inst/doc/CIP_importance.R

## ----setup, include=FALSE-----------------------------------------------------
knitr::opts_chunk$set(echo = TRUE)

## ----results = 'hide',warning = FALSE, message = FALSE------------------------
# load libraries
library(LTFGRS)
library(dplyr)
library(stringr)
library(lubridate)
library(ggplot2)
library(rmarkdown)
# setting seed
set.seed(555)

## -----------------------------------------------------------------------------
h2 = 0.5 # heritability
K = 0.1  # population prevalence

## -----------------------------------------------------------------------------
family = tribble(
  ~id, ~momcol, ~dadcol,
  "pid", "mom", "dad",
  "sib", "mom", "dad",
  "mom", NA, NA,
  "dad", NA, NA)

## -----------------------------------------------------------------------------
# constructing covariance matrix for the family
tmp_graph = prepare_graph(family,
                          icol = "id",
                          mcol = "momcol",
                          fcol = "dadcol")
cov_mat_obj = get_covmat(fam_graph = tmp_graph,
                         h2 = 0.5,
                         add_ind = TRUE,
                         index_id = "pid")

SIGMA = cov_mat_obj[rev(colnames(cov_mat_obj)),rev(colnames(cov_mat_obj)) ]

## -----------------------------------------------------------------------------
# number of families to construct (iterations)
nsim = 2e3

# add an identifieer to each family member per simulation
# format is: <name><iteration>
trio = lapply(1:nsim, function(i) {
  tmp_res = lapply(seq_along(family), function(col) {
    paste(family[[col]], i, sep = "")
  })
  names(tmp_res) = colnames(family)
  tmp_res %>% as_tibble()
}) %>% bind_rows

# setting NAs to be NAs again
trio = trio %>% mutate(across(everything(), ~ ifelse(str_detect(.x, "NA"), NA, .x)))

print(trio)

## -----------------------------------------------------------------------------
# simulated liabilities for each individual
liabs = MASS::mvrnorm(n = nsim, mu = rep(0, nrow(SIGMA)), Sigma = SIGMA)

# generating age of onset and status info
liabs2 = as_tibble(liabs) %>%
  mutate(across(-pid_g,
                ~   round(purrr::map_dbl(.x = .x, .f = LTFGRS:::convert_liability_to_aoo, pop_prev = K)), .names = "{col}_aoo"),
         # adding age columns with random ages for controls and aoo for cases
         across(contains("aoo"),
                ~ ifelse(is.na(.x),
                         sample(1:100, size = sum(is.na(.x)), replace = TRUE),
                         .x), .names = "{col}_age"),
         # generating status
         across(6:9,
                ~ (!is.na(.x)) + 0, .names = "{col}_status"))
colnames(liabs2) = str_replace(colnames(liabs2), "aoo_", "")

## ----echo=FALSE---------------------------------------------------------------
paged_table(liabs2, options = list(max.print = 100))

## -----------------------------------------------------------------------------
# if we assume the population is fully observed, i.e. no censoring. Then we can calculate the cIP as:
CIP = lapply(c("pid", "sib", "mom", "dad"), function(x) {
  ph = liabs2 %>% select(contains(x)) %>% 
    # we do not need the liability
    select(contains("aoo"), contains("age"), contains("status"))
  colnames(ph) = str_replace_all(colnames(ph), paste0(x, "_"), "")
  ph
}) %>% 
  bind_rows() %>% 
  group_by(age) %>% 
  summarise(n_cases = sum(status),
            n_total = n()) %>% 
  ungroup() %>% 
  arrange(age) %>% 
  mutate(cip = cumsum(n_cases) / sum(n_total))

## -----------------------------------------------------------------------------
CIP_ds = CIP %>% 
  mutate(cip = cip * 5)  

## ----out.width="90%", fig.alt = "Plot of the two different CIPs as a function of age."----
plot_data_CIPs = 
  bind_rows(
    CIP %>% mutate(type = "Population Representative"),
    CIP_ds %>% mutate(type = "Over-Ascertained")
  )

# how does the CIP look like in both scenarios?
plot_data_CIPs %>% ggplot(aes(x = age, y = cip)) +
  geom_point() + 
  facet_grid(~type) +
  theme_classic()

