Computational Molecular Biology 2026, Vol.16, No.3, 159-180 http://bioscipublisher.com/index.php/cmb 163 primarily focus on additive genetic variance, while dominance, epistasis, and their interactions with environmental factors are often not explicitly modeled (Abney et al., 2001; Chen et al., 2015; Zhu et al., 2015). These effects may be partially absorbed into genetic variance estimates in pedigree-based analyses, but are difficult to identify in SNP-based analyses using unrelated individuals. In summary, “missing heritability” is more appropriately understood as a difference in the identifiability of genetic variance under different statistical frameworks, rather than as an actual absence of genetic mechanisms. Pedigree-based and SNP-based heritability estimates reflect different aspects of genetic architecture; their discrepancy provides important insights into the multi-layered genetic basis of complex traits, rather than constituting mutually contradictory evidence. To facilitate a systematic comparison between traditional marker-assisted approaches and genome-wide statistical genetic methods in terms of research objectives, statistical assumptions, and application scenarios, representative methods-including linkage analysis, candidate gene approaches, and GWAS/GCTA-GREML-are summarized in Table 1. Table 1 Comparison between traditional marker-assisted approaches and genome-wide statistical genetic methods Comparison dimension Traditional approaches (Linkage/Candidate gene) Genome-wide approaches (GWAS/GCTA-GREML) Research starting point Hypothesis-driven candidate regions or genes Genome-wide, hypothesis-free scanning Primary data type A limited number of molecular markers (e.g., RFLP, SSR) High-density SNPs or whole-genome sequencing data Study population Structured populations or pedigrees Natural populations or breeding populations Scale of genetic signal Single loci or local linkage intervals Genome-wide, multi-locus signals Core statistical assumptions Strong prior assumptions with limited multiple testing Explicit modeling of population structure and multiple testing Main analytical objective Identification of QTLs or candidate genes Estimation of heritability and genetic architecture Interpretation of results Locus-specific effects and biological interpretation Variance decomposition and predictability assessment Suitability for complex traits Limited power for highly polygenic traits Well suited for highly polygenic traits Role in breeding Marker-assisted selection and locus validation Guiding genomic selection and breeding strategy design Representative methods Linkage analysis, candidate gene analysis GWAS、GCTA、GREMLGWAS, GCTA, GREML Methodological limitations Limited resolution, power depends on population design Sample-size dependent, limited causal interpretation Comparison dimension Traditional approaches (Linkage/Candidate gene) Genome-wide approaches(GWAS/GCTA-GREML) Note: Traditional marker-assisted approaches rely mainly on linkage analysis and candidate gene strategies to identify QTLs or functional loci using a limited number of molecular markers in structured populations (Fang et al., 2001). Genome-wide methods, represented by GWAS and GCTA/GREML, use dense genome-wide markers to build statistical models for estimating heritability and dissecting the genetic architecture of complex traits. Although these approaches differ substantially in statistical assumptions and analytical scale, they are historically and conceptually connected in crop genetic improvement (Fang and Wu, 2026). 3 Principles of Constructing the Genome-wide Relationship Matrix (GRM) 3.1 Standardized genotype matrix The construction of the genome-wide relationship matrix (GRM) is fundamentally based on a standardized genotype matrix. For each SNP locus in diploid species, genotypes are typically encoded as 0, 1, or 2, representing the number of copies of the reference allele carried by an individual. However, directly using these raw genotype encodings may introduce bias, because differences in allele frequencies across loci can lead to heterogeneity in variance (Forni et al., 2011; Wang et al., 2025). To avoid such bias, genotype data must be standardized. Let the population frequency of the reference allele at a given locus be p. The observed genotype xfor an individual at that locus is standardized as: z= x-2p √2p(1-p)
RkJQdWJsaXNoZXIy MjQ4ODYzNA==