TL;DR
This paper introduces RADpainter and fineRADstructure, a new software package that makes powerful population structure inference methods accessible to researchers working with RADseq data, a common and cost-effective genomic tool for nonmodel organisms. It overcomes previous limitations by effectively handling the unique characteristics of RADseq data, including missing data and the absence of a reference genome.
What Problem Does This Paper Address?
Inferring recent population structure and recent shared ancestry is crucial for understanding evolutionary processes. Methods like chromoPainter/fineSTRUCTURE excel at this by building a "coancestry matrix" based on nearest neighbor haplotype relationships. However, these methods traditionally require high-quality, genome-wide haplotype data, which is often unavailable for nonmodel organisms. The rise of RADseq (Restriction-site-associated DNA sequencing) has made genomic data accessible for these organisms, but its inherent characteristics, such as missing data and the lack of a reference genome, have made it incompatible with existing coancestry inference tools. This paper addresses this gap by developing a method specifically designed for RADseq data.
Methods
The core innovation is RADpainter, a program designed to infer the coancestry matrix from RADseq data. Unlike previous methods that relied on SNP data, RADpainter utilizes the sequence information from all variable sites (SNPs) within each RAD locus to define alleles. It then identifies the nearest neighbor haplotype for each allele across all individuals. This information is summed across all RAD loci to construct the coancestry matrix.
This RADpainter program is then combined with the previously published fineSTRUCTURE Markov chain Monte Carlo (MCMC) clustering algorithm into a complete package called fineRADstructure. This package provides a user-friendly and fast solution for population inference. Key methodological aspects include:
- Handling arbitrary ploidy: The method can accommodate varying ploidy levels, even within the same analysis, making it suitable for diverse organisms, particularly plants.
- Leveraging linked SNPs: By considering all SNPs within a RAD locus, RADpainter gains higher resolution in identifying nearest neighbor relationships compared to methods using single SNPs.
- No reference genome required: This is a significant advantage for nonmodel organisms where reference genomes are often unavailable.
- Robustness to missing data: The paper details strategies for handling missing nucleotides (Ns) and allele dropout, a common issue in RADseq data.
Datasets
The paper uses two example datasets to illustrate the capabilities of fineRADstructure:
- A single-digestion RAD data set: This dataset includes 120 individuals from 12 populations of the alpine plant species complex Heliosperma pusillum. This dataset was assembled using the Stacks pipeline without a reference genome.
- A double digest RAD (ddRAD) data set: This dataset comprises 76 samples from two different subspecies and three hybrid individuals. This dataset was used to specifically assess the robustness of fineRADstructure to nonrandom missingness and allele dropout.
The paper does not introduce a new benchmark dataset but rather uses existing datasets to demonstrate the performance of their new software.
Key Insights and Ideas
- Democratizing advanced population genetics: RADpainter and fineRADstructure make sophisticated population structure inference, previously limited to high-quality genome-wide data, accessible to a much wider range of researchers, especially those working with nonmodel organisms and RADseq data.
- Leveraging RADseq data's strengths: The method effectively utilizes the information within each RAD locus (linked SNPs) and is designed to be robust to the common challenges of RADseq data, such as missingness.
- Uncovering complex relationships: The approach can reveal intricate population structures, including evidence of gene flow and parallel ecological divergence, which might be missed by other methods.
- Visualizing complex data: The package includes R scripts for plotting results, such as clustered coancestry matrices and trees, which are crucial for interpreting complex population structures.
Key Results
- Successful inference of population structure: fineRADstructure successfully identified distinct populations and substructure in the Heliosperma pusillum dataset, revealing relationships that were not clearly resolved by other methods like fastSTRUCTURE or PCA.
- Robustness to missing data: The analysis of the ddRAD dataset demonstrated that fineRADstructure's inference of genetic structure was not dominated by nonrandom missingness due to batch effects or allele dropout. Hybrid individuals were also clearly identified.
- No reference genome needed: The method's ability to work without a reference genome is a significant practical advantage for many research applications.
- Improved resolution: By using all SNPs within a RAD locus, RADpainter provides higher resolution in identifying nearest neighbor relationships compared to methods that use only single SNPs.
Limitations and Future Work
- Interpretation of cluster number: The authors caution against interpreting the number of clusters identified by fineRADstructure as a definitive estimate of the "correct" number of populations. They suggest users may gain better intuition by running the clustering with varying sensitivity (c parameter values) and interpreting the results jointly.
- Ad hoc tree building: While the ad hoc tree building approach is useful for illustrating relationships, it is intended to be illustrative rather than a definitive representation of true population history. The authors suggest that when true historical relationships are known, the coancestry matrix can be ordered accordingly to provide insights into recent gene flow.
- Assumption of independence between RAD loci: While the method accounts for linkage within loci, it assumes independence between different RAD loci. The authors note that this assumption is not always realistic, especially when loci are not sorted by genomic position. They provide a script to reorder loci based on LD to address this.
- Future work: The paper highlights the utility of the software for plant research communities due to its ability to handle arbitrary ploidy. The authors also suggest that the coancestry matrix can provide valuable information on recent sample relatedness, complementing older historical relationships, especially in cases of recent gene flow.
Summary updated