Omics landscapes in molecular mechanisms with Amomum tsaoko as an example

Abstract

To improve the quality and efficiency of cultivating Amomum tsaoko (AT), a non-model plant, it is crucial to understand the intrinsic molecular mechanisms underlying its growth. This review summarizes the significance of multi-omics in the study of plant molecular mechanisms and illustrates how multi-omics technology can solve the practical problems of non-model plants using AT as an example. In this review, we argue that nonlinear dimensionality reduction is more suitable for data organization in multi-omics because it is compatible with the nonlinear relationship between the components of systems biology. Subsequently, researchers have verified the strong vitality of multi-omics from three perspectives: the natural communication, breeding, and shade tolerance mechanisms of AT. Finally, we summarized some of the current commonly used plant genome databases and analyzed their utility for such research. We believe that our study makes a significant contribution to the literature because this review summarizes the multi-omics research process in detail, from data processing to application to the use of public databases, and illustrates the potential for the application of multi-omics with the example of a non-model plant, AT.

Highlights

• •Nonlinear dimensionality reduction was the main processing method in multiomics.
• •Volatile organic compounds were the hubs of Amomum tsaoko communication networks.
• •Cultivation density affected the growth efficiency of Amomum tsaoko.
• •A molecular breeding system was constructed for Amomum tsaoko.
• •bHLH family has been a key regulator of shade avoidance and tolerance.

1. Introduction

Molecular regulatory networks in plants are complex and varied and interact with each other. Changes in the physiological environment can induce regulation at multiple levels. When plants are injured by heat stress, they accumulate flavonoid glycosides and anthocyanins; further, the HsfA1 protein activates the HSR gene, acting as a positive regulator (Kan et al., 2023). Heat stress is an abiotic stress in nature that is co-regulated by multiple pathways; hormone signaling pathways, such as the jasmonate pathway, simultaneously regulate multiple biological pathways in plants. The jasmonate signaling pathway is a key hub for many processes in plants; jasmonate hormones act as signaling molecules that can affect the proteome and phosphoproteome of etiolated seedlings, inhibit or promote seed germination, and regulate stomatal closure in Arabidopsis thaliana through the STOMATA1 protein (Ghorbel et al., 2021; Zander et al., 2020). Thus, plant life activities are highly complex and interrelated, and single-omics is often limited to only one facet of plant biological processes. Integrated multi-omics involves a large amount of data and multiple biological processes, enabling the effective monitoring and analysis of complex molecular regulatory networks in plants. The integrated multi-omics approach involves deciphering defense mechanisms under various stresses (abiotic stresses, such as heat stress, and biotic stresses, such as pests and diseases) in plants; a series of pathways associated with plant defense mechanisms have been discovered thus far (Kan et al., 2023). Using localized trait-related genes, many high-yield, high-quality, and highly resistant crops have been bred (Bisht et al., 2023; Iqbal et al., 2021; Peng et al., 2020). This is key to achieving sustainable agriculture. Nowadays, the innovation of single-cell and spatial multi-omics techniques and analytical methods has helped researchers systematically study the genes and metabolism of specialized tissues and cell types at specific developmental stages (Depuydt et al., 2023).

Amomum tsaoko (AT) is a plant from the family Zingiberaceae; it is a herb that serves as a traditional Chinese medicine and a popular plant spice, belonging to one of the “five flavors” of food seasoning. The whole plant has a pungent taste, while its fruit is pungent and warming, with a strong aroma. It is traditionally used as an aromatic stomachic to relieve epigastric distension and pain, recurrent vomiting, food accumulation, and other related ailments (Fu et al., 2025).

