Prediction of complex traits: Conciliating genetics and statistics

Area of Science:

• Quantitative genetics and genomics
• Statistical modeling and machine learning

Background:

• Predicting complex traits is crucial in various biological and agricultural fields.
• Traditional methods often rely on pedigree and phenotype data.
• Advances in molecular and epigenetic technologies provide new data sources.

Purpose of the Study:

• To review and categorize methods for predicting complex traits.
• To discuss the evolution of statistical approaches in trait prediction.
• To highlight the impact of diverse data sources on prediction accuracy.

Main Methods:

• Review of prediction methodologies: deterministic vs. stochastic.
• Analysis of information sources: genealogies, phenotypes, nucleotide sequences, expression data, epigenetics.
• Overview of statistical methods: conditional expectation, best linear unbiased prediction (BLUP), Bayesian methods, machine learning (ML).

Main Results:

• Prediction methods vary in their nature, data requirements, and statistical underpinnings.
• Diverse data sources, from traditional to omics, enhance prediction capabilities.
• Statistical methods have progressed from basic linear models to sophisticated ML techniques.

Conclusions:

• The field of complex trait prediction has evolved significantly, driven by methodological and data advancements.
• Machine learning and diverse data integration represent the current frontier in trait prediction.
• Continued methodological innovation is essential for accurate complex trait prediction.