PLIERsparse: sparse PLIER function (experimental)

View source: R/Allfuncs.R

PLIERsparseR Documentation

sparse PLIER function (experimental)

Description

sparse PLIER function (experimental)

Usage

PLIERsparse(
  data,
  priorMat,
  svdres = NULL,
  k = NULL,
  L1 = NULL,
  L2 = NULL,
  L3 = NULL,
  frac = 0.7,
  max.iter = 350,
  trace = F,
  scale = T,
  Chat = NULL,
  maxPath = 10,
  doCrossval = T,
  penalty.factor = rep(1, ncol(priorMat)),
  glm_alpha = 0.9,
  minGenes = 10,
  tol = 1e-06,
  seed = 123456,
  allGenes = F,
  rseed = NULL,
  pathwaySelection = c("complete", "fast"),
  sparseL = 0.01,
  sparseType = "SCAD"
)

Arguments

data

the data to be processed with genes in rows and samples in columns. Should be z-scored or set scale=T

priorMat

the binary prior information matrix with genes in rows and pathways/genesets in columns

svdres

Pre-computed result of the svd decomposition for data

k

The number of latent variables to return, leave as NULL to be set automatically using the num.pc "elbow" method

L1

L1 constant, leave as NULL to automatically select a value

L2

L2 constant, leave as NULL to automatically select a value

L3

L3 constant, leave as NULL to automatically select a value. Sparsity in U should be instead controlled by setting frac

frac

The fraction of LVs that should have at least 1 prior inforamtion association, used to automatically set L3

max.iter

Maximum number of iterations to perform

trace

Display progress information

scale

Z-score the data before processing

Chat

A ridge inverse of priorMat, used to select active pathways, expensive to compute so can be precomputed when running PLIER multiple times

maxPath

The maximum number of active pathways per latent variable

doCrossval

Whether or not to do real cross-validation with held-out pathway genes. Alternatively, all gene annotations are used and only pseudo-crossvalidation is done. The latter option may be preferable if some pathways of interest have few genes.

penalty.factor

A vector equal to the number of columns in priorMat. Sets relative penalties for different pathway/geneset subsets. Lower penalties will make a pathway more likely to be used. Only the relative values matter. Internally rescaled.

glm_alpha

Set the alpha for elastic-net

minGenes

The minimum number of genes a pathway must have to be considered

tol

Convergence threshold

seed

Set the seed for pathway cross-validation

allGenes

Use all genes. By default only genes in the priorMat matrix are used.

rseed

Set this option to use a random initialization, instead of SVD

pathwaySelection

Pathways to be optimized with elstic-net penalty are preselected based on ridge regression results. "Complete" uses all top pathways to fit individual LVs. "Fast" uses only the top pathways for the single LV in question.

sparseL

the lambda for sparsity on Z, default 0.02

sparseType

sparsity inducing penalty to use (SCAD or L1)


wgmao/PLIER documentation built on Sept. 1, 2024, 10:25 p.m.