RELEASE HISTORY OF HTSCluster PACKAGE
             ========================================

		CHANGES IN HTSCluster VERSION 2.0.8-2.0.11

- Updated all imports to function calls outside of base		
- Added possibility to input pre-calculated normalization factors (optional)
and to choose a subset of genes for co-expression analysis in the PoisMixClus
and PoisMixClusWrapper functions (thanks to S. Thevenon for the suggestions!).
- Package-wide, the lib.size argument has been removed and the lib.type argument has been 
replaced by the argument "norm", which can take the values of different normalization 
types (none, TC, UQ, Med, DESeq, TMM) or can be a vector of pre-calculated normalization
factors.
- The subset.index argument has been added to PoisMixClus and PoisMixClusWrapper.
- Removed metadata from HTSCluster package overview help file

		CHANGES IN HTSCluster VERSION 2.0.6-2.0.7
		
- Updated citation information throughout (paper published in Bioinformatics).
- Fixed bug in PoisMixClusWrapper function when the number of clusters run is < 10.
- Fixed bug for check that the data matrix is made up of integers, i.e., changed
sum(y) to sum(as.numeric(y)) to avoid integer overflow.
- Removed poisson.glm.mix from DEPENDS for integer overflow bug listed above (and
added the function logLikePoisMix to exported functions from HTSCluster)

		CHANGES IN HTSCluster VERSION 2.0.5
		
- Change contrast for capushe to -log.like/n for slope heuristics in
PoisMixClusWrapper function (implies no change to model selection results
in previous version of package).
- Updated reference to Papastamoulis et al. (2014)
- Minor changes to manual files and recompiled vignettes.
  
		CHANGES IN HTSCluster VERSION 2.0.4
		
- Minor changes to speed up code in logLikePoisMixDiff, PoisMixMean,
  probaPost, and PoisMixClus and avoiding recalculating library sizes multiple 
  times across several functions, as well as improving compile time for Init Rd file.
- Import capushe package and added capushe results and plotting capabilities for
  PoisMixClusWrapper, and removed ggplot2 and RColorBrewer dependencies.
- Completed vignette.
- Minor change to summary function for PoisMixClusWrapper objects.

		CHANGES IN HTSCluster VERSION 2.0.3
		
- Fixed typo in calculation of entropy for ICL (thanks to M. Gallopin)
- Added mylogLikePoisMixObs function to calculate log-likelihood for each 
  observation.
- Default number of init.runs now fixed at 1
- Added EM.verbose option to include more informative output about the algorithm
- Added iterations and logLikeDiff into the output of the PoisMixClus and
  PoisMixClusWrapper functions.
  
		CHANGES IN HTSCluster VERSION 2.0.2
		
- Added poisson.glm.mix to package IMPORTS to use the function mylogLikePoisMix
  for all likelihood calculations in order to avoid numerical difficulties (and
  deleted logLikePoisMix from the PoisMixClus and probaPostInit functions).
- Set up initial file for vignette.
- Changed splitSmallEM initializing strategy so as to split the cluster with the
  largest per-cluster entropy (-sum_{i \in k} \log(t_{ik}) rather than randomly
  choosing a cluster to split.
					
		CHANGES IN HTSCluster VERSION 2.0.1
		
- Changed entropy calculation for ICL from -2*sum*(z*log(t)) to -2*sum(t*log(t)).

		CHANGES IN HTSCluster VERSION 1.99.0

- Created vignette and doc directories.
- Moved edgeR to IMPORTS rather than DEPENDS.
- Additional plotting option for lambda using ggplot2 and RColorBrewer, both of
  which have been added as dependencies in the namespace.
- Behavior of clusters can now be interpreted with respect to overall gene expression
  (interpretation = "sum" in the PoisMixClus and PoisMixClusWrapper functions) or with
  respect to mean gene expression (interpretation = "mean").
- Added user-specific parameters init.runs (number of small-EM algorithms to
  run for initialization) and init.iter (number of iterations to run within each
  small-EM algorithm).
- Added highDimensionARI function
- Renormalized posterior probabilities after smoothing in probaPost function.
- Removed smoothing option from logLikePoisMix() and implemented numerical
  technique suggested by P. Papastamoulis to calculate log likelihood.
- Mean.filter argument removed from PoisMixClus function.
- Changed "Q" to "UQ" (Upper Quantile), and added "Med" (Median),
  "DESeq" (normalization in DESeq package), and "TMM" (Trimmed Means
  of M values, from the edgeR package) for library size normalization.
- Changed PoisMixClus to estimate a Poisson mixture model for a single
  number of clusters. To estimate a PMM for a series of cluster sizes and
  perform model selection, added the wrapper function PoisMixClusWrapper,
  which enables the user to perform initialization using the small EM strategy
  (as before) or the initialization strategy proposed by Panos Papastamoulis via
  the functions splitEMInit and probaPostInit.
- Added the possibility to estimate model for equally proportioned
  clusters, as well as to include a cluster with fixed parameter values.
- Creation of S3 class "HTSClusterWrapper" and associated S3 methods
  plot.HTSClusterWrapper and summary.HTSClusterWrapper.
- The number of free parameters for the BIC and ICL criteria now does not include
  the n per-gene w parameters, as these are considered to be fixed.

		CHANGES IN HTSCluster VERSION 1.1

- Smoothing option (default = TRUE) has been added to logLikePoisMix() to
  alleviate numerical errors.
- Creation of NAMESPACE, NEWS, and CITATION information.
- Creation of S3 class "HTSCluster" and associated S3 methods plot.HTSCluster
  and summary.HTSCluster. The plot.HTSCluster function now depends on the
  plotrix package, which has been added to Depends in DESCRIPTION.
- Fixed initialization bug for data with non-numerical condition names in
  kmeanInit function.
- License updated to GPL (>=3)

		CHANGES in HTSCLuster VERSION 1.0

This is the first release (7 November 2011).