Pipe operator

Description

See magrittr::%>% for details.

Usage

lhs %>% rhs

Arguments

Value

The result of calling 'rhs(lhs)'.

A synthetic mRNA expression dataset.

Description

A matrix containing simulated mRNA expression levels for 358 subjects (rows) and 500 features (columns).

Usage

X1

Format

An object of class matrix (inherits from array) with 358 rows and 500 columns.

A synthetic miRNA expression dataset.

Description

A matrix containing simulated miRNA expression levels for 358 subjects (rows) and 100 features (columns).

Usage

X2

Format

An object of class matrix (inherits from array) with 358 rows and 100 columns.

A synthetic phenotype dataset.

Description

A matrix containing simulated quantitative phenotype measures for 358 subjects (rows).

Usage

Y

Format

An object of class matrix (inherits from array) with 358 rows and 1 columns.

Aggregate and Save Cross-validation Result for Single-omics Analysis

Description

Saves cross-validation results in a table with the user-defined directory and outputs penalty term with the highest testing canonical correlation, lowest prediction error, and lowest scaled prediction error.

Usage

aggregateCVSingle(
  CVDir,
  SCCAmethod = "SmCCA",
  K = 5,
  NumSubsamp = 500,
  verbose = FALSE
)

Arguments

Value

A vector of length 3 with indices of the penalty term that (1) maximize the testing canonical correlation, (2) minimize the prediction error and (3) minimize the scaled prediction error.

Evaluation of Binary Classifier with Different Evaluation Metrics

Description

Evaluate binary classifier's performance with respect to user-selected metric (accuracy, auc score, precision, recall, f1 score) for binary phenotype.

Usage

classifierEval(
  obs,
  pred,
  EvalMethod = "accuracy",
  BinarizeThreshold = 0.5,
  print_score = TRUE
)

Arguments

Value

An evaluation score corresponding to the selected metric.

Examples

# simulate observed binary phenotype
obs <- rbinom(100,1,0.5)
# simulate predicted probability
pred <- runif(100, 0,1)
# calculate the score
pred_score <- classifierEval(obs, pred, EvalMethod = 'f1', print_score = FALSE)

preprocess a omics dataset before running omics SmCCNet

Description

Data preprocess pipeline to: (1) filter by coefficient of variation (cv), (2) center or scale data and (3) adjust for clinical covariates.

Usage

dataPreprocess(
  X,
  covariates = NULL,
  is_cv = FALSE,
  cv_quantile = 0,
  center = TRUE,
  scale = TRUE
)

Arguments

Value

Processed omics data with the size of nxp.

Examples

X1 <- as.data.frame(matrix(rnorm(600, 0, 1), nrow = 60))
covar <- as.data.frame(matrix(rnorm(120, 0, 1), nrow = 60))
processed_data <- dataPreprocess(X = X1, covariates = covar, is_cv = TRUE, 
cv_quantile = 0.5, center = TRUE, scale = TRUE)

Automated SmCCNet to Streamline the SmCCNet Pipeline

Description

Automated SmCCNet automatically identifies the project problem (single-omics vs multi-omics), and type of analysis (CCA for quantitative phenotype vs. PLS for binary phenotype) based on the input data that is provided. This method automatically preprocesses data, chooses scaling factors, subsampling percentage, and optimal penalty terms, then runs through the complete SmCCNet pipeline without the requirement for users to provide additional information. This function will store all the subnetwork information to a user-defined directory, as well as return all the global network and evaluation information. Refer to the automated SmCCNet vignette for more information.

Usage

fastAutoSmCCNet(
  X,
  Y,
  AdjustedCovar = NULL,
  preprocess = FALSE,
  Kfold = 5,
  EvalMethod = "accuracy",
  subSampNum = 100,
  DataType,
  BetweenShrinkage = 2,
  ScalingPen = c(0.1, 0.1),
  CutHeight = 1 - 0.1^10,
  min_size = 10,
  max_size = 100,
  summarization = "NetSHy",
  saving_dir = tempdir(),
  ncomp_pls = 3,
  tuneLength = 5,
  tuneRangeCCA = c(0.1, 0.5),
  tuneRangePLS = c(0.5, 0.9),
  seed = 123,
  verbose = FALSE
)

Arguments

Value

This function returns the global adjacency matrix, omics data details, network clustering outcomes, and cross-validation results. Pruned subnetwork modules are saved in the directory specified by the user.

