metsInt: Stratified and interaction metabolite association models
In buddha2490/BERGMets: Metabolomics data and analysis in the BERG

Description Usage Arguments Details Value Author(s) See Also Examples

This function will run stratified or interaction models for metabolites with a continuous or binary outcome. The choice of stratified vs simple interaction depends on the type of interaction variable. See details.

1 2	metsInt(dat,biochem,outcome,metabolites,intvar, covariates=NULL,normalize=T)

`dat`	data frame containing your metabolites and covariates
`biochem`	The biochemical metadata file. Typically this is in the $biochem element of the metabolites list. You can subset this data frame any way you wish, some variables are probably unneeded. Recommend that you subset to only include COMP_ID, BIOCHEMICAL, SUPER_PATHWAY, SUB_PATHWAY.
`outcome`	Outcome for the analysis. Logistic regression must be coded [0,1]; linear regression must be continuous numeric. Function will identify which is which and will provide an error message if you code it wrong.
`compid`	Vector of COMP_IDs used for the models
`intvar)`	Name of interaction variable. If class(intvar)=="numeric", simple interaction models will be fit coded as metabolite*intvar. If class(intvar)=="factor", the function will return estimates for the metabolites stratified by the intvar.
`covariates`	Optional. Vector of covariate names from the dat data frame.
`normalize`	Logical. Set to TRUE if you haven't already mean-centered and glog transformed your metabolite data. Default is TRUE.

In order to use metsInt(), users need to take care to code their variables appropriately. The outcome variable determines the type of model fit: a continuous outcome variable will fit a linear model; a binary outcome will fit a logistic model. If the variables are coded incorrectly, an error/warning will be returned.

Users also must code the intvar correctly. If intvar is a factor variable, the function will return metabolite associations stratified across each level of of intvar. If intvar is coded as a numeric variable, the function will return associations for the interaction term. Both models will return a p-value for the interaction based on a likelihood ratio test comparing the interaction model with a reduced model.

A data frame containing the results. Results are merged with the biochem-metadata file and estimates/pvalues are nicely formatted.

Stratified models will return estimates across each level of the interaction variables.

Interaction models will return only the estimate and p-values associated with the interaction term.

Both will return an overall p-value for the interaction.

Brian Carter

normalizeMets, metsAssoc, metsModels,

# Data
breast_metabolomics <- getMetabolites("breast_metabolomics")
data(survey)
df <- left_join(survey,breast_metabolomics$metabolites,"ID")
covars <- c("AGE_INT","LASTATE")
comp.id <- names(breast_metabolomics$metabolites)[-1]
biochem <- breast_metabolomics$biochem[,c("COMP_ID","BIOCHEMICAL")]

#### stratified models

     # categorize my strata variables (quartiles):
     df$agecat <- gtools::quantcut(df$AGE_INT,4)
     metsInt(dat=df,
                     biochem=biochem,
                     outcome="BMI",
                     compid=comp.id,
                     intvar="agecat",
                     covariates=covars,
                     normalize=T)

#### Interaction models
     metsInt(dat=df,
               biochem=biochem,
               outcome="BMI",
               compid=comp.id,
               intvar="AGE_INT",
               covariates=covars,
               normalize=T)