roi_met.R

library(ggplot)
library(gamm4)
library(dplyr)
library(checkmate) # for expect_subset dataframe names test

met_res_for_roi <- function(MRSI_input, this_roi, met_name, minage=18, maxage=25) {
  # have columns we need
  expect_subset(c("roi", "age", "dateNumeric", met_name), names(MRSI_input))
  
  roi_met <- MRSI_input %>% 
     filter(roi %in% this_roi, age >=minage, age <=maxage) %>%
     mutate(met = .data[[met_name]],  # map whatever metabolite we're using (met_name) onto the 'met' column
            gamresids  = gam(met ~ s(age, k=3), data=.) %>% residuals,
            dateAdjust = gam(gamresids ~ s(dateNumeric, k=3), data=.) %>% predict,
            met_adj = met - dateAdjust)

  return(roi_met)
}

plot_met_adjusted <- function(roi_met) {
  # have columns we need
  expect_subset(c("age", "gamresids", "dateAdjust", "met", "met_adj", "visitnum", "subjID"),
                names(roi_met))
  qassert(roi_met$met_adj, "n+") # nonzero numeric vector in met_adj
  
  cowplot::plot_grid(  
        ggplot(roi_met) + aes(x=age, y=gamresids) + geom_smooth(method="loess") + geom_point(),
        ggplot(roi_met) + aes(x=vdate, y=dateAdjust) + geom_point() + geom_smooth(method="loess"),
        ggplot(roi_met) + aes(x=vdate, y=met_adj, color=visitnum, group=visitnum) + geom_point() + geom_smooth(method="loess"),
        ggplot(roi_met) + aes(x=age, y=met_adj) + geom_point() + stat_smooth(method='loess', se=F) + geom_path(aes(group = subjID)),
        ggplot(roi_met) + aes(x=age, y=met) + geom_point() + stat_smooth(method='loess', se=F) + geom_path(aes(group = subjID))
  )
}

test_data <- data.frame(subjID=1:32, roi=1, age=rep(10:25,2), gaba=randu[1:32,1], dateNumeric=sample(rep(1:8,4)),visitnum=1) %>% mutate(vdate=dateNumeric)
adj_df <- met_res_for_roi(test_data, 1, "gaba")
plot_met_adjusted(adj_df)