Using the keyword “Amomum tsaoko,” a total of 844 related articles were searched in the Web of science and CNKI databases as of August 2025. The omics research on AT has only been developing recently, and since the publication of the first one in 2012, only a few omics articles have been published every year. Only in 2021 did the number exceed ten; the publication volumes were zero in 2015 and 2016 (Fig. 1). At present, the main fields of research on AT are “chemical composition analysis,” “volatile oil extraction,” “antibacterial,” “antioxidant,” “industrial developmental problems and solutions,” and so on. Currently, research with the help of multi-omics approaches is mainly based on metabolomic-based chemical composition studies (Fig. 2). Regarding the structure of the research field, relevant research at the molecular level, that is, information regarding the series of molecular regulatory networks within AT, is seriously lacking. As the economic value of AT has gradually been explored, people are paying an increasing amount of attention and are eager to have more mature theoretical systems to support the further development of the AT industry. However, AT cultivation is currently facing multiple challenges, such as the frequent occurrence of pests and diseases, a lack of high-quality germplasm resources, and incomplete resource evaluation systems. Problems such as seedling degradation, prominent pests and diseases, and insufficient water and fertilizer have resulted in an average yield of less than 4500 kg/hm² in AT-fruiting areas (Song et al., 2023b); furthermore, the semi-wild cultivation mode has led to unstable yields and serious damage to natural resources (Yang et al., 2020). Light, temperature, humidity, pollinators, and soil fertility affect the fruiting rate, growth, and development of AT. AT grows on slopes at an altitude of 1200–1800 m; such regions are associated with poor infrastructure, i.e., poor systems for irrigation and water conservation, and high transport costs during the harvesting season (Song et al., 2023a). Growing the same variety of AT in the same field for a long time has caused the degradation of the variety and quality, reduction of the resistance of AT to pests and diseases, and ultimately, a decrease in AT yield (Li, 2021; Lian et al., 2020). Although certain measures have been taken to solve these problems, there have been no substantial effects thus far. To improve the quality and efficiency of the AT industry and achieve high-quality AT cultivation, it is crucial to understand the intrinsic molecular mechanisms underlying the physiological processes of AT.

Fig. 1.

Sankey diagram and chord diagram of the number of publications of Amomum tsaoko related literature and multi-omics related literature by year, with red color indicating the number of Amomum tsaoko articles published by year and blue color indicating the number of Amomum tsaoko multi-omics articles published.

Fig. 2.

Current areas of research in AT multi omics.

In the digital information age, multi-omics data are extremely large, highly dimensional, and noisy, and have exceeded the statistical analysis capabilities of traditional classical models (Reel et al., 2021). Principal component analysis (PCA), which can be considered as an example, projects high-dimensional data in a low-dimensional space while preserving the original information of the data. However, PCA can only capture linear relationships within datasets and cannot effectively handle complex non-linear relationships in multi-omics data (Takefuji, 2025). In contrast, owing to its superior ability to handle large-scale, unstructured, and complex datasets, machine learning (ML) has been widely applied to the analysis of high-dimensional biological datasets (Samek et al., 2021; Wang et al., 2023). According to the central law of molecular biology, organisms view the process of information recycling as a computer-like processing of biological information (Furusawa, 2025). ML is the “visualization” of the information we need. Advanced biology and genetics have aided the identification of thousands of highly therapeutic genome sequences and numerous gene–phenotype associations; however, knowledge regarding complex traits remains limited (Han et al., 2023). Scientists have assigned transcription factors (TFs) to different cell types based on ML models and prioritized candidate regulators by combining cell-specific expression, network-neutrality measurements, functional annotation, and expression specificity (Ferrari et al., 2022). In recent years, deep-learning (DL)-based models have demonstrated competitive or even better performance than traditional linear regression models in big data-driven genome prediction, owing to their ability to automate feature extraction and enhance the representation of high-dimensional datasets (Gao et al., 2025). Thus, the convergence of biological big data and ML algorithms provides an unprecedented opportunity to understand complete molecular biology networks.

Various databases have enabled data sharing and reduced data redundancy, making centralized management and real-time updating possible. For example, PlantPAN4.0 (http://PlantPAN.itps.ncku.edu.tw/), which constructs a comprehensive network for transcriptional regulation, contains the latest 3428 nonredundant matrices for 18,305 TFs, facilitating the exploration of combinational and nucleotide variants of cis-regulatory elements in conserved non-coding sequences (Chow et al., 2024). Databases are often created as a centralized account of a particular type or category of research results, and can be considered as a large, informative, and highly relevant “reviews.” They are often added to research articles as evidence for conclusions or comparative analyses; examples include downloading gene sequences to build developmental trees (Gong et al., 2022), pathway analysis of genes and proteins (An et al., 2024), network pharmacology analysis (Yu et al., 2024), ML (Liu et al., 2024), and metabolite annotation (Otify et al., 2023).

This review summarizes the multi-omics research process in detail, from data processing to application to the use of public databases, and illustrates the potential for the application of multi-omics with the example of the non-model plant AT. In this review, we argue that nonlinear dimensionality reduction (DR) is more suitable for data organization in multi-omics because it is compatible with the nonlinear relationship between the components of systems biology. Subsequently, based on the current development of AT, researchers have verified the strong vitality of multi-omics from three perspectives: natural communication, breeding, and shade-tolerance (ST) mechanisms. Finally, we have summarized some of the current commonly used plant genome databases and analyzed their applicability and the challenges associated with their use.

