Package netclass. July 2, 2014

Transcription

1 Package netclass Jul 2, 2014 Version Date Title netclass: An R Package for Network-Based Biomarker Discover Author Yupeng Cun Maintainer netclass is an R package for network-based feature (gene) selection for biomarkers discover via integrating biological information. This package adapts the following 5 algorithms for classifing and predicting gene epression data using prior knowledge: 1) average gene epression of pathwa (aep); 2) pathwa activities classification (PAC); 3) Hub network Classification (hubc); 4) filter via top ranked genes (FrSVM); 5) network smoothed t-statistic (stsvm). Depends R (>= 2.14), kernlab Imports AnnotationDbi, Matri, ROCR, graph, igraph, samr Suggests parallel, Biobase, KEGG.db, pathclass License GPL (>= 2) LazLoad es NeedsCompilation no Repositor CRAN Date/Publication :44:46 1

2 2 netclass-package R topics documented: netclass-package ad.matri calc.diffusionkernelp classif.aep classif.frsvm classif.hubc classif.pac classif.stsvm cv.aep cv.frsvm cv.hubc cv.pac cv.stsvm EN2SY epr getgeneranking getgraphrank Gs pgenerank pofhubs predictaep predictfrsvm predicthubc predictpac predictstsvm probeset2pathwa probeset2pathwatrain probeset2pathwatst train.aep train.frsvm train.hubc train.pac train.stsvm Inde 35 netclass-package An R package for network-based microarra Classification We implemented average gene epression of pathwa (aep), pathwa activitive classification (PAC), Hub network Classsifccation, filter via top ranked genes(frsvm), smoothed t-statistic(stsvm) for two classes microarr classification which emploed the prior information.

3 netclass-package 3 Details

4 4 ad.matri Package: netclass Tpe: Package Version: 1.2 Date: License: GPL (>= 2) LazLoad: es Yupeng Cun Maintainer: References Yupeng Cun, Holger Frohlich (2013) netclass: An R-package for network based, integrative biomarker signature discover. ad.matri An adjacenc matri of a sample graph... An adjacenc matri of a sample graph Details An adjacenc matri of a random graph with some random Entre ID of Protein for use in eample files and the vignette

5 calc.diffusionkernelp 5 calc.diffusionkernelp Computing the Random Walk Kernel matri of network Computing the Random Walk Kernel matri of network calc.diffusionkernelp(l, is.adjacenc = TRUE, p = 3, a = 2) L is.adjacenc p a an adjacenc matri that represents the underling biological network. using adjacenc of graph or not #(p) random walk step(s) of random walk kernel constant value of random walk kernel R Return a Random Walk Kernel matri of given network, L. References Kondor, R. I., & Laffert, J. (2002, Jul). Diffusion kernels on graphs and other discrete input spaces. In MACHINE LEARNING-INTERNATIONAL WORKSHOP THEN CONFERENCE- (pp ). as classif.stsvm librar(netclass) data(ad.matri) #dk= calc.diffusionkernelp(l=ad.matri, is.adjacenc=true, p=2,a=1)

6 6 classif.aep classif.aep Training and predicting using aepsvm (aepsvm) classification methods Training and predicting using aepsvm (aepsvm) classification methods classif.aep(fold, cuts, Cs,,, cv.repeat, int, =, Gsub) fold cuts Cs cv.repeat int Gsub number of -folds cross validation (CV) list for randoml divide the training set in to --folds CV soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). gene epression data class labels model for one CV training and predicting Intersect of genes in network and gene epression profile. show debugging information in screen more or less. an adjacenc matri that represents the underling biological network. fold auc train feat the recored for test fold The AUC values of test fold The tranined models for traning folds The feature selected b each b the train References Guo et al., Towards precise classification of cancers based on robust gene functional epression profiles. BMC Bioinformatics 2005, 6:58. as cv.aep #See cv.aep

