LOO: Leave-one-out analysis for quantitative evaluation of the...

In snaketron/IgGeneUsage: Differential gene usage in immune repertoires

Leave-one-out analysis for quantitative evaluation of the probability of DGU

Description

IgGeneUsage detects differential gene usage (DGU) in immune repertoires that belong to two biological conditions.

To quantify the robustness of the estimated probability of DGU (pmax), IgGeneUsage has a built-in procedure for a fully Bayesian leave-one-out (LOO) analysis. In each LOO step we discard the data of one of the repertoires, and use the remaining data to analyze for DGU with IgGeneUsage. In each step we record pmax for all genes. Finally, we evaluate the variability of pmax for a given across the different LOO steps. Low variability in pmax: robust DGU; high variability: unrobust DGU.

For datasets that include many repertoires (e.g. 100) LOO can be computationally costly.

Usage

LOO(ud, 
    mcmc_warmup, 
    mcmc_steps,
    mcmc_chains, 
    mcmc_cores, 
    hdi_lvl,
    adapt_delta, 
    max_treedepth,
    paired = FALSE)

Arguments

ud	Data.frame with 4 columns: 'individual_id' = character, name of the donor (e.g. Pt1) 'condition' = character, name of biological conditions (e.g. tumor) 'gene_name' = character, Ig gene name (e.g. IGHV1-69) 'gene_usage_count' = number, frequency (=usage) of rearrangements from individual_id x condition x gene_name [optional] 'replicate' = character or number. Replicate id, if more than one repertoire (biological replicates) is available per individual ud can also be be a SummarizedExperiment object. See dataset 'data(Ig_SE)' for more information.
mcmc_chains, mcmc_warmup, mcmc_steps, mcmc_cores	Number of MCMC chains (default = 4), number of cores to use (default = 1), length of MCMC chains (default = 1,500), length of adaptive part of MCMC chains (default = 500).
hdi_lvl	Highest density interval (HDI) (default = 0.95).
adapt_delta	MCMC setting (default = 0.95).
max_treedepth	MCMC setting (default = 12).
paired	should a paired samples differential Ig gene analaysis be performed (default = FALSE)?

Details

IgGeneUsage invokes the function DGU in each LOO step. For more details see help for DGU or vignette 'User Manual: IgGeneUsage'.

Value

loo

DGU statistics for each Ig gene for specific LOO step: es = effect size statistics: mean, median, standard error (se), standard deviation (sd), L/H (low/high boundary of HDI) contrast = direction of the effect pmax = DGU probability loo_id (LOO step) Neff (effective sample size), Rhat (potential scale reduction factor)

Author(s)

Simo Kitanovski <[e-mail]>

See Also

DGU, Ig, IGHV_Epitopes, IGHV_HCV, Ig_SE, d_zibb_1, d_zibb_2, d_zibb_3

Examples

# input data:
data("Ig", package = "IgGeneUsage")
head(Ig)

# run leave-one-out (LOO)
L <- LOO(ud = Ig,
         mcmc_warmup = 500,
         mcmc_steps = 2000,
         mcmc_chains = 3,
         mcmc_cores = 1,
         hdi_lvl = 0.95,
         adapt_delta = 0.99,
         max_treedepth = 10,
         paired = FALSE)

# how many LOOs?  
names(L)

# elements in first LOO, see vignette about how to extract results
names(L[[1]])

snaketron/IgGeneUsage documentation built on April 23, 2024, 2:22 a.m.