Plants generate a diverse array of metabolites that are instrumental in their growth, productivity, and resilience to biotic and abiotic stresses. Importantly, these metabolites also contribute to human health (Fang et al., 2019). Specifically, flavonoids and polyphenolic compounds synthesized via the phenylpropanoid pathway represent an important group of such metabolites. They are prevalent throughout the plant kingdom (Yonekura-Sakakibara et al., 2019). Notably, flavonoids exhibit numerous biological activities in humans, including combating oxidative stress. They also play a key role in plant growth regulation (Kumar and Pandey, 2013; Mierziak et al., 2014). Polyphenols, on the other hand, are utilized to defend against attacks from pathogens and herbivores (Zacares et al., 2007; Kaur et al., 2010; Gaquerel et al., 2014; Alamgir et al., 2016). Additionally, in the Brassicaceae family, glucosinolates are crucial for defensive purposes and maintaining auxin balance. Remarkably, they also possess properties that can aid in preventing cancer in humans (Mithen et al., 2000; Kliebenstein et al., 2005).

Deciphering the genetic foundations of metabolites is essential for altering or reconstructing metabolic pathway for specific molecules. These molecules have major impacts on both plant physiological characteristics and human health. The emergence of advanced metabolite detection technologies and next-generation sequencing has enabled numerous studies exploring the regulatory mechanisms of targeted metabolites in various species. These efforts provide theoretical frameworks for the modulation of plant metabolism (Fernie and Tohge, 2017). Notably, by modifying phytoene synthase, a critical enzyme in the beta-carotene synthesis pathway, the total carotenoids in "Golden Rice 2" increased up to 23-fold compared to the original Golden Rice (Paine et al., 2005). Beyond manipulating specific metabolite synthesis pathways, the modulation or modification of transferases, which are involved in modifying a group of acceptors, also provides an alternate route for adjusting plant metabolism (Wang et al., 2022; Wu et al., 2022).

Building a metabolic network and identifying candidate genes for each reaction are key tasks for understanding the genetic foundations of metabolites. Unfortunately, the construction process of a metabolic network can be very time-consuming, especially for non-model organism. Traditionally, the metabolic network construction starts with functional annotation of genes. Based on homology searches, several computational approaches, such as Pathologic (Karp et al., 2002) and AUTOGRAPH-method (Notebaart et al., 2006), have been developed to automate the reconstruction procedure. According to those approaches, several wide used metabolite network databases have been established and widely utilized in metabolomics, such as KEGG (Kanehisa et al., 2023), PMN (Hawkins et al., 2021), Reactome (Milacic et al., 2024), the Cyc databases (Notebaart et al., 2006) and PathBank (Wishart et al., 2024). After the established network, reverse genetic approaches and top-down methods were then widely employed to identify candidate genes for each reaction (Fernie and Tohge, 2017). With the rapid development of multi-omics, the integrative pathway analysis techniques are required for genetic regulatory analysis of metabolite pathway in the multi-omics and/or multi-cohort setup (Maghsoudi et al., 2022). Towards this goal, many surveyed methods were developed to computing the summary statistics for the genes/markers or performing pathway enrich pathway enrichment analysis in the setup, such as ActivePathways (Paczkowska et al., 2020), iODA (Yu et al., 2020), PaintOmics 3 (Hernandez-de-Diego et al., 2018), ReactomeGSA (Griss et al., 2020). With the help of those metabolite network databases and integrative pathway analysis techniques, investigations into global molecular changes, including metabolite profiling and transcriptomics, deepen our understanding of the genetic underpinnings and regulatory mechanisms of plants (Yu et al., 2014; Dai et al., 2015; Stefan et al., 2015; Han et al., 2017; Li et al., 2021; Li et al., 2022).

Wheat (Triticum aestivum L.) is a staple cereal crop, accounting for approximately 20% of human calorie intake (Ling et al., 2013). While numerous studies using metabolomic and transcriptomic profiling have increased the understanding of the developmental regulatory network of wheat kernels (Yu et al., 2014; Dai et al., 2015; Stefan et al., 2015; Han et al., 2017; Li et al., 2021; Li et al., 2022), the regulatory mechanisms remain elusive. This is due to a vast array of basic molecules undergo chemical modifications in the plant world, making it both more challenging and labor-intensive to construct a comprehensive metabolic network, particularly for non-model organisms (Wang et al., 2019). Specific reactions producing species-specific metabolite cannot be predicted only based on traditional homology searches. In addition, the slow rate of structure elucidation for metabolites and the high false positive rate of candidate genes, screened merely based on correlation analysis between transcripts and metabolites. In this study, a total of 625 known metabolites in wheat kernels were examined at three developmental stages: kernels during grain filling (FK), mature kernels (MK), and germinating kernels (GK) in two varieties, Aikang 58 (AK58) and Jingdong 8 (JD8). By integrating RNA-Seq data, a broader metabolic network based on structure-guided chemical modifications, was created for candidate gene exploration. The validation of two candidate genes further underscored our approach as an effective tool for constructing the metabolic regulatory network.