The integration of genomic resources into breeding pipelines has reshaped crop improvement across many species [171,172,173,174], and similar approaches are increasingly being applied to industrial hemp [175,176]. For much of its cultivated history, industrial hemp breeding operated on phenotypic selection alone, a process that is inherently slow, labor-intensive, and inefficient for complex traits influenced by numerous loci and strong genotype-by-environment interactions [177,178]. The emergence of high-quality reference genomes, pangenomic resources, genome-wide markers, and trait-mapping studies has created opportunities to transition from phenotype-driven selection toward predictive, genome-informed breeding strategies [21,27,58].

The modern molecular breeding pipeline begins with the identification of genomic regions associated with target traits through linkage mapping, genome-wide association studies (GWASs), transcriptomics, and pangenomic analyses [21,106,113]. These approaches generate candidate loci controlling economically important traits such as cannabinoid composition, flowering time, sex determination, disease resistance, fiber quality, seed oil composition, and abiotic stress tolerance [75,116,179]. Functional validation using transient expression systems, gene silencing technologies, and genome editing subsequently provides biological evidence linking genotype to phenotype [106,152]. Once validated, these loci can be converted into molecular markers that support selection within breeding populations, thereby creating a direct pathway from genomic discovery to cultivar development.

Marker-assisted selection (MAS) remains one of the earliest applications of molecular breeding in industrial hemp breeding. By using tightly linked markers to identify favorable alleles at early developmental stages, breeders can substantially reduce reliance on labor-intensive phenotyping and accelerate breeding cycles. Marker systems associated with cannabinoid profiles, sex determination, and powdery mildew resistance illustrate the potential of MAS for traits controlled by major-effect loci [72,175,180]. However, MAS is most effective for traits controlled by a small number of loci with large effects, and its utility is limited for highly polygenic traits [181]. Moreover, the development and validation of minimal SNP genotyping arrays in Cannabis sativa have enhanced the ability to differentiate among various cultivars and germplasms at relatively affordable costs [148]. The development of SNP-based genotyping platforms and Kompetitive Allele Specific PCR (KASP) markers has further enhanced the efficiency of germplasm characterization, cultivar identification, and marker-assisted breeding [182].

To overcome these limitations, genomic selection (GS) has emerged as a promising strategy for improving complex traits controlled by many small-effect loci [183,184]. Unlike MAS, GS uses genome-wide marker information to predict the breeding values of individuals, allowing selection without direct phenotyping. This approach captures both major and minor genetic effects and enables selection prior to phenotypic evaluation. In industrial hemp, GS holds significant promises for traits such as fiber yield, seed yield, and stress tolerance, where traditional selection is challenging. Because many commercially important industrial hemp traits exhibit complex genetic architectures and substantial environmental influence, GS can accelerate breeding far faster than traditional methods. While still emerging in industrial hemp, recent studies have demonstrated the feasibility of GS and related predictive approaches for traits such as flowering time, biomass accumulation, and cannabinoid composition [75,161]. Nevertheless, successful implementation requires large training populations, accurate phenotypic datasets, robust statistical models, and extensive marker coverage across diverse germplasm [185].

Developments in machine learning and artificial intelligence further expand the predictive capabilities of molecular breeding [186,187]. Multi-trait prediction models that integrate genomic, phenotypic, and environmental data have demonstrated improved accuracy compared with traditional statistical approaches [187]. As multi-omics datasets become increasingly available, integrating genomic, transcriptomic, metabolomic, and environmental information will enable more precise prediction of complex phenotypes and improve selection efficiency across diverse environments.

Molecular breeding strategies are also increasingly being integrated with speed breeding and controlled-environment agriculture [188,189,190]. By accelerating generation turnover, speed breeding enables more rapid population advancement and increases the efficiency of both MAS and GS [191,192]. When combined with genomic prediction and high-throughput phenotyping data, these methods create a self-improving breeding cycle where genomic data guide selection, phenotyping validates predictions, and new data continuously improve predictive models [193,194]. Such approaches have the potential to substantially reduce the time required to develop elite genotypes [189,193,194].

Before molecular breeding can be routinely implemented across industrial hemp breeding programs, several significant challenges still need to be overcome. Compared with major crops, Cannabis breeding has only recently gained access to high-quality reference genomes and large-scale genotyping resources [20,22,64,147,182]. Many economically important traits exhibit complex genetic architectures and strong genotype-by-environment interactions that are poorly understood [26]. In addition, limited access to diverse public germplasm collections, fragmented breeding efforts, inconsistent phenotyping protocols, and a lack of large multi-environment datasets have constrained the development and validation of robust molecular breeding tools [10,26]. As a result, the accuracy of marker-assisted and genomic selection approaches remains lower than that achieved in other major crop breeding systems.

Future progress will require the integration of pangenomics, functional genomics, high-throughput phenotyping, environmental data, and advanced predictive modeling into unified breeding frameworks (Figure 4). The development of large, well-phenotyped populations across diverse environments will be essential to improving genomic prediction accuracy and identifying stable marker–trait associations. Although still in its early stages in Cannabis sativa, controlled-environment cultivation systems and photoperiod manipulation strategies are already being applied to enable rapid generation cycling and accelerated breeding through speed breeding approaches [195,196]. These approaches have the potential to significantly shorten breeding cycles by enabling multiple generations per year, thereby increasing selection intensity and accelerating the fixation of favorable alleles. However, their efficiency in Cannabis remains genotype- and environment-dependent and requires further optimization across diverse industrial hemp germplasm [195]. Furthermore, combining genomic selection with speed breeding, controlled-environment agriculture, and genome-editing technologies will enable more rapid deployment of favorable alleles and accelerate genetic gain. Emerging machine learning approaches that integrate genomic, transcriptomic, metabolomic, and environmental datasets are expected to improve prediction of complex traits and support data-driven breeding decisions [197,198,199].

Looking ahead, the future of Cannabis molecular breeding will likely move toward integrated genome-to-phenome platforms that combine pangenome-informed allele discovery, functional validation of candidate genes, genomic prediction modeling, and targeted genome engineering. As advanced tools become available, including genotype-independent transformation systems, transgene-free genome editing, multiplex editing, and precision breeding technologies, direct manipulation of complex genetic networks that control cannabinoid biosynthesis, flowering time, stress adaptation, fiber quality, and seed composition will be achievable. Together, these integrated pipelines have the potential to transform Cannabis improvement from a largely selection-based method to a more predictive, engineering-driven discipline, accelerating the development of elite cultivars optimized for fiber, grain, cannabinoid, and resilience to changing climates.