rCorrSieve

Michael G. Campana
Smithsonian's National Zoo & Conservation Biology Institute

R implementation of CorrSieve

Licensing

Original R source code (CorrSieve versions <= 1.6-8) copyright (c) Michael G. Campana, 2010-2013 is licensed under the GNU General Public License (version 3 or later). See included LICENSE file for details.

Public domain updates by Michael G. Campana (2019, 2022) to the GitHub documentation (CorrSieve versions >= 1.6-8) and package metadata (CorrSieve version 1.6-9) are United States government works.

Introduction

CorrSieve is a Ruby and R package that filters Q value output from the programs STRUCTURE (Pritchard et al. 2000) and INSTRUCT (Gao et al. 2007) by correlation values. It outputs matrices showing significant correlations between individual runs for each K. It can also calculate ΔK (Evanno et al. 2005), mean F_STs and ΔF_ST. These measures help identify meaningful values of K.

Installation and Compatibility

rCorrSieve is compatible with Windows, Linux, and UNIX (including macOS) operating systems. It requires the R statistical environment available from the Comprehensive R Archive Network (CRAN: http://cran.r-project.org). Install the appropriate environment for your operating system.

Load the R environment.

The latest stable rCorrSieve version is available from CRAN. Install this version using:
install.packages('corrsieve')

The development rCorrSieve version is hosted here on GitHub. Installation of this version requires the remotes package:
install.packages('remotes')
library(remotes)
install_github('campanam/rCorrSieve')

Usage

1. Prepare input for CorrSieve. CorrSieve reads directly from STRUCTURE and INSTRUCT output files, but requires that all files be in a single folder. Do not place other files in this folder. All files should end in ‘_f’ without an extension, e.g:
   TEST_11_f
   TEST_12_f
   TEST_13_f
   TEST_21_f
   TEST_22_f
   TEST_23_f

2. Load the package:
   library(corrsieve)

3. The command to read a folder of STRUCTURE output for summarising F_ST and ln P(D) and calculating ΔK and ΔF_ST is read.struct(filepath = “./”, instruct = FALSE) The file path default is the current directory. The default data file format is the STRUCTURE format. Setting instruct to TRUE indicates the presence of INSTRUCT data. This will create a table of data in the R workspace. For example, the command:
   data <- read.struct(“C:/Results/”)
   will create the table ‘data’ containing all the STRUCTURE output in the folder C:/Results/.

4. To summarise the F_ST and ln P(D) data from the data read with read.struct, the commands summarise.Fst and summarise.lnPD are used respectively. For example, using the data we read in previously, the commands:
   lnpd <- summarise.lnPD(data)
fst <- summarise.Fst(data)
   will create the tables ‘lnpd’ and ‘fst’ containing the summarised ln P(D) and F_ST data respectively.

Since the assigned clusters will not necessarily be in the same order for each STRUCTURE/ INSTRUCT run, summarise.Fst has a second argument stdevopt to best order the F_ST data. Setting stdevopt = 1 (the default) uses no optimisation procedure, setting stdevopt = 2 orders the F_ST data by value, whilst setting stdevopt = 3 will use correlation matrices between runs to determine the optimal F_ST data ordering. Using the correlation matrices will prompt the user to enter the file path for the original files again.

NOTE: F_ST statistics are only available from STRUCTURE data generated under the admixture model. Output generated in INSTRUCT (even under the admixture model) or under other STRUCTURE models will cause an error.

5. To calculate, ΔK and ΔF_ST from the summarised ln P(D) and F_ST data, the command calc.delta(input, Fst = FALSE) is used. The default is ln P(D) data. The argument Fst determines whether the input data is F_ST or ln P(D) data. For example, the commands:
   deltaK <- calc.delta(lnpd)
deltaF <- calc.delta(fst, Fst= TRUE)
   will create the tables ‘deltaK’ and ‘deltaF’ containing the calculated ΔK and ΔF_ST statistics as well as the intermediary statistics L′(K), L′′(K), F′(K) and F′′(K) used to calculate ΔK and ΔFST (Campana et al. 2011; Evanno et al. 2005). These statistics are denoted by LprK, LdprK, FprK, and FdprK respectively.

6. To perform Q-matrix correlations, the commandcorr.Qmatrix(filepath = "./", rowncol = TRUE, avmax = TRUE, pvalue = FALSE, raw = TRUE, r = 0.99, p = 0.05) is used. The file path default is the current directory. The default data file format is the STRUCTURE format. Setting instruct to TRUE indicates the presence of INSTRUCT data. The arguments rowncol, avmax, and pvalue determine whether the rows-and-columns criterion (Campana et al. 2011), the average max correlation criterion (Cockram et al. 2008), permutation tests are used. The argument raw determines whether unfiltered Q matrix correlation matrices are outputted. The arguments r and p determine the minimum Pearson coefficient and the maximum p values to be considered a significant correlation. The data generated by corr.Qmatrix are entered into an S4 object of class ‘QmatrixFilt’. For example:
   Qmat <- corr.Qmatrix(“C:/Results/”, pvalue = TRUE)
   will generate an S4 object named ‘Qmat’ in which the STRUCTURE output in the folder C:/Results/ are correlated and filtered both by the average maximum correlation and the rows-and-columns criteria. Permutation tests are also used. The unfiltered correlation matrices are outputted. The minimum Pearson coefficient is 0.99 and the maximum p value is 0.05.

7. To view, the various correlation matrix outputs, the commands @rowncol, @avmaxcorr and @rawcorr are used. For example: Qmat@rowncol will show the output of the rows-and-columns criterion
   Qmat@avmaxcorr will show the output of the average maximum correlation criterion
   Qmat@rawcorr will show the unfiltered correlation matrices.

Bugs and Contributing

Please report all bugs (and any suggestions for improvements) to Michael G. Campana ([email protected]).

CorrSieve Citation

Campana et al. 2011. CorrSieve: software for summarizing and evaluating Structure output. Mol. Ecol. Resour. 11:349-352. https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1755-0998.2010.02917.x

Acknowledgments

Rita Cannas helpfully checked the method for calculating ΔK. Dent Earl , Elena López, and Michał Żmihorski identified bugs in the software.

References

Campana et al. 2011. CorrSieve: software for summarizing and evaluating Structure output. Mol. Ecol. Resour. 11:349-352. https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1755-0998.2010.02917.x

Cockram et al. 2008. Association mapping of partitioning loci in barley. BMC Genet. 9: 16–29. https://bmcgenet.biomedcentral.com/articles/10.1186/1471-2156-9-16.

Evanno et al. 2005. Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study. Mol. Ecol. 14: 2611-2620. https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1365-294X.2005.02553.x.

Gao et al. 2007. A Markov Chain Monte Carlo approach for joint inference of population structure and inbreeding rates from multilocus genotype data. Genetics. 176: 1635-1651. https://www.genetics.org/content/176/3/1635.

Hester et al. 2019. remotes: R package installation from remote repositories, including 'GitHub'. R package version 2.1.0. https://CRAN.R-project.org/package=remotes.

Pritchard et al. 2000. Inference of population structure using multilocus genotype data. Genetics. 155: 945–49. https://www.genetics.org/content/155/2/945.