CHANGES IN VERSION 1.7.38 (2012-03-28) SIGNIFICANT USER-VISIBLE CHANGES o added the option to ignore network in detect.responses (netw=NULL); then the methods assumes fully connected network. Without speedups the performance may be very slow. o added (optional) initial mutual information based filtering of the network edges also in the first stage where pairwise similarities are calculated for all node pairs This can give considerable speedups with large networks o merged netresponse.visualization package in the netresponse main package o added mode = hard in sample2response function o plot.associations -> plotAssociations o plot.pca -> plotPCA o get.gofz -> getqofz BUG FIXES o changed dependency RBGL into RColorBrewer o changed dependencies multicore and doMC into parallel o Modified mk.hp.posterior so as to accommodate 'max number of responses' (c.max) option. Validations in tests/vdpmixture.R ok o now allowing max.subnet.size = 1 in detect.responses o order.responses: fixed minor bug which occurs when no enrichments are detected CHANGES IN VERSION 1.7.34 (2012-03-28) o changed dependency RBGL into RColorBrewer CHANGES IN VERSION 1.7.33 (2012-03-13) o Modified mk.hp.posterior so as to accommodate 'max number of responses' (c.max) option. Validations in tests/vdpmixture.R ok CHANGES IN VERSION 1.7.32 (2012-02-21) o added the option to ignore network in detect.responses (netw=NULL); then the methods assumes fully connected network. Without speedups the performance may be very slow. CHANGES IN VERSION 1.7.03 (2012-02-02) o merged netresponse.visualization package in the netresponse main package CHANGES IN VERSION 1.7.02 (2012-02-01) o switched to R-2.14.1 o now allowing max.subnet.size = 1 in detect.responses o order.responses: fixed minor bug which occurs when no enrichments are detected o added mode = hard in sample2response function o added (optional) initial mutual information based filtering of the network edges also in the first stage where pairwise similarities are calculated for all node pairs This can give considerable speedups with large networks o plot.associations -> plotAssociations o plot.pca -> plotPCA o get.gofz -> getqofz CHANGES IN VERSION 1.5.14 (2011-10-28) o removed compiler package and functions as they are included in R-2.14 CHANGES IN VERSION 1.5.12 (2011-06-29) o added options for read.network CHANGES IN VERSION 1.5.11 (2011-06-28) o compiler added o corrected weight estimation from vdp.mixt CHANGES IN VERSION 1.5.09 (2011-06-13) o updated toydata o compute.weights: now operating in log-domain to avoid occasional floating point errors CHANGES IN VERSION 1.5.08 (2011-05-27) o added plot.pca function to visualize subnets in 2D with ellipsoid confidence intervals and annotation colors CHANGES IN VERSION 1.5.04 (2011-05-27) BUG FIXES o fixed discretization NAMESPACE issues CHANGES IN VERSION 1.5.02 (2011-05-18) SIGNIFICANT USER-VISIBLE CHANGES o added speedup option speedup.max.edges in detect.responses. To consider only this many most similar neighborghs in merging, similarity is evaluated by empirical mutual information estimate with sqrt(n) bins where n is data sample size o changed AIC to default information criterion (previously BIC). Both remain to be available through information.criterion argument CHANGES IN VERSION 1.3.43 (2011-04-13) o updated vignette CHANGES IN VERSION 1.3.41 (2011-04-08) o updated documentation for ICMg functions CHANGES IN VERSION 1.3.39 (2011-04-07) o optimized the code (vdp.mixt, compute.weight, detect.responses:subgraph) CHANGES IN VERSION 1.3.37 (2011-04-06) o read.sif function improved CHANGES IN VERSION 1.3.36 (2011-04-04) SIGNIFICANT USER-VISIBLE CHANGES o added default speedups (speedup, speedup.max.edges options in detect.responses) CHANGES IN VERSION 1.3.35 (2011-04-04) o added functions for the ICMg (Interaction Component Model for finding Gene modules) package by Juuso Parkkinen CHANGES IN VERSION 1.1.34 (2011-04-03) o fixed a bug in updatePosterior which caused halt; now trying to increase implicit noise for calculating free energy, if this does not help, merging of the investigated subnetwork pair is prevented (previously all calculations were stopped at this point, which is not practical) CHANGES IN VERSION 1.1.33 (2011-03-30) o added various information criteria for model selection to detect.responses and find.similar.features (AIC, AICc, BIC) o added vignette PDF windows buildcheck seems to require it CHANGES IN VERSION 1.1.31 (2011-03-26) o polished vignette o optimized memory usage CHANGES IN VERSION 1.1.29 (2011-03-23) SIGNIFICANT USER-VISIBLE CHANGES o package: xadded support for reading in SIF network files through read.sif function o detect.responses: added support for igraph and graphNEL network formats; in detect.responses