Examples

 library(SmCCNet)
 set.seed(123)
 Y <- rnorm(50)
 X1 <- matrix(rnorm(2500), nrow = 50, ncol = 50)
 colnames(X1) <- paste0("Gene_", 1:50)
 X1[, 1:10] <- X1[, 1:10] + matrix(rep(Y, 10), ncol = 10)
 X2 <- matrix(rnorm(1000), nrow = 50, ncol = 20)
 colnames(X2) <- paste0("miRNA_", 1:20)
 X2[, 1:5] <- X2[, 1:5] + matrix(rep(Y, 5), ncol = 5)
 Y_binary <- ifelse(Y > median(Y), 1, 0)
 ## single-omics CCA
 result <- fastAutoSmCCNet(X = list(X1), Y = Y,
                           Kfold = 3, preprocess = FALSE,
                           subSampNum = 10,
                           DataType = c('Gene'),
                           saving_dir = tempdir(),
                           summarization = 'NetSHy',
                           CutHeight = 1 - 0.1^10,
                           min_size = 5)
 ## single-omics PLS
 result <- fastAutoSmCCNet(X = list(X1),
                           Y = as.factor(Y_binary),
                           Kfold = 3, subSampNum = 10,
                           DataType = c('Gene'),
                           saving_dir = tempdir(),
                           EvalMethod = 'auc',
                           summarization = 'NetSHy',
                           CutHeight = 1 - 0.1^10,
                           min_size = 5, ncomp_pls = 3)

Calculate similarity matrix based on canonical weights.

Description

Compute the similarity matrix based on the outer products of absolute canonical correlation weights, can be used for both single and multi-omics setting.

Usage

getAbar(Ws, FeatureLabel = NULL)

Arguments

Value

A p\times p symmetric non-negative matrix.

Examples

w <- matrix(rnorm(6), nrow = 3)
Ws <- apply(w, 2, function(x)return(x/sqrt(sum(x^2))))
abar <- getAbar(Ws,  FeatureLabel = c('omics1', 'omics2', 'omics3'))

Internal functions called by getRobustPseudoWeights_single.

Description

Internal functions called by getRobustPseudoWeights_single.

Usage

getCCAout_single(X1, Trait, Lambda1, trace = FALSE)

Arguments

Canonical Correlation Value for SmCCA

Description

Calculate canonical correlation value for SmCCA given canonical weight vectors and scaling factor

Usage

getCanCorMulti(X, CCcoef, CCWeight, Y)

Arguments

Value

A numeric value of the total canonical correlation

Examples

library(SmCCNet)
data("ExampleData")
getCanCorMulti(list(X1,X2), CCcoef = c(1,1,1), 
CCWeight = list(rnorm(500,0,1), rnorm(100,0,1)), Y = Y)

Get Canonical Weight SmCCA Algorithm (No Subsampling)

Description

Run Sparse multiple Canonical Correlation Analysis (SmCCA) and return canonical weight vectors.

Usage

getCanWeightsMulti(
  X,
  Trait = NULL,
  Lambda,
  CCcoef = NULL,
  NoTrait = TRUE,
  trace = FALSE,
  TraitWeight = FALSE
)

Arguments

Value

A canonical weight vector with size of p by 1.

Examples

 
 x1 <- matrix(rnorm(1000), nrow = 50)
 x2 <- matrix(rnorm(1000), nrow = 50)
 y <- matrix(rnorm(50), nrow = 50)
 X <- list(x1,x2)
 result <- getCanWeightsMulti(X, Trait = y, Lambda = c(0.5,0.5), NoTrait = FALSE)
 result <- getCanWeightsMulti(X, Trait = NULL, Lambda = c(0.5,0.5), NoTrait = TRUE)
 cccoef <- c(1,10,10)
 result <- getCanWeightsMulti(X, Trait = y, CCcoef = cccoef, 
                              Lambda = c(0.5,0.5), NoTrait = FALSE)
                              

Extract Omics Modules based on Similarity Matrix.