7 classif.frsvm 7 classif.frsvm Training and predicting using FrSVM Training and predicting using FrSVM classif.frsvm(fold, cuts,,, cv.repeat, =, Gsub = Gsub, d = d, op = op,aa = aa, Cs = Cs) fold number of folds to perform cuts list for randoml divide the training set in to --CV epression data a factor of length p comprising the class labels. cv.repeat model for one CV training and predicting show debugging information in screen more or less. Gsub an adjacenc matri that represents the underling biological network. d damping factor for GeneRank, defaults value is 0.5 op the uper bound of top ranked genes aa the lower bound of top ranked genes Cs soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). fold auc train feat the recored for test fold The AUC values of test fold The tranined models for traning folds The feature selected b each b the train References Yupeng Cun, Holger Frohlich (2012) Integrating Prior Knowledge Into Prognostic Biomarker Discover Based on Network Structure.arXiv: Winter C, Kristiansen G, Kersting S, Ro J, Aust D, et al. (2012) Google Goes Cancer: Improving Outcome Prediction for Cancer Patients b Network-Based Ranking of Marker Genes. PLoS Comput Biol 8(5): e doi: /journal.pcbi

8 8 classif.hubc as cv.frsvm #see cv.frsvm classif.hubc Training and predicting using hub nodes classification methods Training and predicting using hub nodes classification methods classif.hubc(fold, r, cuts,,, cv.repeat, Gsub = Gsub, =, ghub = ghub, hubs = hubs, nperm = nperm, node.ct = node.ct, Cs = Cs) fold number of -fold cross validation (CV) cuts list for randoml divide the training set in to --fold CV Gsub an adjacenc matri that represents the underling biological network. gene epression data. a factor of length p comprising the class labels. cv.repeat model for one CV training and predicting show debugging information in screen more or less. r repeat order for CV ghub Subgraph of hubs of graph Gs hubs Hubs in graph Gs nperm number of permutation test steps node.ct cut off value for select highl quantile nodes in a nwtwork. Defaults to 0.98). Cs Soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). fold auc train feat the recored for test fold The AUC values of test fold The tranined models for traning folds The feature selected b each b the train

9 classif.pac 9 References Talor et al.(2009)dnamic modularit in protein interaction networks predicts breast cancer outcome, Nat. Biotech.: doi: /nbt.1522 See cv.hubc #See cv.hubc classif.pac Training and predicting using PAC classification methods Training and predicting using PAC classification methods classif.pac(fold, cuts,,, cv.repeat, Gsub, int, = FALSE) fold cuts Gsub cv.repeat int number of -folds cross validation (CV) list for randoml divide the training set in to --folds CV an adjacenc matri that represents the underling biological network. gene epression data a factor of length p comprising the class labels. model for one CV training and predicting Intersect of genes in network and gene epression profile. show debugging information in screen or not. fold auc train feat the recored for test fold The AUC values of test fold The tranined models for traning folds The feature selected b each b the train

10 10 classif.stsvm References Lee E, Chuang H-Y, Kim J-W, Ideker T, Lee D (2008) Inferring Pathwa Activit toward Precise Disease Classification. PLoS Comput Biol 4(11): e doi: /journal.pcbi as cv.pac #see cv.pac classif.stsvm Training and predicting using stsvm classification methods Training and predicting using stsvm classification methods classif.stsvm(fold, cuts, e.sum,, p, a,, cv.repeat, =, Gsub=Gsub, op.method=op.method, op = op, aa = aa, dk = dk, dk.tf = dk.tf, seed = seed, Cs = Cs) fold cuts e.sum a p cv.repeat Gsub op.method op number of folds to perform list for randoml divide the training set in to --folds CV epression data epression data constant value of random walk kernel random walk step(s) of random walk kernel a factor of length p comprising the class labels. model for one CV training and predicting show debugging information in screen more or less. an adjacenc matri that represents the underling biological network. Method for selecet optimal feature subgoups: pt is permutation test, sp is span bound. optimal on top op

