NEWS.md

decoupleR 2.1.

Changes

• likelihood param is deprecated, from now on, weights (positive or negative) should go to the mor column of network. Methods will still run if likelihood is specified, however they will be set to 1.

• Added minsize argument to all methods, set to 5 by default. Sources containing less than this value of targets in the input mat will be removed from the calculations.

• Changed default behavior of the decouple function. Now if no methods are specified in the statistics argument, the function will only run the top performers in our benchmark (mlm, ulm and wsum). To run all methods like before, set statistics to 'all'. Moreover, the argument consensus_stats has been added to filter statistics for the calculation of the consensus score. By default it only uses mlm, ulm and norm_wsum, or if statistics=='all' all methods returned after running decouple.

• viper method:

  • Now properly handles weights in mor by normalizing them to -1 and +1.

• ulm/mlm/udt/mdt methods:

  • Changed how they processed the input network. Before the model matrix only contained the intersection of features between mat and network's targets, now it incorporates all features coming from mat ensuring a more robust prediction. Prediction values may change slightly from older versions.
  • Deprecated sparse argument.

• ora method:

  • Now takes top 5% features as default input instead of 300 up and bottom features.
  • Added seed to randomly break ties

• consensus method:

  • No longer based on RobustRankAggreg. Now the consensus score is the mean of the activities obtained after a double tailed z-score transformation.

• Discarded filter_regulons function.

• Moved major dependencies to Suggest to reduce the number of dependencies needed.

• Updated README by adding:

  • Kinase inference example
  • Graphical abstract
  • Manuscript and citation
  • New vignette style

New features

• Added wrappers to easily query Omnipath, one of the largest data-bases collecting prior-knowledge resources. Added these functions:

  • show_resources: shows available resources inside Omnipath.
  • get_resource: gets any resource from Omnipath.
  • get_dorothea: gets the DoRothEA gene regulatory network for transcription factor (TF) activity estimation. Note: this version is slightly different from the one in the package dorothea since it contains new edges and TFs and also weights the interactions by confidence levels.
  • get_progeny: gets the PROGENy model for pathway activity estimation.

• Added show_methods function, it shows how many statistics are currently available.

• Added check_corr function, it shows how correlated regulators in a network are. It can be used to check for co-linearity for mlm and mdt.

• Added new error for mlm when co-variables are co-linear (regulators are too correlated to fit a model).

Bugfixes

• wmean and wsum now return the correct empirical p-values.

• ulm, mlm, mdt and udt now accept matrices with one column as input.

• Results from ulm and mlm now correctly return un-grouped.

• Methods correctly run when mat has no column names.

decoupleR 2.0

Changes

• Some method's names have been changed to make them easier to identify:

  • pscira now is called Weighted Sum (wsum).
  • mean now is called Weighted Mean (wmean).
  • scira now is called Univariate Linear Model (ulm).

• The column name for tf in the output tibbles has been changed to source.

• Updated documentation for all methods.

• Updated vignette and README.

• decouple function now accepts order mismatch between the list of methods and the list of methods's arguments.

• Moved benchmark branch to a separate repository as its own package: https://github.com/saezlab/decoupleRBench

New features

• New methods added:

  • Fast Gene Set Enrichment Analysis (fgsea).
  • AUCell.
  • Univariate Decision Tree (udt).
  • Multivariate Decision Tree (mdt).
  • Multivariate Linear Model (mlm).

• New decoupleR manuscript repository: https://github.com/saezlab/decoupleR_manuscript

• New consensus score based on RobustRankAggreg::aggregateRanks() added when running decouple with multiple methods.

• New statistic corr_wmean inside wmean.

• Methods based on permutations or statistical tests now return also a p-value for the obtained score (fgsea, mlm, ora, ulm, viper, wmean and wsum).

• New error added when network edges are duplicated.

• New error added when the input matrix contains NAs or Infs.

decoupleR 1.1

New features

All new features allow for tidy selection. Making it easier to evaluate different types of data for the same method. For instance, you can specify the columns to use as strings, integer position, symbol or expression.

Methods

• New decouple() integrates the various member functions of the decoupleR statistics for centralized evaluation.

• New family decoupleR statists for shared documentation is made up of:

  • New run_gsva() incorporate a convinient wrapper for GSVA::gsva().
  • New run_mean() calculates both the unnormalized regulatory activity and the normalized (i.e. z-score) one based on an empirical distribution.
  • New run_ora() fisher exact test to calculate the regulatory activity.
  • New run_pscira() uses a logic equivalent to run_mean() with the difference that it does not accept a column of likelihood.
  • New run_scira() calculates the regulatory activity through the coefficient $\beta_1$ of an adjusted linear model.
  • New run_viper() incorporate a convinient wrapper for viper::viper().

Converters

• New functions family convert_to_ variants that allows the conversion of data to a standard format.
  • New convert_to_() return the entry without modification.
  • New convert_to_gsva() return a list of regulons suitable for GSVA::gsva().
  • New convert_to_mean() return a tibble with four columns: tf, target, mor and likelihood.
  • New convert_to_ora() returns a named list of regulons; tf with associated targets.
  • New convert_to_pscira() returns a tibble with three columns: tf, target and mor.
  • New convert_to_scira() returns a tibble with three columns: tf, target and mor.
  • New convert_to_viper() return a list of regulons suitable for viper::viper()