparsification
  • runPredictor_nestedCV() now has flag for univariate filtering of features.
  • Better control of ProfileToNetworkDriver through user-defined parameters passed to it.
  • All code for examples run in the netDx methods paper and PSN review article (Pai S and GD Bader (2018) J Mol Biol) now moved out of this repo, leaving the code purely related to the netDx software package and tutorial examples.

** A release will follow once the code has been tested on a continuous-valued predictor and CNV-based predictor.

Details

  • DESCRIPTION: Version changed to 1.0.23
  • New feature: corrFeatWithOutcome.R: Correlates PC projections of features with phenotypes, allowing correlation of individual features with outcome. Tool to prioritize selected features based on which also correlate with outcome.
  • New feature: sparsify3: Faster sparsifier for large datasets.
  • GM_createDB.R: New parameters P2N_threshType and P2N_maxMissing: allow user control of GeneMANIA's ProfileToNetwork
  • getFeatureScores.R: Option to return the full matrix of pathway scores for all splits, including those with NA. Useful when features are not consistently used across splits (e.g. univariate filtering for pathways)
  • makePSN_NamedMatrix.R:
    • previous sparsify method deprecated. useSparsify2 switch now decides whether to use sparsify2 (loop-based) or sparsify3 (matrix-based, faster).
    • sparsify_edgeMax, sparsify_maxInt: parameters to control input network sparsification
    • enforces writeProfiles=TRUE when similarity metric is pearson or MI
    • similarity metric can now be "MI", without supplying custom function
    • simMetric=pearson now requires minimum 5 measures per feature.
  • nWay_netSum.R: Bug fix that was causing crash. Change does not afect logic.
  • runPredictor_nestedCV:
    • Added option to start with rngNum that is not 1, useful for predictors that aborted after a certain number of splits.
      • Added preFilter flag, which performs univariate lasso filtering when set.
  • New sparsification methods:
  • sparsify2 (to be deprecated once tests on sparsify3 are complete)
  • sparsify3 (faster, matrix-based version of sparsify2)
  • writeEMapInput: No longer assigns colnames to netInfo.
Assets 2
Loading