-
Notifications
You must be signed in to change notification settings - Fork 2
R Classes for Population Genetic Data
Christine Ewers-Saucedo edited this page Mar 9, 2015
·
12 revisions
The goal of this page is to discuss ways to code population genetic data. Below is a table of the currently available classes to store population genetic data and their strengths and weaknesses.
| Class (package) | Strengths | Weaknesses |
|---|---|---|
| DNAbin (ape) | stores all sets of sequences (aligned or not) | less compact than 2-bit coding (but by a factor 4 at most) |
uses matrices (aligned) or lists so usual R's commands (names, rownames, [, [[, $) can be used |
||
many as.DNAbin methods in ape (inc. from BioConductor) |
||
efficient functions in ape (dist.dna, seg.sites, base.freq, read.FASTA) and in pegas (haplotype) |
||
| loci (pegas) | low memory usage | not really appropriate for some analyses (e.g., multivariate analyses) |
| all levels of ploidy and any number of alleles | needs to improve the treatment of NA's (especially when data are read with read.vcf() | |
| genotypes can be phased | ||
| any kind of individual data can be associated in the data frame | ||
| efficient to compute genotype and allele frequencies | ||
| genind (adegenet) | stores allelic frequencies; ideal for multivariate analyses | requires more memory |
| additional slots for individual data | less efficient to compute frequencies | |
| all levels of ploidy | ||
| genpop (adegenet) | equivalent to genind at group level; ideal for multivariate analysis | requires more memory |
| genlight (adegenet) | stores binary SNPs using bit-level coding; very memory efficient | more computationally intensive to handle; less functionalities |
| all levels of ploidy | assumes bi-allelic loci | |
| genclone (poppr) | stores population metadata (@hierarchy slot); users don't have keep track of multiple objects for the same data |
inherits genind object; all the same weaknesses plus slightly more memory |
stores multilocus genotype definitions (@mlg slot) for clonal populations |
||
| genambig (polysat) | stores microsatellite data with ambiguous ploidy | does not handle any other data type |
| exports to genpop objects | cannot easily be transferred to any other object | |
| phyDat (phangorn) | very general inspired by R data.frame, factor and contrasts, can contain any discrete data type; nucleotides, amino acids and codons have some more support |
designed having phylogenetic analysis in mind; requires alignments, where all sequences have same length |
can be converted to and from DNAbin objects (as.DNAbin / as.phyDat) |
||
a few generic functions work on it: c, unique, subset and utility functions baseFreq, allSitePattern, etc. |
data are not necessarily very memory efficient (as integer + contrast matrix), but stores only unique site patterns and their weights (as double) | |
| "efficient" maximum likelihood, maximum parsimony and distance functions in phangorn | ||
| gtype (strataG) | a simple R list containing a matrix where the first column is a stratification scheme and columns afterward are either haplotypes or diploid loci. If haploid data, the gtype object can also contain a list of DNA sequences. |
Can likely be made more efficient in terms of storage and preprocessing for other analytical routines in package |
can be converted to data.frame or matrix with appropriate as. functions. |
||
has manipulation functions like subset which will select certain strata and/or loci, merge to combine mulitple gtypes, and summary. |
||
| can create input files for Genepop, STRUCTURE, fastsimcoal, Arlequin, MEGA, and PHASE | ||
| vareff (VarEff) | this format is the only input format for the package VarEff, which is the only R package I could find to estimate the effective population sizes from present to ancestral time using microsatellite markers under the coalescence - AKA skyline plots | it is very difficult to convert any genotype data format into the vareff format |
| other |
Possible lines of action:
- improve the efficiency of
loci2genindandgenind2lociin pegas - explain, in the targeted vignette, why it's useful to have different classes
- consider methods data transfer between of several objects akin to
loci2genind - improve
pegas::read.vcf - recode some parts to be more efficient
- give the possibility to select some loci without reading all the file (support for tabix?)
- give the possibility to read the extra informations (e.g, genotype likelihoods)
- support for BCF files?