11 cv.aep 11 aa dk dk.tf seed Cs permutation test steps Random Walk Kernel matri of network cut off p-value of permutation test seed for random sampling. Soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). fold auc train feat the recored for test fold The AUC values of test fold The tranined models for traning folds The feature selected b each b the train References Yupeng Cun, Holger Frohlich (2013) Network and Data Integration for Biomarker Signature Discover via Network Smoothed T-Statistics. PLoS ONE 8(9): e doi: /journal.pone see cv.stsvm #see cv.stsvm cv.aep Cross validation for aepsvm (aepsvm) Cross validation for aepsvm (aepsvm) using SAM to select significant differential epressed genes cv.aep(,, folds = 10, repeats = 5, parallel = FALSE, cores = 2, = TRUE, Gsub = matri(1, 100, 100), Cs = 10^(-3:3), seed = 1234)

12 12 cv.aep folds repeats parallel cores Gsub Cs seed a p n matri of epression measurements with p samples and n genes. a factor of length p comprising the class labels. number of -folds cross validation (CV) number of CV repeat times paralle computing or not cores used in parallel computing show more results or not Adjacenc matri of Protein-protein interaction network soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). seed for random sampling. a LIST for Cross-Validation results auc fits feat labels The AUC values of each test fold The tranined models for traning folds The feature selected b each b the fits the original lables for training References Guo et al., Towards precise classification of cancers based on robust gene functional epression profiles. BMC Bioinformatics 2005, 6:58. librar(netclass) data(epr) data(ad.matri) <- epr$genes <- epr$ librar(kegg.db) #r.aep <- cv.aep([,1:500],, folds=3, repeats=1, parallel=false,cores=2, # Gsub=ad.matri,Cs=10^(-3:3),seed=1234,=TRUE)

13 cv.frsvm 13 cv.frsvm Cross validation for FrSVM Cross validation for FrSVM, an R algorithm, which integrates protein-protein interaction network information into gene selection for microarr classification cv.frsvm(,, folds = 10, Gsub = matri(1, 100, 100), repeats = 5, parallel = FALSE, cores = 2, = FALSE, d = 0.85, top.uper = 10, top.lower = 50, seed = 1234, Cs = 10^c(-3:3)) gene epression data class labels folds number of -folds cross validation (CV) Gsub Adjacenc matri of Protein-protein intersction network repeats number of CV repeat times parallel paralle computing or not cores cores used in parallel computing show more results or not d damping factor for GeneRank, defaults value is 0.5 top.uper the uper bound of top ranked genes top.lower the lower bound of top ranked genes seed Seed for random sampling. Cs soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). a LIST for Cross-Validation results auc fits feat labels The AUC values of each test fold The tranined models for traning folds The feature selected b each b the fits the original lables for training

14 14 cv.hubc References Yupeng Cun, Holger Frohlich (2012) Integrating Prior Knowledge Into Prognostic Biomarker Discover Based on Network Structure, arxiv: Winter C, Kristiansen G, Kersting S, Ro J, Aust D, et al. (2012) Google Goes Cancer: Improving Outcome Prediction for Cancer Patients b Network-Based Ranking of Marker Genes. PLoS Comput Biol 8(5): e librar(netclass) data(epr) data(ad.matri) <- epr$genes <- epr$ ### r.frsvm <-cv.frsvm([,1:200],, folds=3,gsub=ad.matri, repeats=1, parallel=false, cores=2, =TRUE,d=.85,top.uper=5,top.lower=15,seed=1234,Cs=10^c(-3:3)) cv.hubc Cross validation for hub nodes classification Cross validation for hub nodes classification, which described in Talor et al.(2009). cv.hubc(,, folds = 10, repeats = 5, parallel = TRUE, cores = NULL, = TRUE, nperm = 500, node.ct = 0.98, Gsub = matri(1, 100, 100), Gs = Gs, seed = 1234, Cs = 10^c(-3:3)) folds repeats parallel cores nperm a p n matri of epression measurements with p samples and n genes. a factor of length p comprising the class labels. number of -folds cross validation (CV) number of CV repeat times paralle computing or not cores used in parallel computing show more results or not number of permutation test steps node.ct cut off value for select highl quantile nodes in a nwtwork. Defaults to 0.98).

