BASiCS_DetectVG: Detection method for highly (HVG) and lowly (LVG) variable...

In catavallejos/BASiCS: Bayesian Analysis of Single-Cell Sequencing data

Detection method for highly (HVG) and lowly (LVG) variable genes

Description

Functions to detect highly and lowly variable genes. If the BASiCS_Chain object was generated using the regression approach, BASiCS finds the top highly variable genes based on the posteriors of the epsilon parameters. Otherwise, the old approach is used, which initially performs a variance decomposition.

Usage

BASiCS_DetectVG(
  Chain,
  Task = c("HVG", "LVG"),
  PercentileThreshold = NULL,
  VarThreshold = NULL,
  ProbThreshold = 2/3,
  EpsilonThreshold = NULL,
  EFDR = 0.1,
  OrderVariable = c("Prob", "GeneIndex", "GeneName"),
  Plot = FALSE,
  MinESS = 100,
  ...
)

BASiCS_DetectLVG(Chain, ...)

BASiCS_DetectHVG(Chain, ...)

Arguments

Chain	an object of class BASiCS_Chain
Task	Search for highly variable genes (Task="HVG") or lowly variable genes (Task="LVG").
PercentileThreshold	Threshold to detect a percentile of variable genes (must be a positive value, between 0 and 1). Default: PercentileThreshold = NULL.
VarThreshold	Variance contribution threshold (must be a positive value, between 0 and 1). This is only used when the BASiCS non-regression model was used to generate the Chain object. Default: VarThreshold = NULL.
ProbThreshold	Optional parameter. Posterior probability threshold (must be a positive value, between 0 and 1). If EFDR = NULL, the posterior probability threshold for the test will be set to ProbThreshold.
EpsilonThreshold	Threshold for residual overdispersion above which
EFDR	Target for expected false discovery rate related to HVG/LVG detection. If EFDR = NULL, EFDR calibration is not performed and the posterior probability threshold is set equal to ProbThreshold. Default EFDR = 0.10.
OrderVariable	Ordering variable for output. Possible values: 'GeneIndex', 'GeneName' and 'Prob'. Default ProbThreshold = 'Prob'
Plot	If Plot = TRUE error control and expression versus HVG/LVG probability plots are generated.
MinESS	The minimum effective sample size for a gene to be included in the HVG or LVG tests. This helps to remove genes with poor mixing from detection of HVGs/LVGs. Default is 100. If set to NA, genes are not checked for effective sample size the tests are performed.
...	Graphical parameters (see par).

Details

In some cases, the EFDR calibration step may fail to find probability threshold that controls the EFDR at the chosen level. In cases like

See vignette

Value

An object of class BASiCS_ResultVG.

Author(s)

Catalina A. Vallejos [e-mail]

Nils Eling [e-mail]

References

Vallejos, Marioni and Richardson (2015). PLoS Computational Biology.

See Also

BASiCS_Chain

Examples

# Loads short example chain (non-regression implementation)
data(ChainSC)

# Highly and lowly variable genes detection (within a single group of cells)
DetectHVG <- BASiCS_DetectHVG(ChainSC, VarThreshold = 0.60,
                              EFDR = 0.10, Plot = TRUE)
DetectLVG <- BASiCS_DetectLVG(ChainSC, VarThreshold = 0.40,
                              EFDR = 0.10, Plot = TRUE)
                              
# Loads short example chain (regression implementation)
data(ChainSCReg)

# Highly and lowly variable genes detection (within a single group of cells)
DetectHVG <- BASiCS_DetectHVG(ChainSCReg, PercentileThreshold = 0.90,
                              EFDR = 0.10, Plot = TRUE)
DetectLVG <- BASiCS_DetectLVG(ChainSCReg, PercentileThreshold = 0.10,
                              EFDR = 0.10, Plot = TRUE)

## Highly and lowly variable genes detection based on residual overdispersion
## threshold
DetectHVG <- BASiCS_DetectHVG(ChainSCReg, EpsilonThreshold = log(2), Plot = TRUE)
DetectLVG <- BASiCS_DetectLVG(ChainSCReg, EpsilonThreshold = -log(2), Plot = TRUE)

catavallejos/BASiCS documentation built on March 27, 2024, 12:49 a.m.