2. Integration of multi-omics data

2.1. The necessity of DL in multi-omics data processing

Different omics approaches, such as genomics, epigenomics, transcriptomics, and proteomics, are often complementary and can collectively address overlapping or distinct fundamental biological questions when integrated. Therefore, the integration of multi-omics data contributes towards providing a comprehensive overview of systems biology. DL is an important field in bioinformatics and can be used for multi-omics data mining, bioinformatics data analysis, and bioinformatics model construction. In addition, it can automatically discover patterns and regularities in multi-omics data, which can help researchers analyze and study data more quickly and accurately (Chen et al., 2025). According to the mode of supervision, ML algorithms can be categorized into the supervised, unsupervised, and semi-supervised models. Data labeling is the essential difference between these three methods: supervised learning requires labeled data for training; unsupervised learning searches for structure and regularity in data without relying on data labels; and semi-supervised learning uses a small amount of unlabeled data and a large amount of labeled data for training (Kozamernik & Bračun, 2025). Unsupervised models, such as NMF (Nonnegative Matrix Factorization), K-means, PCA, SVD (Singular Value Decomposition), and MDS (Multidimensional scalingmodels), are mainly used for the classification of samples. The commonly used supervised models are the SVM (Support Vector Machines), KNN (K- Nearest Neighbor), NB (Naive Bayes), CNN (Convolutional Neural Network), CCA (Canonical Correlation Analysis), and LSTM (Long Short-Term Memory) models (Table 1). Semi-supervised methods represent a combination of the supervised and unsupervised methods (Greener et al., 2022). Among the multi-omics data of plants, phenotypic data and their predictions are the largest beneficiaries of DL. Over recent decades, knowledge acquired from basic research in plant biology has greatly (MacNish et al., 2025). However, existing gaps between basic research and breeding practice in plants still have to be overcome if we are to ultimately achieve the goal of precision-designed plant breeding expedited the progress of plant breeding and accelerated crop improvement. ML may translate biological knowledge and data into precision-designed plant breeding,mainly through two pathways. One path is to facilitate omics sciences in plant biology and expedite the discovery of agronomically utilizable genes and mutations to achieve knowledge-driven molecular design breeding. The other path is to directly apply ML techniques in commercial breeding programs to construct a variety of predictive models for achieving data-driven genomic design breeding (Crossa et al., 2025; Yan & Wang, 2023). Predicting plant phenotypes has been one of the main applications of machine learning. For instance, in a recent study, multispectral imaging and ML models were used to predict phenotypes such as potato leaf structure, growth habits, plant height, and leaf color, as well as diseases (Lapajne et al., 2025).

Table 1.

ML Linear and nonlinear DR modes. (Ayesha et al., 2020; J. Bao et al., 2024; Bergenstråhle et al., 2022; Cai et al., 2023; Reel et al., 2021; Samek et al., 2021; Siblini et al., 2019; Yan & Wang, 2023).

DR	Supervised	Unsupervised	Semi-supervised
Linear	DT, SVM, RankSVM, LDA, PLS, Orthogonal PLS, NB，KNN, LASSO, MDDM, SDR, GBFA, DSNPE, LSTM	PCA, Random PCA, SVD, ICA, CCA, NPP, LSI, MAF, SFA, ICA, FA, LSA, PP	Semi-CCA, SLDA, RER, LGC
Nonlinear	SVM, ANN, DNN, SVR, LASSO, LLP, LPP, Orthonormal LPP, MLKNN, SLFN, Isomap, LVQ, CNN, RF	LLE, AE, Joint NMF, KPCA, MDS, LLE, SOM, t-SNE, Lemon-Tree, SNF, BCC, NEMO, Kernel PCA, LE, Hessian LLE, K-means	CNMF, SSLE, Cop-Kmeans, PCKmeans, SSFCM, CCL, SSCEV,GFHF