15 cv.pac 15 Gsub Gs seed Cs an adjacenc matri that represents the underling biological network. Undirected of graph with adjacenc matri Gsub. Seed for random sampling. Soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). auc fits feat labels The AUC values of each test fold The tranined models for traning folds The selected features of each training folds the original lables for training References Talor et al.(2009)dnamic modularit in protein interaction networks predicts breast cancer outcome, Nat. Biotech.: doi: /nbt.1522 data(ad.matri) #data(gs2) librar(netclass) data(epr) <- epr$genes <- epr$ # r.hubc <- cv.hubc(=, =, folds=3, repeats=1, parallel=false, cores=2, =TRUE, # nperm=2, Gsub=ad.matri,Gs=Gs2,node.ct=0.5,Cs=10^(-3:3)) cv.pac Cross validation for Pathwa Activities Classification(PAC) Cross validation for Pathwa Activities Classification(PAC) using Logistic regression model for classification. Implementation of the Pathwa Activities Classification b CROG algorithm. cv.pac(=, =, folds=10, repeats=5, parallel = TRUE, cores = NULL, =TRUE, Gsub=matri(1,100,100), seed=1234)

16 16 cv.pac folds repeats parallel cores Gsub seed a p n matri of epression measurements with p samples and n genes. a factor of length p comprising the class labels. number of -folds cross validation (CV) number of CV repeat times paralle computing or not cores used in parallel computing show debugging information in screen or not. Adjacenc matri of Protein-protein intersction network seed for random sampling. a LIST for Cross-Validation results auc fits feat labels The AUC values of each test fold The tranined models for traning folds The feature selected b each b the fits the original lables for training References Lee E, Chuang H-Y, Kim J-W, Ideker T, Lee D (2008) Inferring Pathwa Activit toward Precise Disease Classification. PLoS Comput Biol 4(11): e librar(netclass) data(epr) data(ad.matri) <- epr$genes <- epr$ librar(kegg.db) r.pac <- cv.pac(=, =, folds=3, repeats=1, parallel=false, cores=2, =TRUE, Gsub=ad.matri,seed=1234)

17 cv.stsvm 17 cv.stsvm Cross validation for smoothed t-statistic to select significant top ranked differential epressed genes Cross validation for smoothed t-statistic to select significant top ranked differential epressed genes cv.stsvm(=,.mi=null,=, folds=5,gsub=matri(1,100,100),op.method=c("pt","spb"), repeats=3, parallel=false, cores=2,=true, pt.pvalue=0.05,op=0.85, aa=1000,a=1,p=2,allf=true, seed=1234,cs=10^c(-3:3)).mi folds Gsub op.method repeats parallel cores pt.pvalue op aa a p allf seed Cs A p n matri of epression measurements with p samples and n genes. A p m matri of epression measurements with p samples and m mirnas. A factor of length p comprising the class labels. Folds number of folds to perform An adjacenc matri that represents the underling biological network. Method for selecet optimal feature subgoups: pt is permutation test, sp is span bound. Number of how often to repeat the -fold cross-validation Use parallel computing or not Number of cores will used when parallel is TRUE Show debugging information in screen more or less. Cut off p-value of permutation test Optimal on top op permutation test steps for permutation test (pt); low bounds top op constant value of random walk kernel random walk step(s) of random walk kernel Using all features (TRUE) or onl these genes mapped to prior information (FALSE). seed for random sampling. Soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). a LIST for Cross-Validation results auc fits feat labels The AUC values of each test fold The tranined models for traning folds The feature selected b each b the fits the original lables for training

