Interspecific Sample Prioritization Can Improve QTL Detection With Tree-Based Predictive Models

Abstract

Due to increasing demand for new advanced crops, considerable efforts have been made to explore the improvement of stress and disease resistance cultivar traits through the study of wild crops. When both wild and interspecific hybrid materials are available, a common approach has been to study two types of materials separately and simply compare the quantitative trait locus (QTL) regions. However, combining the two types of materials can potentially create a more efficient method of finding predictive QTLs. In this simulation study, we focused on scenarios involving causal marker expression suppressed by trans-regulatory mechanisms, where the otherwise easily lost associated signals benefit the most from combining the two types of data. A probabilistic sampling approach was used to prioritize consistent genotypic phenotypic patterns across both types of data sets. We chose random forest and gradient boosting to apply the prioritization scheme and found that both facilitated the investigation of predictive causal markers in most of the biological scenarios simulated.

Association mapping analysis of oil palm interspecific hybrid populations and predicting phenotypic values via machine learning algorithms

“…Additionally, another competitive machine learning method, gradient boosting, was used in a sugarcane study to predict yield grade successfully (Charoen-Ung & Mittrapiyanuruk, 2018). Gradient boosting is a machine learning method with high robustness and flexibility, and is often a top ranking machine learning method in comparative evaluations (Shin & Nuzhdin, 2021;Zhang et al, 2021).…”