International Journal of Plant & Soil Science

Crop improvement increasingly depends on resolving how inherited sequence variation is converted into context-dependent molecular states and, ultimately, agronomic phenotypes. Genomics supplies stable information on allelic and structural variation, whereas transcriptomics, proteomics and metabolomics progressively capture regulatory activity, functional molecular abundance and biochemical state. Their integration is therefore attractive, but the central breeding question is not whether more data can be combined; it is when added molecular layers improve causal inference, trait prediction or selection decisions beyond a strong genomic baseline. This critical narrative review evaluates evidence linking the four core omics layers in crop research, with emphasis on pangenome-aware genomics, expression and association mapping, proteome and post-translational regulation, metabolic endophenotypes, multi-block statistical integration, network models and machine-learning approaches. The literature indicates strong value for cross-omics integration in candidate-gene prioritisation, pathway reconstruction and mechanistic interpretation, particularly when genomic associations are anchored to expression, protein or metabolite variation measured in relevant tissues and environments. Evidence for routine prediction gains is less uniform. Metabolomic and transcriptomic profiles can be highly predictive in some populations, yet comparative studies show that combined multi-omics models may fail to outperform genomic prediction when molecular measurements are noisy, environmentally contingent, redundant with genetic relationships or evaluated without stringent out-of-environment validation. Proteomic integration adds biologically important information that is not reliably inferred from messenger RNA, but coverage, cost and post-translational complexity remain substantial constraints. The field is moving towards graph pangenomes, matched longitudinal sampling, interpretable multimodal models and integration with deep phenotyping and environmental covariates. The most defensible near-term strategy is consequently selective rather than maximal integration: deploy the molecular layer whose biological position and measurement reliability match the breeding question, validate improvements against genomic baselines, and prioritise designs that can demonstrate transfer across environments and breeding cycles.