DT, Decision Tree; SVD, Singular Value Decomposition; PCA, Principal Components Analysis; SVM, Support Vector Machines (Linear and nonlinear DR methods can be selected according to the characteristics of data sets); ICA, Independent Component Analysis; CCA, Canonical Correlation Analysis, Semi-CCA, Semi-supervised CCA; LLE, Locally Linear Embedding; LDA, Linear Discriminant Analysis; PLS, Partial Least Squares; NB, Naive Bayes; ANN, Artificial Neural Network; KNN, K- Nearest Neighbor; DNN, Deep Neural Network; SVR, Support Vector Regression; LASSO, Least Absolute Shrinkage and Selection Operator; AE, Autoencoder; LLP, Locality Preserving Projection; Multi-label Dimensionality Reduction via Dependence Maximization; CNMF, Constrained Non-negative Matrix Factorization; LPP, Locality Preserving Projections; NPP, Neighborhood Preserving Projection; SLFN, Single-hidden Layer Feedforward Neural Network; LSI, Latent Semantic Indexing; MDS, Multidimensional scaling; MAF, Maximum Autocorrelation Factors; SFA, Slow Feature Analysis; SDR, Sufficient Dimensionality Reduction; ICA,Independent Component Analysis; FA, Factor Analysis; LSA, Latent Semantic Analysis; PP, Projection Pursuit; KPCA, Kernel Principal Component Analysis; LLE, Locally Linear Embedding; SOM, Self-Organizing Map; LVQ, Learning Vector Quantization; t-SNE, t-Stochastic Neighbor Embedding; SNF, Similarity Network Fusion; BCC, Bayesian consensus clustering; NEMO, NEighborhood based Multi-Omics clustering; SLDA, Semi-Supervised Linear Discriminant analysis; LE, Laplacian Eigenmap; SSLE, Semi-Supervised Laplacian Eigenmap; GBFA, Graph Based Fisher Analysis; DSNPE, Discriminant Sparse Neighborhood Preserving Embedding; Cop-Kmeans, Constrained K-means; PCKmeans, Pairwise Constraints K-means; SSFCM, Semi-Supervised Standard Fuzzy C-means Clustering; CCL, Constrained Complete-Link; SSCEV, Semi-Supervised Clustering Ensemble by Voting; RER, Robust Embedding Regression; GFHF, Gaussian Fields and Harmonic Functions; LGC, Local and Global Consistency; CNN, Convolutional Neural Network; LSTM, Long Short-Term Memory; RF, Random Forest.

2.2. Linear DR

Multi-omics data are always accompanied by high dimensionality; therefore, DR or feature selection is usually performed before training ML models. Wrapper feature selection includes forward and recursive feature selection. In this regard, intrinsic feature-selection methods, such as the decision tree (DT), RF, and Gaussian Navier Bayes methods, are performed; these methods do not require specialized domain knowledge but may lead to the loss of important features (Mariammal et al., 2022; Sun et al., 2024; Yan & Wang, 2023). Additionally, several feature dimensions in different modalities may interfere with the final classification performance (Zheng et al., 2024). DR relies on a range of ML algorithms and provides alternative methods for feature extraction. Widely used linear DR methods include PCA, OPLS-DA (Orthogonal Partial Least Squares Discriminant Analysis), and LDA (Linear Discriminant Analysis) (Yan & Wang, 2023). PCA relies on the features of a linear combination to construct a principal subspace, which contains several feature vectors, and OPLS-DA is applied to the feature extraction and visualization of multi-omics data. PCA and OPLS-DA form a complementary combination in multi-omics data downscaling, assuming the core roles of exploratory analysis and target modeling, respectively (Pan et al., 2025). In a previous study, PCA (a supervised model) and OPLS-DA (an unsupervised model) were used together to classify seven tomato species; the combination of these two methods amplified the differences between tomato samples (Marukatat, 2023; Yan & Wang, 2023). The core advantage of LDA in multi-omics dimensionality reduction is maximising class separability through supervised learning, which is particularly suited to scenarios where predefined groups need to be clearly distinguished (e.g., disease diagnosis, breed identification). Its mathematical nature (optimising inter/intra-class variance ratios) allows it to outperform PCA in discriminative power and capture cross-modal discriminative patterns in joint multi-omics analyses (Wani et al., 2025). For example, using multiple brain imaging datasets consisting of more than 150 million features, and several genomics datasets with more than 500,000 features, Linear Optimal Low-Rank Projection outperforms other scalable linear dimensionality reduction techniques in terms of accuracy, while only requiring a few minutes on a standard desktop computer (Vogelstein et al., 2021).

