This R package provides functions to design and evaluate seqsuencing panels. A sequencing panel is a collection of genomic regions selected for targeted sequencing, such as cancer hotspot mutations. Given a cohort of individuals with associated genomic features (e.g., patients with somatic mutations), the typical objective is to design a panel that captures the largest number of individuals while using the smallest possible subset of features.
This package depended on the Biocondoctor packages GenomicRanges and GenomeInfoDb. Install it by
install.packages("BiocManager") # if needed
BiocManager::install(c("GenomicRanges", "GenomeInfoDb"))This R package is not yet on CRAN or Biodonducotr. Therefore, you have to install it form this GitHub repository.
# install.packages("remotes")
remotes::install_github("tron-bioinformatics/paneldesign")This is a basic example to show you how to use the package:
library(GenomicRanges)
library(paneldesign)Given a data set of mutations in patients, the goal is to select
mutations that are present in a maximum number of patients. We use a toy
data set consisting of patients p1, p2, and p3 of which each has 3
mutations.
mut_toy
#> # A tibble: 9 × 6
#> patient_id mut_id chr start end gene
#> <chr> <chr> <chr> <dbl> <dbl> <chr>
#> 1 p1 m01 1 1000 1000 g1
#> 2 p1 m02 1 2000 2000 g1
#> 3 p1 m03 2 3000 3000 g2
#> 4 p2 m01 1 1000 1000 g1
#> 5 p2 m04 2 4000 4000 g2
#> 6 p2 m05 3 5000 5000 g3
#> 7 p3 m02 1 2000 2000 g1
#> 8 p3 m04 2 4000 4000 g2
#> 9 p3 m06 2 6000 6000 g2Here, we use a greedy algorithm to select a minimal set of mutations that covers all patients.
select_greedy(mut_toy)
#> 100% covered by 2 sets.
#> # A tibble: 2 × 8
#> mut_id n n_samples rank order n_cum coverage coverage_cum
#> <chr> <int> <int> <dbl> <int> <int> <dbl> <dbl>
#> 1 m01 2 3 1 1 2 66.7 66.7
#> 2 m02 1 3 2 2 3 33.3 100This results in two mutations m01 and m02.
Here we use the following toy genomic region set.
gr_toy
#> GRanges object with 2 ranges and 0 metadata columns:
#> seqnames ranges strand
#> <Rle> <IRanges> <Rle>
#> r1 1 1000 *
#> r3 2 3000 *
#> -------
#> seqinfo: 2 sequences from an unspecified genome; no seqlengthsFirst, we associate each region to patients by overlap with the mutation data
reg_to_patient <- panel_to_patient(gr_toy, mut_toy)
reg_to_patient
#> # A tibble: 6 × 10
#> reg_id reg_chr reg_start reg_end mut_id patient_id chr start end gene
#> <chr> <chr> <int> <int> <chr> <chr> <chr> <dbl> <dbl> <chr>
#> 1 r1 1 1000 1000 m01 p1 1 1000 1000 g1
#> 2 r1 1 1000 1000 m01 p2 1 1000 1000 g1
#> 3 r1 1 1000 1000 m01 p1 1 1000 1000 g1
#> 4 r1 1 1000 1000 m01 p2 1 1000 1000 g1
#> 5 r3 2 3000 3000 m03 p1 2 3000 3000 g2
#> 6 <NA> <NA> NA NA <NA> p3 <NA> NA NA <NA>This results in the above-shown data set. The patient p3 is not
covered because there are no mutations in p3 that overlap with the
input regions. This results in ‘NA’ values for some region or
mutation-specific columns.
Next, we evaluate each region from the panel for the patient coverage.
eval_regions(reg_to_patient)
#> # A tibble: 2 × 15
#> reg_id reg_chr reg_start reg_end n_region reg_size reg_size_cum n_patient
#> <chr> <chr> <int> <int> <int> <dbl> <dbl> <int>
#> 1 r1 1 1000 1000 1 1 1 2
#> 2 r3 2 3000 3000 2 1 2 1
#> # ℹ 7 more variables: n_patients_cum <dbl>, total_patients <int>,
#> # percent_patients <dbl>, percent_patients_cum <dbl>, n_patient_gain <dbl>,
#> # n_mut <int>, n_mut_cum <dbl>This assumes some order of the input regions by decreasing priority and
in this way outputs additionally cumulative values such as
percent_patients_cum which the cumulative percent of input patients
that are covered by the current or any previous region. This is useful
to decide where to cut the panel.
Lastly, we can get some overall performance metrics of the entire panel:
eval <- eval_panel(reg_to_patient)
eval
#> # A tibble: 1 × 9
#> n_reg size_total total_patients n_patient percent_patient n_mut
#> <int> <dbl> <int> <int> <dbl> <int>
#> 1 2 2 3 2 66.7 2
#> # ℹ 3 more variables: mut_per_patient_median <int>, mut_per_patient_mean <dbl>,
#> # mut_per_patient_df <list>This data set contains a nested data set with information on how many mutations each patient has represented on the panel.
eval$mut_per_patient_df[[1]]
#> # A tibble: 3 × 2
#> patient_id mut_per_patient_n
#> <chr> <int>
#> 1 p1 2
#> 2 p2 1
#> 3 p3 0Structural variants (SVs) are represented as pairs of breakpoint
coordinates. And to be covered by a panel region either a single
(sv_mode = "single") or both (sv_mode = "both") breakpoints must be
included in a region of the panel.
An toy example SV dataset might look like this
sv_toy
#> # A tibble: 2 × 6
#> patient_id mut_id bp1_chr bp1_pos bp2_chr bp2_pos
#> <chr> <chr> <chr> <dbl> <chr> <dbl>
#> 1 p4 sv01 1 1000 1 1500
#> 2 p5 sv02 1 2000 2 2000We can evaluate a dataset with mutations and SVs by passing it as
additional input to the panel_to_patient() function
reg_to_patient <- panel_to_patient(gr_toy, mut_toy, sv_toy, sv_mode = "single")Now the reg_to_patient table includes additonal colums for the SV
breakpoint coordinates and includes here an association of SV sv01
from patient p4 with region r1. In contrast patient p5is not
covered by the panel.
reg_to_patient
#> # A tibble: 9 × 14
#> reg_id reg_chr reg_start reg_end mut_id patient_id chr start end gene
#> <chr> <chr> <int> <int> <chr> <chr> <chr> <dbl> <dbl> <chr>
#> 1 r1 1 1000 1000 m01 p1 1 1000 1000 g1
#> 2 r1 1 1000 1000 m01 p2 1 1000 1000 g1
#> 3 r1 1 1000 1000 m01 p1 1 1000 1000 g1
#> 4 r1 1 1000 1000 m01 p2 1 1000 1000 g1
#> 5 r3 2 3000 3000 m03 p1 2 3000 3000 g2
#> 6 r1 1 1000 1000 sv01 p4 <NA> NA NA <NA>
#> 7 r3 2 3000 3000 <NA> <NA> <NA> NA NA <NA>
#> 8 <NA> <NA> NA NA <NA> p3 <NA> NA NA <NA>
#> 9 <NA> <NA> NA NA <NA> p5 <NA> NA NA <NA>
#> # ℹ 4 more variables: bp1_chr <chr>, bp1_pos <dbl>, bp2_chr <chr>,
#> # bp2_pos <dbl>