18 18 EN2SY References Yupeng Cun, Holger Frohlich (2013) Network and Data Integration for Biomarker Signature Discover via Network Smoothed T-Statistics. PLoS ONE 8(9): e librar(netclass) data(epr) data(ad.matri) <- epr$genes <- epr$ r.stsvm <- cv.stsvm(=[,1:500],.mi=null,=,folds=3,gsub=ad.matri,op.method="pt", repeats=1, parallel=false, cores=2,=true,pt.pvalue=0.05,op=0.9, aa=5,a=1,p=2,allf=true, seed=1234,cs=10^(-3:3)) EN2SY An list for mapping gene entre ids to smbol ids An list for mapping gene entre ids to smbol ids Details An list for mapping gene Entre ID of Smbol ID

19 epr 19 epr Two epression profile matris and their labels Details Two epression profile matris and thei labels Two epression profile matris and thei labels of random samples. epr$genes is the epression profile with Entrez ID of genes; epr$ is labels of the epression profile. getgeneranking Get gene ranking based on generank algorithm. Get the ranking of differential epression of genes on graph using generank algorithm. getgeneranking( =, =, Gsub = Gsub, d = d) gene epression data class labels Gsub Adjacenc matri of Protein-protein intersction network d damping factor for GeneRank, defaults value is 0.5 r ranking of each gebes on graph as pgenerank

20 20 getgraphrank librar(netclass) data(epr) data(ad.matri) e.sum <- epr$genes <- epr$ #r= getgeneranking( = e.sum, =, Gsub = ad.matri, d = 0.5) getgraphrank Random walk kernel matri smoothing t-statistic Using Random walk kernel matri of network to smooth t-statistic of each gene getgraphrank( =, =, Gsub = Gsub, sca = TRUE) Gsub sca a matri of epression measurements with p samples and n genes. a factor of length p comprising the class labels. Random Walk Kernel matri of network Sacling data or not r return a smoothed t-statistic of each gene References Yupeng Cun, Holger Frohlich (2013) Network and Data Integration for Biomarker Signature Disvover via Network Smoothed T-Statistics as getgraphrank #See also \code{classf.stsvm}

21 Gs2 21 Gs2 An subgraph of hub nodes An subgraph of hub nodes, which using igraph to generate from hubs Details An adjacenc matri of hubs of a random graph was used to constructed a sub-graph of hubs using igraph pgenerank GeneRANK Ranking gene based on Googles s PageRank algorithm pgenerank(w, e, d, ma.degree = Inf) W adjacenc matri of graph e the fold change/ diffiencicial epression of genes d damping factor for GeneRank, defaults value is 0.5 ma.degree Ma degree of graph r ranking of each gebes on graph

22 22 pofhubs References Morrison, Julie L., et al. "GeneRank: using search engine technolog for the analsis of microarra eperiments." BMC bioinformatics 6.1 (2005): 233. Page, Lawrence, et al. "The PageRank citation ranking: bringing order to the web." (1999). as classif.frsvm # as {classif.frsvm} pofhubs Computing p value of hubs using the permutation test Computing p value of hubs using the permutation test pofhubs( =, =, ghub = ghub, hubs = hubs, nperm = nperm) ghub hubs nperm gene epression data a factor of length p comprising the class labels. Subgraph of hubs of graph Gs Hubs in graph Gs number of permutation test steps pval hub Permutation test Pvalues of each hub name of hubs # see \code{pofhubs}

23 predictaep 23 predictaep Predicting the test tdata using aep trained model Predicting the test data using aep trained model predictaep(train = train,,, = FALSE, Gsub = Gsub) train Gsub trained model gene epression data for testing class labels show debugging information in screen more or less. an adjacenc matri that represents the underling biological network. The value returned auc The AUC values of test fold as cv.aep #see cv.aep

24 24 predicthubc predictfrsvm Predicting the test data using frsvm trained model Predicting the test data using frsvm trained model predictfrsvm(train = train, =, =, = FALSE) train trained model epression data for testing class labels show debugging information in screen more or less. auc The AUC values of test fold as cv.frsvm #see cv.frsvm predicthubc Predicting the test data using hubc trained model Predicting the test data using hubc trained model predicthubc(train = train, =, =, = FALSE)

25 predictpac 25 train trainied model bases on hub nodes. gene epression data for predicting. Class labels show debugging information in screen more or less. The value returned auc The AUC values of test fold as cv.hubc #See cv.hubc predictpac Predicting the test data using pac trained model Predicting the test data using pac trained model predictpac(train = train, =, =, int = int, = FALSE) train int gene epression data for the testing data a factor of length p comprising the class labels. Intersect of genes in network and gene epression profile. show debugging information in screen or not.

