Fine-mapping analysis pipeline by finemapr

Andrey Ziyatdinov

2018-02-19

About finemapr

finemapr is an R package that provides an interface to fine-mapping tools:

By using finemapr, your input files are automatically prepared for each tool, the analysis workflow is tool-independent; and exploration of fine-mapping results is powered by R in printing/plotting/data export.

Tool-independent scheme of analysis workflow

# set up
options(finemapr_<tool> = "<path to fine-mapping tool>")

# read input files
my_zscores <- read_zscores("<my_scores.tab>")
my_ld <- read_ld("<my_ld.tab>")

# run analysis
out <- run_<tool>(my_zscores, my_ld, args = "<custom arguments>")

# explore results
print(out)
head(out$snp) # main table of results
plot(out)

# export results
write.table(out$snp, "<my_results.tab>")

Installation

The user needs to download and install a fine-mapping tool before the analysis. An example of installation commands used in finemapr by default is given here.

After installing, for example, the FINEMAP tool, the user specify for finemapr where the tool is located:

options(finemapr_finemap = "~/apps/finemap/finemap")

Load packages

We load packages for the analysis conducted in this document.

library(devtools)
load_all("~/git/variani/finemapr")
#> Loading finemapr

library(magrittr)
library(dplyr)  
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(ggplot2)  
theme_set(theme_linedraw())

Example data

We load example data copied from the FINEMAP website (http://www.christianbenner.com/). This simulated dataset has two causal variants rs15 and rs47.

file1_z <- system.file("extdata/region1.z", package = "finemapr")
file1_ld <- system.file("extdata/region1.ld", package = "finemapr")

z1 <- read_zscore(file1_z)
ld1 <- read_ld(file1_ld, snps = z1$snp)
n1 <- 5363

Explore z-scores

Top 5 z-scores:

z1 %>% arrange(-abs(zscore)) %>% head(5) %>% kable(digits = 1)
snp zscore
rs15 10.9
rs47 8.8
rs23 5.8
rs17 -5.6
rs42 -5.0 
ggplot(z1, aes(zscore)) + geom_histogram()
#> `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

mutate(z1, pval = pchisq(zscore^2, df = 1, lower.tail = FALSE)) %>%
  ggplot(aes(pval)) + geom_histogram()
#> `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Run tools

Run FINEMAP

options(finemapr_finemap = "~/apps/finemap/finemap")

out_finemap <- run_finemap(z1, ld1, n1, args = "--n-causal-max 3")
#> Parsed with column specification:
#> cols(
#>   index = col_integer(),
#>   snp = col_character(),
#>   snp_prob = col_double(),
#>   snp_log10bf = col_double()
#> )
#> Parsed with column specification:
#> cols(
#>   rank = col_integer(),
#>   config = col_character(),
#>   config_prob = col_double(),
#>   config_log10bf = col_double()
#> )
print(out_finemap)
#>  - command: ~/apps/finemap/finemap --sss --log --n-causal-max 3 --in-files region.master 
#>  - see log output in `log`
#>  - tables of results: `config`, `snp`, `ncausal`
#>  - config:
#> # A tibble: 9,007 x 4
#>    rank config         config_prob config_log10bf
#>   <int> <chr>                <dbl>          <dbl>
#> 1     1 rs15,rs47           0.607            42.7
#> 2     2 rs15,rs42,rs47      0.0326           43.1
#> 3     3 rs15,rs34,rs47      0.0222           42.9
#> # ... with 9,004 more rows
plot(out_finemap, label_size = 3, grid_ncol = 1)

Run CAVIAR

options(finemapr_caviar = "~/apps/caviar/CAVIAR")

out_caviar <- run_caviar(z1, ld1, args = "-c 3")
#> Parsed with column specification:
#> cols(
#>   SNP_ID = col_character(),
#>   Prob_in_pCausalSet = col_double(),
#>   Causal_Post._Prob. = col_double()
#> )
print(out_caviar)
#>  - command: ~/apps/caviar/CAVIAR -c 3 -z region.z -l region.ld -o log 
#>  - tables of results: `snp`
#>  - snp:
#> # A tibble: 50 x 4
#>    rank snp   snp_prob_set snp_prob
#>   <int> <chr>        <dbl>    <dbl>
#> 1     1 rs15        0.439    1.00  
#> 2     2 rs47        0.439    1.00  
#> 3     3 rs42        0.0120   0.0274
#> # ... with 47 more rows
#>  - 95%-causal set (ordered): rs15, rs47, rs42, rs34, rs20, rs5, rs27, rs45, rs38, rs17, rs25, rs18, rs11, rs19, rs44, rs40, rs8, rs24
plot(out_caviar, label_size = 3)

Run PAINTOR

options(finemapr_paintor = "~/apps/paintor/PAINTOR")

out_paintor <- run_paintor(z1, ld1, n1, args = "-enumerate 3")
#> Parsed with column specification:
#> cols(
#>   snp = col_character(),
#>   zscore = col_double(),
#>   Posterior_Prob = col_double()
#> )
print(out_paintor)
#>  - command: ~/apps/paintor/PAINTOR -input region.master -Zhead zscore -LDname ld -annotations dummy_ones -enumerate 3 -num_samples 5363 
#>  - tables of results: `snp`
#>  - snp:
#> # A tibble: 50 x 4
#>    rank snp    zscore snp_prob
#>   <int> <chr>   <dbl>    <dbl>
#> 1     1 rs15   10.9     1.00  
#> 2     2 rs47    8.80    1.00  
#> 3     3 rs25  - 0.133   0.0200
#> # ... with 47 more rows
#>  - annotations: dummy_ones 
#>  - logBF (proportional to the model likelihood): 100.7274
plot(out_paintor, label_size = 3)

Conclusions

All three fine-mapping tools estimated the poterior causal probabilities of the two variants, rs15 and rs47, very close to 1.