Single-trait Fine-mapping with FineBoost
In colocboost: Multi-Context Colocalization Analysis for QTL and GWAS Studies

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  dpi = 50
)

This vignette demonstrates how to perform single-trait fine-mapping analysis using FineBoost, a specialized single-trait version of ColocBoost, with both individual-level data and summary statistics. Specifically focusing on the 2nd trait with 2 causal variants (194 and 589) from the Ind_5traits and Sumstat_5traits datasets included in the package.

library(colocboost)

1. Fine-mapping with individual-level data

In this section, we demonstrate how to perform fine-mapping using individual-level genotype (X) and phenotype (Y) data. This approach uses raw data directly to identify causal variants.

# Load example data
data(Ind_5traits)
X <- Ind_5traits$X[[2]]
Y <- Ind_5traits$Y[[2]]

res <- colocboost(X = X, Y = Y)
colocboost_plot(res)

2. Fine-mapping with summary statistics

This section demonstrates fine-mapping analysis using summary statistics along with a proper LD matrix.

# Load example data
data(Sumstat_5traits) 
sumstat <- Sumstat_5traits$sumstat[[2]]
LD <- get_cormat(Ind_5traits$X[[2]])

res <- colocboost(sumstat = sumstat, LD = LD)
colocboost_plot(res)

3. LD-free fine-mapping with one causal variant assumption

In scenarios where LD information is unavailable, FineBoost can still perform fine-mapping under the assumption that there is a single causal variant. This approach is less computationally intensive but assumes that only one variant within a region is causal.

# Load example data
res <- colocboost(sumstat = sumstat)
colocboost_plot(res)

Note: Weak learners SEL in FineBoost may capture noise as putative signals, potentially introducing false positives to our findings. To identify and filter spurious signals, we discard fine-tunned the threshold of $\Delta L_l$ using extensive simulations to balance sensitivity and specificity. This threshold is set to 0.025 by default for ColocBoost when detect the colocalization, but we suggested a less conservative threshold of 0.015 for FineBoost when performing single-trait fine-mapping analysis (check_null_max = 0.015 as we suggested).