RAD-seq is a method for subsampling a reduced represenation of a genome by selecting and sequencing genomic fragments near restriction enzyme recognition sites. The goal is to efficiently obtain orthologous genomic regions/loci across many samples to identify genetic variation, without the need to sequence entire genomes.

In 2010 Emerson et al. applied RAD-sequencing to identify variants in natural populations of a mosquito to reconstruct recent phylogeography. This was the first application of RAD-seq to study natural variation, as opposed to genotyping crosses. They also introduced the stacks software tool to assemble RAD data.

Variant methods differ in their cost, complexity, and efficiency.

Why not just sequence the entire genome at low-coverage?
Low coverage whole genome sequencing (WGS) has had a resurgence in popularity as the cost of sequencing has decreased, and new imputation methods have been developed for genotyping very low coverage data.

However, many organisms have very large genomes which can make the cost prohibitive. Also, the analysis of WGS data requires a good reference genome, and imputation of low coverage data additionally requires a high quality/depth reference panel, which is a large additional cost.

For biodiversity research on non-model organisms it is faster, easier, and usually sufficient, to use RAD-seq rather than develop many other additional genetic resources.

Why is missing data a problem in RAD-seq?
Part of the efficiency of RAD-seq is accomplished through multiplexing samples, and errors in this step can lead to variable coverage across samples. In addition, some restriction fragments will be present in only some samples and not others. As a consequence, assembled RAD data sets are often sparse, containing many loci that are present in only a subset of samples, and fewer loci present in every single sample.

This problem is not unique to RAD-seq, but has received significant attention. It does need to be accommodated in many downstream analyses. We will discuss methods for this tomorrow in our analysis workshop.

Can I detect selection, or perform GWAS using RAD-seq?
Yes, RAD-seq can be used for both of these goals (e.g., Nadeau et al. 2013). However, the goal of your study are important determinants of whether or not RAD-seq is more efficient than WGS. Detecting selection or trait associations is much better when mapping RAD or WGS to a reference genome, as opposed to anonymous de novo loci.

At what phylogenetic scale can I use RAD-seq?
RAD-seq data generated for distantly related samples are expected to share fewer orthologous fragments in common, leading to increased missing data. At what scale is RAD-seq no longer useful? It depends on many factors: rate of substitutions and genome size change between samples; how many fragments, i.e., which enzymes are used; the type of protocol (shearing vs digestion); and sequencing coverage.

However, RAD-seq has been successfully applied to phylogenetic questions spanning >30 Mya (oaks; Eaton et al. 2015), >60 Mya (Viburnum; Eaton et al. 2017).

The efficiency of RAD vs. [others] depends on the scale and density of your taxon sampling, and planned analyses.

RAD-seq can be an efficient method for generating both population-scale and phylogenetic-scale datasets simultaneously. This is especially useful for biodiversity research on clades.

For example, you may obtain tens of thousands of loci with shared data among close relatives, allowing for fine-scale inference of population structure, admixture, phylogeography, and selection.

Among more distant clades perhaps only a few thousand loci are shared on average among samples, but this still sufficient to infer phylogenetic relationships.