pKWmEB: integration of Kruskal–Wallis test with empirical Bayes under polygenic background control for multi-locus genome-wide association study

Abstract

Although nonparametric methods in genome-wide association studies (GWAS) are robust in quantitative trait nucleotide (QTN) detection, the absence of polygenic background control in single-marker association in genome-wide scans results in a high false positive rate. To overcome this issue, we proposed an integrated nonparametric method for multi-locus GWAS. First, a new model transformation was used to whiten the covariance matrix of polygenic matrix K and environmental noise. Using the transferred model, Kruskal–Wallis test along with least angle regression was then used to select all the markers that were potentially associated with the trait. Finally, all the selected markers were placed into multi-locus model, these effects were estimated by empirical Bayes, and all the nonzero effects were further identified by a likelihood ratio test for true QTN detection. This method, named pKWmEB, was validated by a series of Monte Carlo simulation studies. As a result, pKWmEB effectively controlled false positive rate, although a less stringent significance criterion was adopted. More importantly, pKWmEB retained the high power of Kruskal–Wallis test, and provided QTN effect estimates. To further validate pKWmEB, we re-analyzed four flowering time related traits in Arabidopsis thaliana, and detected some previously reported genes that were not identified by the other methods.

Genome-Wide Association Study Reveals Novel Marker-Trait Associations (MTAs) Governing the Localization of Fe and Zn in the Rice Grain

“…Nevertheless, MLM being a single locus method that allows testing of one marker locus at a time, had an inherent limitation in matching the real genetic architecture of the complex traits that are under the influence of multiple loci acting simultaneously (Kaler and Purcell, 2019). Multi-locus models like FASTmrEMMAa (Zhang et al, 2018), LASSO (Xu et al, 2017), BLASSO (Tamba et al, 2017), FarmCPU, pLARmEB (Zhang et al, 2018), and pKWmEB (Ren et al, 2018) are being used to overcome the limitation above. A few recent studies on plant height and flowering time (Wallace et al, 2016), ear traits (Zhu et al, 2018), and starch pasting properties (Xu et al, 2018) in maize, yield-related features in wheat (Ward et al, 2019), stem rot resistance in soybean (Wei et al, 2017), agronomic traits in foxtail millet (Jaiswal et al, 2019), and panicle architecture in sorghum (Zhou et al, 2019), have demonstrated the power of the FarmCPU model that uses both fixed effect and random effect models iteratively to effectively control the false discovery.…”

Genomic Prediction Accuracy of Seven Breeding Selection Traits Improved by QTL Identification in Flax

“…Thus, alternative multi-locus methods have been proposed [14], including the multi-locus random-SNP-effect mixed linear model (mrMLM) [9,15], the FAST multi-locus random-SNP-effect EMMA (FASTmrEMMA) [16], the polygene-background-control-based least angle regression plus empirical Bayes (pLARmEB) [17], the iterative modified-sure independence screening EM-Bayesian LASSO (ISIS EM-BLASSO) [18], and the integration of the Kruskal-Wallis test with empirical Bayes under polygenic background control (pKWmEB). These methods adapt statistical models that simultaneously test multiple markers and, doing so, substantially increase the statistical power while simultaneously reducing Type 1 errors and running time [9,[15][16][17][18][19]. These methods also usually adapt LOD scores (usually LOD ≥ 3), rather than the stringent Bonferroni correction (0.05/number of SNPs) [19], thus empowering the detection of more large and small effect QTNs [10].…”

“…These methods adapt statistical models that simultaneously test multiple markers and, doing so, substantially increase the statistical power while simultaneously reducing Type 1 errors and running time [9,[15][16][17][18][19]. These methods also usually adapt LOD scores (usually LOD ≥ 3), rather than the stringent Bonferroni correction (0.05/number of SNPs) [19], thus empowering the detection of more large and small effect QTNs [10]. In contrast to these multi-locus models, the fixed and random model circulating probability unification (FarmCPU) [20] still uses Bonferroni correction and mostly detects a few large-effect QTNs [10].…”

mrMLM v4.0: An R Platform for Multi-locus Genome-wide Association Studies

“…To confirm the correctness of our software mrMLM v4.0, the same simulation datasets in Zhang et al [20] (File S4) were re-analyzed by the above six methods and three current methods (GEMMA [21], FarmCPU [14] and EMMAX [4]). As a result, our six methods were better than the three current methods (Tables S9 to S11; Figs.…”

“…Thereafter, Liu et al [14] developed FarmCPU. Based on the advantages of the random model of QTN effect over the fixed model [15], recently, we have developed six multi-locus methods: mrMLM [16], FASTmrMLM [17], FASTmrEMMA [18], ISIS EM-BLASSO [19], pLARmEB [20] and pKWmEB [21] (Files S1 and S2). These methods include two stages.…”