In Jfortin1/neuroCombat_Rpackage: Harmonization of multi-site imaging data with ComBat

Intro

Vignette describing a function in development, neuroCombatFromTraining. The function applies ComBat scanner correction factors estimated from a training dataset to a test dataset. Scanners/sites of the test dataset must be included in the training dataset.

Example dataset

library(neuroCombat)
library(neuroCombatData)
se <- seCorticalThickness
se

We use the example dataset stored in the neuroCombatData package: 10 scans from 2 scanners, with 200 imaging features. Scans come from either a normal or disease brain.

Harmonization

data <- assays(se)[["freesurfer"]]
mod <- model.matrix(~colData(se)$gender+colData(se)$age)
results <- neuroCombat(dat=data,
                       batch=colData(se)$site,
                       mod=mod)

names(results)

neuroCombat::drawPriorDelta(results$estimates, xlim=c(0,2.5))
neuroCombat::drawPriorGamma(results$estimates, xlim=c(-1.5,1.5))

data_standardized <- results[["dat.standardized"]]
col <- as.numeric(as.factor(colData(se)$site))
boxplot(data_standardized, ylim=c(-2,2),col=col)

data_norm <- results[["dat.combat"]]
data_norm <- standardizeData(data_norm,
                             batch=colData(se)$site,
                             mod=mod)
boxplot(data_norm, ylim=c(-2,2),col=col)

References

Jfortin1/neuroCombat_Rpackage documentation built on Nov. 24, 2024, 9:23 a.m.