ENIGMA_L2_max_norm: ENIGMA maximum L2 norm version

In WWXkenmo/ENIGMA_test: dEcoNvolutIon method based on reGularized MAtrix completion

ENIGMA maximum L2 norm version

Description

maximum L2 norm version is the default version of ENIGMA, which is the regularized weighted matrix completion to constraint the maximum L2 norm of inferred cell type-specific gene expression matrix.

Usage

ENIGMA_L2_max_norm(
  object,
  do_cpm = TRUE,
  alpha = 0.5,
  tao_k = 0.01,
  beta = 0.1,
  epsilon = 0.001,
  max.iter = 1000,
  verbose = FALSE,
  pos = TRUE,
  calibrate = TRUE,
  Norm.method = "frac",
  preprocess = "sqrt",
  loss_his = TRUE,
  model_tracker = FALSE,
  model_name = NULL,
  X_int = NULL,
  Normalize = TRUE
)

Arguments

object	ENIGMA object
do_cpm	if perform cpm normalization to the data, strongly recommand to use to make sure each sample numerical value scale is comparable. Default = TRUE
alpha	ENIGMA is a multi-objective optimization problem involve two object function, the distance function between observed bulk RNA-seq and reconstitute RNA-seq generated by weighted combination of CSE, and the distance function beween average CSE expression and cell type reference matrix. The alpha is used to determine weights of these two objects. If the alpha gets larger, the optimization attach greater importance on the the first object. Default: 0.5
tao_k	The step size of each round of gradient decent. Default: 0.01
beta	The regularization parameter to penalize the weight (deconvoluted expression) matrices from being too large. Default: 0.1
epsilon	Determine the stop condition in CSE updating. Default: 0.001
max.iter	The maximum number of iterations. Default: 1000
verbose	Rreturn the information after each step of processing. Default: TRUE
pos	Set all entries in CSE is positive. Default: TRUE
calibrate	calibrate the inferred CSE into input bulk gene expression scale. Default: TRUE
Norm.method	Method used to perform normalization. User could choose PC, frac or quantile
preprocess	Method used to perform variance stablization preprocessing. User could choose none, sqrt or log
loss_his	save the loss value of each round of iteration.
model_tracker	save the model in returned object
model_name	name of the model
X_int	initialization for CSE profiles, an array object with three dimensions (the number of genes * the number of samples * the number of cell types), if user input a matrix (the number of genes * the number of samples), each cell type would be assigned the same start matrix.
force_normalize	when alpha >= 0.9 or profile matrix is not generated from S-mode batch effect correction, ENIGMA would not perform normalization, if user still want to perform normalization, set force_normalize=TRUE. Default: FALSE

Value

ENIGMA object where object@result_CSE contains the inferred CSE profile, object@result_CSE_normalized would contains normalized CSE profile, object@loss_his would contains the loss values of object functions during model training. If model_tracker = TRUE, then above results would be saved in the object@model.

Examples

## Not run: 
egm = ENIGMA_l2max_norm(egm,model_tracker = TRUE, Norm.method="quantile")
egm@result_CSE
egm@result_CSE_normalized

## End(Not run)

WWXkenmo/ENIGMA_test documentation built on March 17, 2023, 4:56 a.m.