Xuanjun Fang , Xuanjun Fang

Author    Correspondence author
Computational Molecular Biology, 0, Vol. 16, No.   
Received: 01 Jan., 1970    Accepted: 01 Jan., 1970    Published: 15 May, 2026

Heritability, as a core concept, plays a critical role in explaining trait variation and predicting selection response. Traditional heritability estimation relies on pedigree information but is limited by pedigree completeness and environmental confounding. With the development of high-throughput genotyping and genome-wide association studies, the restricted maximum likelihood method based on genomic relationship matrices (GCTA/GREML) has provided a new pathway for estimating the heritability of complex traits. This study reviews the theoretical framework and statistical assumptions of the GCTA and GREML families, elucidates their logic in variance decomposition and differences from pedigree-based models, and focuses on analyzing the sources and interpretive boundaries of the “missing heritability” problem. Further, the study explores methodological extensions such as the LOCO strategy, functional annotation partitioning, and bivariate analysis, and discusses their application value in complex trait dissection and crop breeding, supported by both simulation and empirical studies. The results indicate that GCTA/GREML not only promotes a paradigm shift in heritability research but also provides theoretical support for genomic selection and molecular breeding design. In the future, with the accumulation of sequencing data and multi-environment big data, this method is expected to more comprehensively uncover the genetic basis of complex traits. Accordingly, this review focuses on clarifying the statistical interpretation of SNP-based heritability estimation rather than providing a general tutorial. Specifically, we (i) outline the statistical conditions required for meaningful comparisons between SNP-based and pedigree-based heritability estimates; (ii) formally define the estimand targeted by GREML and clarify its relationship to the concept of missing heritability; (iii) organize commonly used GREML extensions into a unified framework based on their inferential goals, assumptions, and diagnostic boundaries; and (iv) propose a workflow-oriented checklist to guide the interpretation of SNP heritability estimates in practice.