2.3. Nonlinear DR

However, the dependencies between biological features can be complex and far from linear. Linear DR approaches may not adequately capture the complexity of diverse cell types, leading to insufficient representation of the data (Liu et al., 2025). Traditional linear models, such as PCA, are widely used for feature reduction and visualization; however, their reliance on linear and parametric assumptions may lead to misleading conclusions in nonlinear proteomic analyses. For instance, although principal component analysis successfully highlighted circadian gene expression patterns in kidney studies, it can oversimplify multifaceted biological interactions that better align with nonlinear behavior (Takefuji, 2025). Single-cell and spatial multi-omics sequencing data alone exhibit extremely high dimensionality and nonlinear features (Bergenstråhle et al., 2022; Yan & Wang, 2023). Potential of heat-diffusion for affinity-based transition embedding (PHATE), uniform manifold approximation and projection (UMAP), and t-SNE (t-Stochastic Neighbor Embedding) are among the downscaling methods that can characterize the nonlinear relationships of tens of thousands of different cell types by visualizing the structures in space (Moon et al., 2019). Among these nonlinear DR methods, t-SNE and UMAP are the most commonly used techniques in single-cell data analysis, but they are primarily employed for visualizing high-dimensional data in low dimensions rather than providing meaningful lower embeddings for downstream analysis. In addition, DNN (Deep Neural Network) can map discrete speckle-based expression profiles to high-resolution morphological images or identify new cell subpopulations missed by individual models (Bergenstråhle et al., 2022). As a nonlinear model, DNN captures complex nonadditive effects; further, it does not require the same sacrifice of goodness-of-fit as linear models to increase predictability or a priori knowledge of the underlying genetic models (Abdollahi-Arpanahi et al., 2020; Wang et al., 2023). The DNN method exhibits good potential for characterizing the relationship between genotypes and phenotypes. With genomic best linear unbiased prediction, light gradient boosting machine, support vector regression, DL-based genomic selection, and DL genome-wide association study (GWAS) as five classical models, DNN can be applied to various histological data to predict phenotypes; it utilizes multilayered hierarchical structures to dynamically learn features from raw data, avoiding overfitting and improving the convergence rate (Wang et al., 2023). Joint multi-omics DR is a sophisticated data processing method that can accurately detect multi-omics data regarding shared biological processes as well as complementary processes from multiple omics sources (Cantini et al., 2021). NMF is a method that decomposes a nonnegative matrix A (m × n) into feature matrices W (m × k) and coefficient matrix H (n × k) is another nonlinear method with the main goal of reducing the dimensionality of the data by reducing a large number of features (Georgaka et al., 2025). IntNMF, as a representative NMF method, the core idea is to jointly decompose multiple histological data layers into a unified biomodular structure by sharing the sample potential space (Hirst et al., 2025). It, at the same time, addresses the key challenges of heterogeneous integration of multi-omics data. It inherits the non-negativity and interpretability advantages of standard NMF, and is particularly good at solving the three core challenges of multi-omics integration: heterogeneous data fusion, cross-omics signal alignment, and interpretable module extraction (Castellano-Escuder et al., 2025; Milite et al., 2025). For example, IntNMF performs well in the task of classifying multi-omics single-cell data, i.e., detecting a large number of variant patterns in a histological dataset. This is observed for both simulated bulk data clustering and single-cell data clustering (Cantini et al., 2021). Nevertheless, the modeling and control of large nonlinearities remain challenging for researchers. Therefore, two modeling approaches based on first-principles models with a mechanical knowledge of the system and data-driven high-fidelity predictive system models have emerged (Zhao et al., 2023).

2.4. General summary

Multi-omics data are typically high-dimensional and characterized by complex, nonlinear relationships that linear dimensionality reduction (DR) methods like PCA may fail to capture effectively. Nonlinear DR techniques such as UMAP, t-SNE, and PHATE better preserve both local and global data structures, enabling more accurate identification of cellular subpopulations and biological trajectories. Furthermore, deep learning models can automatically extract complex nonlinear features, enhancing downstream prediction tasks. Integrative methods like IntNMF allow simultaneous extraction of shared and modality-specific features across multiple omics layers, offering superior data integration compared to linear approaches. Together, these nonlinear methods provide a more powerful framework for multi-omics data analysis, leading to improved biological insights and phenotype predictions.

3. Nonlinear relationships in systems biology networks

Omics involves understanding the complex systems of organisms through the holistic analysis of the genes, RNA, proteins, and metabolites in a sample via an unbiased, untargeted approach. There is an inherent connection between omics studies performed around the central law, the most classical statement being that the genome and transcriptome predict what is going to happen in the future and the metabolome validates this prediction. Next we will illustrate the connections between the omics in sequence from the regulation of gene expression to the production of metabolites according to the order of transmission of biological information.