26 26 predictstsvm The value returned auc The AUC values of test fold as cv.pac #see cv.pac predictstsvm Predicting the test data using stsvm trained model Predicting the test data using stsvm trained model predictstsvm(train = train, =, =, = ) train trained model epression data for testing Class labels show debugging information in screen more or less. The value returned auc The AUC values of test fold as cv.stsvm

27 probeset2pathwa 27 #see cv.stsvm probeset2pathwa Generae a mean gene epression of genes of each pathwa matri Generae a mean gene epression of genes of each pathwa matri probeset2pathwa( =, int = int, siggens = siggens) int siggens gene epression data common genes between pathwa genes and genes in gene epression profile significant gene epression using SAM methods kse an matri with n pathwas and p samples References Guo et al., Towards precise classification of cancers based on robust gene functional epression profiles. BMC Bioinformatics 2005, 6:58. as classif.aep

28 28 probeset2pathwatrain probeset2pathwatrain Search CROG in training data Search CROG in training data, and using these CORG set to make a matri for pathwas. probeset2pathwatrain( =, =, int = int) int gene epression data a factor of length p comprising the class labels. Common genes between gene epression data and interaction network. ap selectedgenes... top ranked pathas CROG genes References Lee E, Chuang H-Y, Kim J-W, Ideker T, Lee D (2008) Inferring Pathwa Activit toward Precise Disease Classification. PLoS Comput Biol 4(11): e doi: /journal.pcbi as pac.cv # as \name{pac.cv}

29 probeset2pathwatst 29 probeset2pathwatst Applied CROG to testing data Applied CORG and pathwas activities lists to make a matri for pathwas for test data. probeset2pathwatst( =, aptrain = aptrain) aptrain gene epression data PAC objet which contain CORG and pathwas activities lists of training data. ap top ranked pathas References Lee E, Chuang H-Y, Kim J-W, Ideker T, Lee D (2008) Inferring Pathwa Activit toward Precise Disease Classification. PLoS Comput Biol 4(11): e doi: /journal.pcbi as pac.cv, probeset2pathwatrain # as \code{pac.cv, probeset2pathwatrain}

30 30 train.aep train.aep Training the data using aep methods Training the data using aep methods train.aep( =, =, = FALSE, int = int, Gsub = Gsub, Cs = 10^(-3:3)) int Gsub Cs epression data for training a factor of length p comprising the class labels. show debugging information in screen more or less. Intersect of genes in network and gene epression profile. an adjacenc matri that represents the underling biological network. soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). The returned lists trained sig.genes The tranined models for traning folds The differential epressed feature References Guo et al., Towards precise classification of cancers based on robust gene functional epression profiles. BMC Bioinformatics 2005, 6:58. as cv.aep #see cv.aep

31 train.frsvm 31 train.frsvm Training the data using frsvm method Training the data using frsvm methods train.frsvm( =, =, = FALSE, Gsub = Gsub, d = 0.85, op = 10, aa = 50, Cs = 10^(-3:3)) Epression data for training Class labels show debugging information in screen more or less. Gsub an adjacenc matri that represents the underling biological network. d damping factor for GeneRank, defaults value is 0.5 op the uper bound of top ranked genes aa the lower bound of top ranked genes Cs soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). The value list returned train feat The tranined models for traning folds The feature selected b each b the train as cv.frsvm #see cv.frsvm

