TSDFGS

This package provides two useful tools for determining a proper training set for genomic selection or prediction. In order to have a better prediction of the genomic estimated breeding values (GEBV), the training set should be optimized as highly genomic correlated with the test set as possible. Several criteria have been published previously, including:

Prediction error variance (PEV; Akdemir et al., 2015)
Generalized coefficient of determination (CD; Laloë et al., 1993)

Our research provides an alternative criterion, r-score, which is derived from Pearson's correlation between GEBVs and phenotypic values of a test set. We could determine both a reasonable training set size and an optimal training set for building a prediction model with the criteria. Both functions are provided in our package.

For more information on the method, please check our published article:

Training set determination for genomic selection (Ou et al., 2019)

Installation

The development version of TSDFGS can be installed from GitHub (recommend):

# library(devtools)
install_github("oumarkme/TSDFGS", dependencies = TRUE, force = TRUE)

You may also install the stable version from CRAN, which the most recent function may not include.

install.packages("TSDFGS")

All functions were developed under r version 4.2.0 and tested in both version 3.6.3 and 4.1.1. However, we recommend you to use this package with R version > 3.6.3.
Rcpp and RcppEigen were used in the package. In addition, the core C++ scripts were published on GitHub for those who want a better performance.

Main functions

r_score: Function for calculating r-score (more).
pev_score: Function for calculating PEV score (more).
cd_score: Function for Calculating CD score (more).
optTrain: Function for determining optimal training set (more).
SSDFGS: Function for determining reasonable training set size (more).

Note that cd_score() and pev_score() are using functions from STPGA package. Try their package for advanced usage.

If you wanted to install the recent version of r_score() function independently, you may download the rscore.cpp script from my GitHub repo and install it by:

library(Rcpp, RcppEigen)
download.file("https://raw.githubusercontent.com/oumarkme/TSDFGS/main/src/rscore.cpp", "rscore.cpp")
Rcpp::sourceCpp("rscore.cpp")

Example dataset

An example data provided for testing this package. The rice genome data was published by Zhao et al. (2011) in their research. Raw dataset is available at the Rice Diversity website. Pre-arranged dataset is available in this GitHub repository and you may loaded in R by

download.file("https://github.com/oumarkme/TSDFGS/raw/main/data/geno.rda", "geno.rda")
load("geno.rda")

Authors

Jen-Hsiang Ou
- Author, maintainer
- E-mail: jen-hsiang.ou@imbim.uu.se
- Department of Medical Biochemistry and Microbiology, Uppsala University, Uppsala, Sweden
Po-Ya Wu
- Author
- E-mail: Po-Ya.Wu@hhu.de
- Institute for Quantitative Genetics and Genomics of Plants, Heinrich Heine University, Düsseldorf, Germany
Chen-Tuo Liao
- Author, thesis advisor
- E-mail: ctliao@ntu.edu.tw
- Department of Agronomy, National Taiwan University, Taipei, Taiwan

Citing this package

If you make use of TSDFGS package in your research, we would appreciate a citation of following papers:

Ou, J.-H., and Liao, C.-T. Training set determination for genomic selection. Theor Appl Genet 132, 2781–2792 (2019). https://doi.org/10.1007/s00122-019-03387-0
Wu, P.-Y., Ou, J.-H., and Liao, C.-T. Sample size determination for training set optimization in genomic prediction. Under review (2022).

xiahui625649 / TSDFGS