## ----results = 'hide'---------------------------------------------------------
thresholds = lapply(c("pid", "sib", "mom", "dad"), function(x) {
  ph = liabs2 %>% 
    mutate(fid = 1:n()) %>% 
    select(fid, contains(x)) %>% 
    # we do not need the liability
    select(fid, contains("aoo"), contains("age"), contains("status")) 
  colnames(ph) = str_replace_all(colnames(ph), paste0(x, "_"), "")
  left_join(ph %>% mutate(age = pmax(pmin(age, 99), 1)),
            select(CIP, age, cip),
            by = join_by(age)) %>% 
    mutate(thr = qnorm(cip, lower.tail = F), 
           lower = ifelse(status == 1, thr, -Inf),
           upper = ifelse(status == 1, Inf, thr),
           pid = paste0(x, fid))
}) %>% 
  bind_rows()

## -----------------------------------------------------------------------------
# using duplicated families to create a population graph
pop_graph = prepare_graph(.tbl = trio,
                          icol = "id",
                          mcol = "momcol",
                          fcol = "dadcol",
                          node_attributes = thresholds %>% select(pid,lower, upper) %>% rename(id = pid))

# autmatically detecting family members:
fam_graphs = get_family_graphs(pop_graph = pop_graph,
                               ndegree = 1,
                               proband_vec = str_subset(trio$id, "pid"))

## -----------------------------------------------------------------------------

pop_rep_fgrs = estimate_liability(
  family_graphs = fam_graphs,
  h2 = h2,
  family_graphs_col = "fam_graph")

## ----results = 'hide'---------------------------------------------------------
# thresholds from over-ascertained CIPs
thresholds_ds = lapply(c("pid", "sib", "mom", "dad"), function(x) {
  ph = liabs2 %>% 
    mutate(fid = 1:n()) %>% 
    select(fid, contains(x)) %>% 
    # we do not need the liability
    select(fid, contains("aoo"), contains("age"), contains("status")) 
  colnames(ph) = str_replace_all(colnames(ph), paste0(x, "_"), "")
  left_join(ph %>% mutate(age = pmax(pmin(age, 99), 1)),
            select(CIP_ds, age, cip),
            by = join_by(age)) %>% 
    mutate(thr = qnorm(cip, lower.tail = F), 
           lower = ifelse(status == 1, thr, -Inf),
           upper = ifelse(status == 1, Inf, thr),
           pid = paste0(x, fid))
}) %>% 
  bind_rows

# using duplicated families to create a population graph
pop_graph_ds = prepare_graph(.tbl = trio,
                             icol = "id",
                             mcol = "momcol",
                             fcol = "dadcol",
                             node_attributes = thresholds_ds %>% select(pid,lower, upper) %>% rename(id = pid))

# autmatically detecting family members:
fam_graphs_ds = get_family_graphs(pop_graph = pop_graph_ds,
                               ndegree = 1,
                               proband_vec = str_subset(trio$id, "pid"), 
                               fam_graph_col = "fam_graph_ds")

# estimating the gentic liabilities:
fgrs_ds = estimate_liability(
  family_graphs = fam_graphs_ds,
  h2 = h2,
  family_graphs_col = "fam_graph_ds")

## ----out.width="90%", fig.alt = "Plot of the true genetic liability against the estimated FGRS using both CIPs."----
plot_data = bind_rows(
  tibble(genetic_liab = liabs2$pid_g, 
         FGRS = pop_rep_fgrs$est,
         type = "Population Representative CIP"),
  tibble(genetic_liab = liabs2$pid_g, 
         FGRS = fgrs_ds$est,
         type = "Over-Ascertained CIP"))

cor_data = group_by(plot_data, type) %>% summarise(r = cor(genetic_liab, FGRS))

plot_data %>% 
  ggplot(aes(x = FGRS, y = genetic_liab)) +
  geom_point() +
  geom_abline(slope = 1, intercept = 0, color = "red") +
  facet_grid(~type) +
  geom_text(data = cor_data,
            aes(x = 1, y = -2, label = paste0("cor(g,FGRS) =", round(r, 3))), size = 5) +
  ylab("True Genetic Liability") + 
  theme_classic()

Try the LTFGRS package in your browser

Any scripts or data that you put into this service are public.

LTFGRS documentation built on Nov. 20, 2026, 5:10 p.m.