32 32 train.hubc train.hubc Predicting the data using hub nodes classification model Predicting the data using hub nodes classification model train.hubc( =, =, = FALSE, Gsub = Gsub, ghub = ghub, hubs = hubs, nperm = 500, node.ct = 0.95, Cs = 10^(-3:3)) gene epression data for training. Class labels show debugging information in screen more or less. Gsub an adjacenc matri that represents the underling biological network. ghub Subgraph of hubs of graph Gs hubs Hubs in graph Gs nperm number of permutation test steps node.ct cut off value for select highl quantile nodes in a nwtwork. Defaults to 0.98). Cs Soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). The list returned trained feat The tranined models for traning folds The feature selected b each b the train as cv.hubc #See cv.hubc

33 train.pac 33 train.pac Training the data using pac methods Training the data using pac methods train.pac( =, =, int = int, = FALSE, Gsub = Gsub) int Gsub gene epression data for the training data a factor of length p comprising the class labels. Intersect of genes in network and gene epression profile. show debugging information in screen or not. an adjacenc matri that represents the underling biological network. the value returned trained The tranined models for traning folds as cv.pac #see cv.pac

34 34 train.stsvm train.stsvm Training the data using stsvm methods Training the data using stsvm methods train.stsvm(=, =, =FALSE,Gsub=Gsub, op.method="sp", op=10,aa=100, dk=dk, dk.tf=0.05,seed = 1234,Cs=10^(-3:3),EN2SY=NULL) Gsub op.method op aa dk dk.tf seed Cs EN2SY epression data for training Class labels show debugging information in screen more or less. an adjacenc matri that represents the underling biological network. Method for selecet optimal feature subgoups: pt is permutation test, sp is span bound. optimal on top op permutation test steps Random Walk Kernel matri of network cut off p-value of permutation test seed for random sampling. Soft-margin tuning parameter of the SVM. Defaults to 10^c(-3:3). A list for mapping gene sbol ids or entez ids. The list returned trained feat The tranined models for traning folds The feature selected b each b the train See cv.stsvm #see cv.stsvm

35 Inde Topic FrSVM getgeneranking, 19 pgenerank, 21 Topic GeneRank train.frsvm, 31 Topic aep classif.aep, 6 cv.aep, 11 predictaep, 23 probeset2pathwa, 27 train.aep, 30 Topic biomarker discover, microarra classification, interaction network netclass-package, 2 Topic data ad.matri, 4 EN2SY, 18 epr, 19 Gs2, 21 Topic frsvm classif.frsvm, 7 cv.frsvm, 13 predictfrsvm, 24 train.frsvm, 31 Topic hubc classif.hubc, 8 cv.hubc, 14 pofhubs, 22 predicthubc, 24 train.hubc, 32 Topic pac classif.pac, 9 cv.pac, 15 predictpac, 25 probeset2pathwatrain, 28 probeset2pathwatst, 29 train.pac, 33 Topic stsvm calc.diffusionkernelp, 5 cv.stsvm, 17 getgraphrank, 20 Topic stsvm classif.stsvm, 10 predictstsvm, 26 train.stsvm, 34 ad.matri, 4 calc.diffusionkernelp, 5 classif.aep, 6 classif.frsvm, 7 classif.hubc, 8 classif.pac, 9 classif.stsvm, 10 cv.aep, 11 cv.frsvm, 13 cv.hubc, 14 cv.pac, 15 cv.stsvm, 17 EN2SY, 18 epr, 19 getgeneranking, 19 getgraphrank, 20 Gs2, 21 netclass (netclass-package), 2 netclass-package, 2 pgenerank, 21 pofhubs, 22 predictaep, 23 predictfrsvm, 24 predicthubc, 24 predictpac, 25 predictstsvm, 26 probeset2pathwa, 27 probeset2pathwatrain, 28 35

36 36 INDEX probeset2pathwatst, 29 train.aep, 30 train.frsvm, 31 train.hubc, 32 train.pac, 33 train.stsvm, 34