SRM data

library(data.table)
library(magrittr)
library(ggplot2)
library(SummarizedExperiment)
library(readxl)
theme_set(theme_minimal())

We will start by reading the data.

gmap <- c(control = "no weight loss", `weight loss` = "weight loss")

srm <- read_excel("data/innovator_SRM_02122020.xlsx", sheet = "data") %>% setDT()
prots <- read_excel("data/innovator_SRM_02122020.xlsx", sheet = "proteins") %>% setDT()

srm <- melt(srm, id.vars = c("replicate_id", "public_client_id", "group"), 
            variable.name = "id", value.name = "abundance")
srm <- prots[srm, on = "id"]
srm[, c("group", "state") := tstrsplit(group, "_")]
srm[, group := gmap[group]]
srm

Let;s join that with our manifest.

manifest <- fread("data/manifest.csv")
classes <- c(microbiome_id = "character")
cohort <- list(
    no_loss = fread("no_weight_loss.csv", colClasses = classes)[, "group" := "no weight loss"],
    loss = fread("successful_weight_loss.csv", colClasses = classes)[, "group" := "weight loss"]
) %>% rbindlist()
cohort[, "state" := c("before", "after")[order(days_in_program)], by = "public_client_id"]
cohort

And let’s combine the two. We will only keep the first transition for each protein to have the same n for each probe.

merged <- cohort[srm, on = c("public_client_id", "group", "state")]

norm <- copy(merged)
norm[, "log_abundance" := log2(abundance)]
norm[, "log_abundance_norm" := log_abundance - mean(log_abundance), by = c("public_client_id", "state")]
ggplot(norm, aes(x=factor(paste(public_client_id, state)), 
                 y=log_abundance, color=group)) + 
    geom_jitter(width=0.2)

Now we can start to model the associations:

model_stats <- function(dt, delta = FALSE) {
    mod <- glm(log_abundance ~ group + bmi + age, data=dt)
    if (delta) {
        mod <- glm(delta ~ group + bmi + age, data=dt)
    }
    coef <- coefficients(mod)[2]
    pval <- summary(mod)$coefficients[2, 4]
    pval_bmi <- summary(mod)$coefficients[3, 4]
    return(data.table(
        id = dt$id[1],
        # gene = dt[1, `gene name`],
        coef_name = names(coefficients(mod))[2],
        coef = coef,
        pval = pval,
        coef_bmi = coefficients(mod)[3],
        pval_bmi = pval_bmi
    ))
}

stats <- norm[state == "after", model_stats(.SD), by = "gene name"]
stats[, padj := p.adjust(pval, method = "fdr")]
stats[, padj_bmi := p.adjust(pval_bmi, method = "fdr")]
stats[order(padj)]

ggplot(norm[state == "after"], aes(x=group, y=log_abundance)) + 
    geom_jitter(width=0.2, alpha = 0.5) + 
    facet_wrap(~ `gene name`, scale="free_y") +
    labs(x = "", y="Δprotein [log2(abundance)/month]") +
    theme(axis.text.x = element_text(angle = 30, vjust = 1, hjust=1))

And we can look at the deltas:

norm[, delta := (log_abundance[state == "after"] - log_abundance[state == "before"]) / span_days * 30.5, by = c("public_client_id", "id")]

stats <- norm[state == "after", model_stats(.SD, delta=T), by = "gene name"]
stats[, padj := p.adjust(pval, method = "fdr")]
stats[, padj_bmi := p.adjust(pval_bmi, method = "fdr")]
stats[order(pval)]

sig <- stats[padj < 0.1, `gene name`]

npep <- merged[, uniqueN(`peptide seq`), by="gene name"]
labels <- npep[, paste0(`gene name`, " [", V1, "]")]
names(labels) <- npep[, `gene name`]

ggplot(norm[state == "before" & `gene name` %in% sig], aes(x=group, y=delta)) + 
    geom_hline(yintercept = 0, lty = "dashed", color = "gray20") +
    geom_jitter(width=0.2) + 
    facet_wrap(~ `gene name`, scale="free_y", nrow=1, labeller=as_labeller(labels)) +
    stat_summary(fun.y = median, geom = "point", size = 2, stroke=1, color = "blue", fill="white", shape = 23) +
    labs(x = "", y="Δprotein [log2 fold-change/month]") +
    theme(axis.text.x = element_text(angle = 30, vjust = 1, hjust=1))

`fun.y` is deprecated. Use `fun` instead.

ggsave("figures/deltas.png", dpi=300, width=12, height=3.5)

timed <- copy(norm)[, state := factor(state, levels=c("before", "after"))]
timed[state == "before", time := 0]
timed[state == "after", time := span_days]
means <- timed[`gene name` == "ADIPOQ", .(log_abundance = mean(log_abundance)), by = c("state", "group", "id", "gene name")]
ggplot(timed[`gene name` == "ADIPOQ"], 
       aes(x=state, y=log_abundance, color = group)) + 
    geom_point(alpha=0.25, size=0.5) + geom_line(aes(group = interaction(public_client_id, id)), alpha=0.25) +
    geom_point(data=means) + geom_line(aes(group = id), data = means, size=1) +
    guides(color = FALSE) + labs(x="", y="abundance [log-scale]") +
    facet_grid(group ~ `gene name`)

ggsave("figures/slopes.png", dpi=300, width=3, height=3.5)