Description

Apply hierarchical tree cutting to the similarity matrix and extract multi/single-omics network modules.

Usage

getOmicsModules(Abar, CutHeight = 1 - 0.1^10, PlotTree = TRUE)

Arguments

Value

A list of multi/single-omics modules.

Examples

set.seed(123)
w <- rnorm(5)
w <- w/sqrt(sum(w^2))
feature_name <- paste0('feature_', 1:5)
abar <- getAbar(w, FeatureLabel = feature_name)
modules <- getOmicsModules(abar, CutHeight = 0.5)

Run Sparse multiple Canonical Correlation Analysis and Obtain Canonical Weights (with Subsampling)

Description

SmCCNet algorithm with multi-omics data and quantitative phenotype. Calculate the canonical weights for SmCCA.

Usage

getRobustWeightsMulti(
  X,
  Trait,
  Lambda,
  s = NULL,
  NoTrait = FALSE,
  SubsamplingNum = 1000,
  CCcoef = NULL,
  trace = FALSE,
  TraitWeight = FALSE
)

Arguments

Value

A canonical correlation weight matrix with p = \sum_{i} p_i rows, where p_i is the number of features for the ith omics. Each column is the canonical correlation weights based on subsampled features. The number of columns is SubsamplingNum.

Examples

## For illustration, we only subsample 5 times.
set.seed(123)
X1 <- matrix(rnorm(600,0,1), nrow = 60)
X2 <- matrix(rnorm(600,0,1), nrow = 60)
Y <- matrix(rnorm(60,0,1), nrow = 60)
# Unweighted SmCCA
result <- getRobustWeightsMulti(X = list(X1, X2), Trait = Y, NoTrait = FALSE,
Lambda = c(0.5, 0.5),s = c(0.7, 0.7), SubsamplingNum = 20)
  

Run Sparse multiple Canonical Correlation Analysis and Obtain Canonical Weights (with Subsampling)

Description

SmCCNet algorithm with multi-omics data and binary phenotype. This is a stepwise approach (1) use SmCCA to identify relationship between omics (exlude phenotype), (2) within highly connected omics features selected in step 1, identify relationship between these selected omics features and phenotype of interest with sparse PLS. First, it computes PLSDA by assuming outcome is continuous to extract multiple latent factors, then uses latent factors to fit logistic regression, and weight latent factor by regression parameters. Refer to multi-omics vignette for more detail.

Usage

getRobustWeightsMultiBinary(
  X,
  Y,
  Between_Discriminate_Ratio = c(1, 1),
  SubsamplingPercent = NULL,
  CCcoef = NULL,
  LambdaBetween,
  LambdaPheno = NULL,
  SubsamplingNum = 1000,
  ncomp_pls = 3,
  EvalClassifier = FALSE,
  testData = NULL,
  verbose = FALSE
)

Arguments

Value

Canonical weight matrix or latent projections depending on EvalClassifier.

Single-omics SmCCA with Quantitative Phenotype

Description

Compute aggregated (SmCCA) canonical weights for single omics data with quantitative phenotype (subampling enabled).

Usage

getRobustWeightsSingle(
  X1,
  Trait,
  Lambda1,
  s1 = 0.7,
  SubsamplingNum = 1000,
  trace = FALSE
)

Arguments

Value

A canonical correlation weight matrix with p_1 rows. Each column is the canonical correlation weights based on subsampled X1 features. The number of columns is SubsamplingNum.

Examples

## For illustration, we only subsample 5 times.
set.seed(123)

# Single Omics SmCCA
W1 <- getRobustWeightsSingle(X1, Trait = Y, Lambda1 = 0.05,
  s1 = 0.7, 
  SubsamplingNum = 5, trace = FALSE)
  
  

Single-omics SmCCA with Binary Phenotype

Description

Compute aggregated (SmCCA) canonical weights for single omics data with quantitative phenotype (subampling enabled).

Usage

getRobustWeightsSingleBinary(
  X1,
  Trait,
  Lambda1,
  s1 = 0.7,
  SubsamplingNum = 1000,
  K = 3
)

Arguments

Value

A partial least squared weight matrix with p_1 rows. Each column is the canonical correlation weights based on subsampled X1 features. The number of columns is SubsamplingNum.

