BSLMM Prediction

Introduction of BSLMM Prediction

Bayesian Sparse Linear Mixed Model (BSLMM) is a mixed model of linear mixed model (LMM) and sparse regression model. Both linear mixed models (LMMs) and sparse regression models are widely used in genetic applications, including multi-gene modeling in genome-wide association studies. BSLMM assumes that the effect of genes on phenotype obeys a mixed distribution composed of two normal distributions, and the degree of mixing of the two normal distributions is determined by different mixing ratios. It should be noted that for case-control studies (assuming the phenotype code is 0-1), BSLMM can directly treat 0-1 as a continuous variable and process it through a linear model, called Linear-BSLMM, or process Probit links function through a generalized linear model, called Probit—BSLMM. The BSLMM phenotype prediction model method can improve the accuracy of genome-wide genetic locus information.

Genome-wide association study (GWAS) has found that tens of thousands of single nucleotide polymorphisms (SNPs) are associated with hundreds of complex diseases. It provides an unprecedented perspective for the study of the genetic basis of phenotypic variation. In addition, in the analysis of cross-omics data, phenotype prediction is also considered to be a key step in integrating functional genome sequencing research with GWAS.A more efficient and interpretable gene set test can be constructed in GWAS, and SNP weights can be set to infer it from the prediction model and used for the mapping study of expression quantitative traits. Researchers have invented many phenotypic prediction model methods, such as BSLMM, etc., to improve the prediction accuracy by using the information of the genetic locus of the whole genome.

Fig 1. Comparison of PVE estimates from LMM (blue), BVSR (red), and BSLMM (purple) in two simulation scenarios. (Xiang Z, Peter C, et al. 2013) Model

Advantages of BSLMM Prediction

• Bayesian Sparse Linear Mixed Model (BSLMM) is a mixed model of linear mixed model (LMM) and sparse regression model. These two approaches make very different assumptions, so are expected to perform well in different situations.
• The Bayesian Sparse Linear Mixed Model is a polygenic model. As the number of effect genes and heritability increase, BSLMM has a strong predictive ability.
• In the process of disease and phenotype analysis, the predictive ability of BSLMM is higher than other sparse models.
• BSLMM uses GEMMA software for analysis and can process large-scale data sets.

Application Filed

Using Bayesian sparse linear mixed model phenotype prediction model method can improve the prediction accuracy of genome-wide genetic locus information. The success of GWAS has greatly promoted the use of genetic information (except environmental and lifestyle information) to carry out complex disease risk prediction and assessment.

• In the population, the use of accurate phenotypic prediction of genetic markers is conducive to early prevention and intervention of complex diseases, and to promote the development of personalized medicines, so as to formulate treatment plans and predict curative effects.
• In animals or plants, accurate phenotypic prediction with genetic markers can help select ideal breeding individuals, thereby improving the effectiveness of breeding programs.

References

• Xiang Z, Peter C, et al. Polygenic Modeling with Bayesian Sparse Linear Mixed Models [J]. PLOS GENETICS, 2013. 9 (2) e1003264.