output the network is now also outputted in graphNEL format. This is more memory-efficient than previous implementation. OTHER CHANGES o within detect.responses now handling the network through edge list; this is considerably more memory-efficient than previous implementation o added dependencies to graph and igraph packages CHANGES IN VERSION 1.1.28 (2011-03-23) o made mixture density functions visible in NAMESPACE as this seems to be required in the new R version even if the functions are used internally. CHANGES IN VERSION 1.1.27 (2011-03-22) o removed nodes and samples from NetResponseModel output CHANGES IN VERSION 1.1.26 (2011-03-22) o moved NetResponseModel functions in AllGenerics.R o removed intermediate matrices from detect.responses.R (costs, Nparams, bic.ind, bic.joint) o added two fields in pick.model.parameters: free.energy and Nparams o added internal function bic for bic calculations o removed costs from NetResponseModel output CHANGES IN VERSION 1.1.22 (2011-03-13) USER-VISIBLE CHANGES o added support for sparse matrices with the Matrix package (from CRAN) CHANGES IN VERSION 1.1.19 (2011-01-25) BUG FIXES o now providing names for the elements in model@params list USER-VISIBLE CHANGES o renamed result.stats into model.stats o now NetResponseModel also return the recalculated models for the responses. No need to recalculate them which speeds up analysis of the end results. Also useful since the result is not completely deterministic due to stochastic initializations in model learning etc., and relearning the models can sometimes produce slightly different results. NEW FEATURE o added find.similar.features function which can be used to find features (genes) that are similar to the subnetwork but not directly interacting with it CHANGES IN VERSION 1.1.18 (2011-01-25) USER-VISIBLE CHANGES o in get.subnets function now using default min.responses=2 and subnet.size=2 (at least 2 genes in a subnetwork) o added 'subnets' field to the NetResponseModel class. This lists the subnetworks by feature names. BUG FIXES o NetResponseModel class now returns the original network used in the experiment (diagonal set to zero, however), previous version returned the agglomerated network NEW FEATURES o added functions order.responses and order.samples BUG FIXES o in result.stats function corrected the minimum responses filtering part CHANGES IN VERSION 1.1.17 (2011-01-19) o users can now give the expression data D in either matrix or data.frame format to detect.responses function, which will automatically convert non-matrix inputs into matrix format. CHANGES IN VERSION 1.1.16 (2011-01-16) BUG FIXES o in the last step of the merging stage the network reduced into a numeric which caused error, now corrected by rejecting further operations with completely merged nets in detect.responses CHANGES IN VERSION 1.1.15 (2011-01-15) BUG FIXES o in split.qofz minimum cluster size check added. clusters smaller than min.size cannot be splitted any more; this caused problems CHANGES IN VERSION 1.1.13 (2011-01-07) BUG FIXES o in split.qofz speedup function added the line 'indices <- indices[rinds]'. Without this, an erroneous subset is selected and the program will eventually fail. CHANGES IN VERSION 1.1.12 (2011-01-02) BUG FIXES o in internals.R some functions (mk.qOFz, mk.E.log.q.p.eta, mk.free.energy, mk.hp.posterior, mk.hp.prior, mk.log.lambda) were defined multiple times. Removed multiple copies. Also ensuring that no empty clusters are returned; this caused problems in split.qOFz and updatePosterior OTHER CHANGES o Code has been optimized for speed o updated citation info CHANGES IN VERSION 0.99.3 (2010-10-03) o Added slots 'datamatrix' and 'network' into NetResponseModel class. These are used to store the input data used to learn the model. Subsequently, removed data inputs from functions such as get.parameters etc. since now they can fetch the data matrix directly from the model object. o Added project URL address to DESCRIPTION file o Added 'speedup' option to vdp.mixt CHANGES IN VERSION 0.99.2 (2010-09-28) o Fixed bug in C code. The previous version had convergence problems related to PCA initialization of component split during stick-breaking process, and mixture component centroids were subsequently not retrieved optimally o changed the output format of function vdp.mixt to focus on variables relevant for the end user. o updated toydata and man pages according to the change to vdp.mixt output o added test/demo.R testing and demonstrating vdp.mixt CHANGES IN VERSION 0.99.0 (2010-09-23) o Improved man pages and examples o Removed greedy and find.best.splitting from user-visible function space o Added pre-calculated NetResponseModel object to toydata CHANGES IN VERSION 0.4.4 (2010-09-17) o Initial release.