Examples

X <- matrix(rnorm(600,0,1), nrow = 60)
Y <- rbinom(60,1,0.5)
Ws <- getRobustWeightsSingleBinary(X1 = X, Trait = as.matrix(Y), Lambda1 = 0.8, 
0.7, SubsamplingNum = 10)

Prunes Subnetwork and Return Final Pruned Subnetwork Module

Description

Prunes subnetworks with network pruning algorithm (see multi-omics vignette for detail), and save the final pruned subnetwork to the user-defined directory. The final subnetwork is an .Rdata file with a name 'size_m_net_ind.Rdata', where m is the final pruned network size, and ind is the index of the subnetwork module after hierarchical clustering.

Usage

networkPruning(
  Abar,
  CorrMatrix,
  data,
  Pheno,
  type,
  ModuleIdx,
  min_mod_size = 10,
  max_mod_size,
  damping = 0.9,
  method = "NetSHy",
  saving_dir = tempdir(),
  verbose = FALSE
)

Arguments

Value

A file stored in the user-defined directory, which contains the following: (1) correlation_sub: correlation matrix for the subnetwork. (2) M: adjacency matrix for the subnetwork. (3) omics_corelation_data: individual molecular feature correlation with phenotype. (4) pc_correlation: first 3 PCs correlation with phenotype. (5) pc_loading: principal component loadings. (6) pca_x1_score: principal component score and phenotype data. (7) mod_size: number of molecular features in the subnetwork. (8) sub_type: type of feature for each molecular features.

Examples

library(SmCCNet)
set.seed(123)
w <- rnorm(20)
w <- w/sqrt(sum(w^2))
X1 <- matrix(rnorm(1000,0,1),nrow = 50)
Y <- matrix(rnorm(50,0,1),nrow = 50)
labels <- paste0('feature_', 1:20)
colnames(X1) <- labels
abar <- getAbar(w, FeatureLabel = labels)
modules <- getOmicsModules(abar, CutHeight = 0.1)
x <- X1
corr <- stats::cor(x)
type <- c(rep(1,20))
# display only example
networkPruning(abar, corr, data = x, Pheno = Y, type = type,
 ModuleIdx = 1,  min_mod_size = 3, max_mod_size = 10, method = 'NetSHy', saving_dir = tempdir()
 )

Scaling Factor Input Prompt

Description

Input the vector of the annotation of each type of dataset in the data list X (e.g., c('gene', 'protein')), and return prompt ask the user to supply the scaling factor for SmCCNet algorithm to prioritize the correlation structures of interest. All scaling factor values supplied should be numeric and nonnegative.

Usage

scalingFactorInput(DataType = NULL)

Arguments

Value

A numeric vector of scaling factors.

Examples

if(interactive()){scalingFactorInput(c('gene','mirna', 'phenotype'))}

NetSHy Summarization Score

Description

Implement NetSHy network summarization via a hybrid approach (Vu et al.,) to summarize network by considering the network topology with Laplacian matrix.

Usage

summarizeNetSHy(X, A, npc = 1)

Arguments

Value

A list consists of (1) subject-level network summarization score, (2) principal component importance information: standard deviation, percent of variance explained, and cumulative proportion of variance explained, and (3) principal component feature-level loadings.

References

Vu, Thao, Elizabeth M. Litkowski, Weixuan Liu, Katherine A. Pratte, Leslie Lange, Russell P. Bowler, Farnoush Banaei-Kashani, and Katerina J. Kechris. "NetSHy: network summarization via a hybrid approach leveraging topological properties." Bioinformatics 39, no. 1 (2023): btac818.

Examples

# simulate omics data
OmicsData <- matrix(rnorm(200,0,1), nrow = 10, ncol = 20)
# simulate omics adjacency matrix
set.seed(123)
w <- rnorm(20)
w <- w/sqrt(sum(w^2))
featurelabel <- paste0('omics',1:20)
abar <- getAbar(w, FeatureLabel = featurelabel)
# extract NetSHy summarization score
netshy_score <- summarizeNetSHy(OmicsData, abar)