plinker

plinker is a R package to interface PLINK over PLINK BED files

Quick Install

install PLINK

The easiest way is to use the precompiled binaries available here: https://www.cog-genomics.org/plink/1.9/

install plinker

To install the package:

using devtools

devtools::install_github('quartzbio/plinker_pkg', subdir = 'plinker')

using the source

clone/download this repository, then type:

• make install: install the package in the default R library
• sudo make install: if you need root access

Philosophy

The objective of this package is to use the powerful and fast algorithms implemented in PLINK without caring about input and ouput file formats, and without relying on PLINK order of operations.

Features

• plinker_bed object:

  • instantaneous loading, .bim file only loaded once if needed
  • is a view of the actual dataset, than can be recursively subsetted
  • a print() method displays relevant information: cardinality, views, annotations...
  • an as.data.frame() method (for small datasets or subsets) in long format

• subsetting

  • The most powerful feature is to be able to only use subsets of the original dataset without creating new files.
  • you can subset samples or SNPs, by indices or by IDs.
  • you can subset recursively: a subset is an actual first-order object
  • the order of the subsetting is kept: you can reorder your dataset (and the subsequent PLINK algorithms outputs) while keeping the origibnal dataset

• alleles lexigraphic order: the problem with MAF based allele order is that this order may be inversed in a subset.

  • all plinker wrapper PLINK algorithms can use the lexicogaphic order

• genotypes:

  • seamless and random access to genotypes (numeric) thanks to BEDMatrix R package (N.B: do not need to load .bim)
  • access to genotype strings (using actual alleles)

• custom annotations: You can attach your custom SNP and/or sample annotations to the object, and define new custom IDs. If defined, these IDs will appear in the PLINK results.

• sample IDs:

  • FAM-like annotations are not always well suited: the sample IDs are split between family ID (FID) and internal ID (IID). Quite often, a dummy value is assigned to FID (e.g. "0"). Plinker enables to use a single sample ID, which is either FID_IID, or just IID depending on the "ignore_fid" parameter , which is automatically inferred by default
  • you may also use a custom ID (cf annotations)

• automatic management of missing value (NA): plinker will encode with a value not present in the dataset

• pure-R implementations: for most algorithms, a pure R implementation with same input/output is provided, that allows:

  • to understand what PLINK actually computes
  • to easily tweak it if you need something slighlty different
  • to validate both implementations

• management of covariates:

  • easy integration (checked merge)
  • automatic creation of dummy variables for categorical variables (e.g. for linear models)

• fully tested (coverage ~ 100%)

old gitlab stuff