A Gaussian process model and Bayesian variable selection for mapping function-valued quantitative traits with...

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• High-dimensional phenotyping now includes time as a phenotypic dimension.
• This enables quantitative trait locus (QTL) studies for function-valued traits like growth and development.
• Current functional trait analysis methods and software tools for QTL mapping are limited.

Purpose of the Study:

• To propose a novel Bayesian Gaussian process (GP) approach for functional QTL mapping.
• To develop an efficient method for analyzing time-varying phenotypic data and identifying genetic variants.
• To provide a flexible and robust software tool for practical implementation.

Main Methods:

• Utilized Gaussian processes (GPs) to model time-varying coefficients of molecular marker effects on quantitative traits.
• Employed an efficient gradient-based algorithm for estimating GP tuning parameters.
• Developed a stepwise algorithm for model selection in genetic variant analysis, using Bayesian posterior probability increase as a stopping rule.
• Addressed incomplete datasets with high missing data rates (over 50%).

Main Results:

• The GP approach demonstrated flexibility in modeling diverse phenotypic trajectories with low computational cost.
• The model selection approach reliably identified putative quantitative trait loci (QTL) in simulated and real datasets.
• The method is applicable to datasets with substantial missing phenotypic data.

Conclusions:

• The proposed Bayesian GP method offers a flexible and computationally efficient solution for functional QTL mapping.
• The developed software package 'GPQTLmapping' facilitates practical implementation and analysis of time-varying traits.
• This approach enhances the ability to study genetic influences on traits over time, particularly in the presence of missing data.