RADpainter and fineRADstructure: Population Inference from RADseq Data.
Malinsky M, Trucchi E, Lawson DJ, Falush D. · 2018
Read RADpainter and fineRADstructure: Population Inference from RADseq Data. by Malinsky M, Trucchi E, Lawson DJ, Falush D. (2018) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5913677.txt.
Introduction
Malinsky and co-authors present tools to infer population structure from RADseq data. Their article appeared in Molecular Biology and Evolution. RADseq is cheap and needs no reference genome. It fuels work on wild species with little prior data. Strong ancestry methods, though, assumed full haplotype data. Non-model researchers could not easily use them.
Their program RADpainter builds a coancestry matrix from RADseq reads. It pairs with an existing clustering method as fineRADstructure. The package is fast and simple to run, they report. Two example data sets show its use and robustness. Missing alleles in double-digest data did not break it. Ecologists get fine ancestry tools at last.
Catalog record
| CREATOR | Malinsky M, Trucchi E, Lawson DJ, Falush D. |
|---|---|
| YEAR | 2018 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
RADpainter and fineRADstructure: Population Inference from RADseq Data By Milan Malinsky, Emiliano Trucchi, Daniel John Lawson, Daniel Falush, Naoko Takezaki Published in: Molecular Biology and Evolution DOI: 10.1093/molbev/msy023 ABSTRACT Abstract Powerful approaches to inferring recent or current population structure based on nearest neighbor haplotype “coancestry” have so far been inaccessible to users without high quality genome-wide haplotype data. With a boom in nonmodel organism genomics, there is a pressing need to bring these methods to communities without access to such data. Here, we present RADpainter, a new program designed to infer the coancestry matrix from restriction-site-associated DNA sequencing (RADseq) data. We combine this program together with a previously published MCMC clustering algorithm into fineRADstructure—a complete, easy to use, and fast population inference package for RADseq data (https://github.com/millanek/fineRADstructure; last accessed February 24, 2018). Finally, with two example data sets, we illustrate its use, benefits, and robustness to missing RAD alleles in double digest RAD sequencing. Keywords: population structure, RAD-seq, inference Understanding of shared ancestry in genetic data sets is often key to their interpretation. The chromoPainter/fineSTRUCTURE package (Lawson et al. 2012) represents a powerful model-based approach to investigating population structure using genetic data. It offers especially high resolution in inference of recent shared ancestry, as shown for example in the investigations of worldwide human population history (Hellenthal et al. 2014) and of genetic structure of the British population (Leslie et al. 2015). Further advantages, when compared with other model-based methods, such as STRUCTURE (Pritchard et al. 2000) and ADMIXTURE (Alexander et al. 2009), include the ability to deal with a very large number of populations, explore relationships between them, and to quantify ancestry sources in each population. The high resolution of chromoPainter/fineSTRUCTURE and related methods derives from utilizing haplotype linkage information and from focusing on the most recent coalescence (common ancestry) among the sampled individuals. This approach derives a “coancestry matrix,” a summary of nearest neighbor haplotype relationships in the data set; that is, of the cases where pairs of individuals had the most similar haplotypes one to another. However, the existing pipeline for coancestry matrix inference was designed for large scale human genetic single nucleotide polymorphism (SNP) data sets, where chromosomal locations of the markers are known, haplotypes are typically assumed to be correctly phased (although it is possible to perform the analysis without this assumption), and missing data also need to have been imputed. Therefore, these methods have so far been generally inaccessible for investigations beyond model organisms. With no requirements for prior genomic information (e.g., no need for a reference genome) and relatively low cost, RADseq data are fuelling a boom in ecological and evolutionary genomics, especially for nonmodel organisms (Andrews et al. 2016), where questions on population structure and relative ancestry are among the most commonly asked. Therefore, we have developed RADpainter, a simple method designed to infer the coancestry matrix from RADseq data. RADpainter is designed to take full advantage of the sequence of all the SNPs from each RAD locus, finding (one or more) closest relatives for each allele. The information about the nearest neighbors of each individual is then summed up into the coancestry similarity matrix. We package RADpainter together with the fineSTRUCTURE Markov chain Monte Carlo (MCMC) clustering algorithm into an easy to use population inference package for RADseq data called fineRADstructure. NEW APPROACHES Briefly, the coancestry matrix is calculated as follows: for each RAD locus and each individual (a recipient), we calculate the number of sequence differences (i.e., SNPs) between that individual’s allele(s) and the alleles in all other individuals (potential donors). The closest relative (donor) for each allele is its nearest neighbor allele, that is, the allele with the least number of differences. In the case of multiple equally distant nearest neighbors, an equal proportion of coancestry is assigned to each “donor.” Finally, we sum these local coancestry values across all loci to obtain the coancestry matrix for the full data set. A basic outline of the coancestry estimation procedure for haploid individuals is shown in supplementary Algorithm S1, Supplementary Material online. Differences in ploidy are handled by averaging coancestry across alleles in the same individual. Figure 1 provides a further illustration of the method, showing co-ancestry matrix calculation at a single locus for four diploid individuals. However, we have implemented RADpainter in a way to handle arbitrary number of alleles per individual (i.e., any ploidy levels), depending on the input; ploidy can even vary across individuals in a single analysis. We expect these features to be of particular use to the plant research community. [Figure: Coancestry matrix example with four diploid individuals and no missing data. (A) For each RAD locus (stack), we use all the variable sites in the stack to define the stack haplotypes H1 and H2. The calculation proceeds by finding the closest relatives for each haplotype. (B) With individual 1 as the recipient (filling the first row of the coancestry matrix), we find that the single closest relative of H1 in individual 1 is the haplotype H1 in individual 2. These haplotypes are identical (ACTG; red). Haplotype H2 in individual 1 has two closest relatives, both with an identical sequence (ATTT; green). (C) Focusing on individual 2 as the recipient (filling the second row of the coancestry matrix) we find that haplotype H1 in individual 2 has a single closest relative: H1 in individual 1, in a mirror image of (B). However, H2 in individual 2 does not have an identical closest relative. Its “donors” are the four haplotypes that differ from it by a single SNP. (D) The coancestry matrix corresponding to the nearest-neighbor haplotype relationships at this locus. Note that the matrix is not symmetrical: row totals are identical for each individual and correspond to the number of RAD loci analyzed. In contrast, column totals vary. Individuals with “outlier” sequences are the least likely donors, CCTT in individual 4 being a clear outlier haplotype.] Concatenating all the variable sites (SNPs) at each locus to define alleles increases the resolution of RADpainter when compared with methods that only use a single SNP from each locus. This can be seen clearly by considering the fact that a single biallelic SNP splits the alleles at a locus into only two groups, whereas multiple linked SNPs typically enable more refined assignment of nearest neighbor relationships, as illustrated in figure 1. Given the short length of each RAD locus (typically < 250 bp), we assume that all SNPs within the locus are in almost complete linkage disequilibrium (LD) (i.e., that D′ ≈1) and, therefore, historical recombination can be disregarded. On the other hand, the summation across loci assumes frequent recombination between different RAD sequences—each RAD locus is counted as if providing independent evidence with regards to coancestry. This assumption is not always realistic, in particular when the data set contains pairs of loci from the two sides of each restriction site. However, we account for any linkage between RAD loci by adjusting the normalizing constant c which is passed on to the fineSTRUCTURE clustering algorithm together with the co-ancestry matrix and determines its sensitivity (Lawson et al. 2012). Briefly, the fineSTRUCTURE clustering algorithm (Lawson et al. 2012) uses a MCMC scheme to explore the space of population configurations (sample