3.1. Genomics and transcriptomics: DNA modifications, histones, and RNAs

The unique arrangement of bases determines the different functions of genes; the task of functional genomics is to annotate these genes, more than a thousand of which have now been identified in plants. Later, scientists realized that a single reference genome does not reflect the genetic diversity of a species and that genes are lost, gained, or structurally mutated during the evolution of a species. Therefore, the genes conserved among species with these variable genes were formed into a more complex pan-genome, which represents the diversity and duplicity of plant genomes; this reflects the necessity of core genomes for species identification. Genes govern all life processes, but the genome cannot completely explain biological processes; other components present downstream of genes are products of the post-genomics era (Sherman & Salzberg, 2020; Shi et al., 2023) that complement the genome. DNA methylation and covalent histone modifications are the most common factors that directly influence gene expression via epigenetic mechanisms (Haws et al., 2020). These chemical modifications carry genetic information that can be inherited in cells without altering the DNA sequences, breaking the classical genetic perception that the sequence of genes determines the genetic information of life; this strongly attacks the “genetic determinism” argument. Metabolites and epigenomes have a close dynamic connection, as discussed by many researchers (Bhatia et al., 2022). This is mainly reflected by the fact that chromatin modification & removal depend on various metabolites, such as central metabolic cofactors that catalyze such activities. Genes affected by epigenetic modifications ultimately alter the abundance of metabolites; many biosynthetic gene clusters also carry epigenetic modifications. The dysregulation of metabolic pathways has also been identified as a cause of epigenetic aging (Cawood & Ton, 2025; Wang et al., 2025). Usually, DNA methylation occurs in promoter regions and silences genes, whereas demethylation usually activates genes (probably achieved by preventing the binding of TF and RNA Polymerase II, thus inhibiting transcription) (Furci et al., 2023; Gardiner et al., 2020). In addition, DNA methylation affects other epigenetic modifications and DNA methylation–histone modification crosstalk is responsible for heterochromatin and open chromatin conformations (Cawood & Ton, 2025). Upon the activation of TFs and non-coding RNAs (ncRNAs), histone modifications recruit RNA polymerase II transcripts and potential RNA-modifying complexes, which, in turn, often alters chromatin conformation (Chen et al., 2025; Bhatia et al., 2022;). In this context, H3K9, H3K27, and H4K20 methylation are generally considered “inactivation” marks, and H3K4 and H3K36 methylation are considered “activation” marks (Chen et al., 2025; Liu, Mu, et al., 2023). The transcription process is also influenced by the DNA structure. Neighboring genes may also add antisense RNAs, interfering with gene expression; 60 % of the seats in rice have antisense transcripts (Chen et al., 2020; Kenchanmane Raju et al., 2019), and the three-dimensional structure of DNA allows TFs to regulate the expression of multiple genes, with regions of aberrant transcription altering the gene structure, enabling the recruitment of TFs (Zhong et al., 2023). At the same time, half of the TFs also interact with RNA in specific binding domains, thereby affecting gene expression (Oksuz et al., 2023). The discovery of X-chromosome inactivation marked the beginning of research regarding the effects of epigenetic modifications of ncRNAs (Bhatia et al., 2022); at present, epitranscriptomics, which includes covalent modifications of RNAs, is well-known to serve as a potential mediator of the post-transcriptional gene silencing of aberrant RNAs (Zhong et al., 2023). Moreover, ncRNAs are also involved in chromatin-mediated gene silencing; DNA rearrangements or the recruitment of the chromatin-modifying complex is indirectly involved in epigenetic modifications by remodeling the structure of chromatin (Chen et al., 2025; Chhabra, 2023). Thus, epigenetic modifications and RNAs can play regulatory roles at the gene and chromosomal levels. In addition, ncRNAs can interact with each other. Long non-coding RNAs (lncRNAs) and micro RNAs share many functions and signaling pathways, and they interact with each other to exert “sponge effects” on other ncRNAs (Panni et al., 2020; Wang et al., 2021). These large, diverse, and functionally complex ncRNAs are also responsible for differences in the number of cellular and protein-coding genes in organisms (Cai et al., 2025; Jung et al., 2025).

3.2. Proteomics: post-translational protein modifications

It has also been proposed that uncharacterized proteins can be systematically linked to proteins with known functions through functional proteomics, bridging the gene–protein annotation gap and aiding the discovery of potential proteins (Ponzini et al., 2022). Therefore, the annotation of protein function relies on high-quality genomic or transcriptional information, and proteins and metabolites are also important components of functional genomics. Proteomics initially arose in the context of the inability of genomics to explain the pattern of activities such as modification and processing of proteins, structural changes, and interactions among proteins or between proteins and other biomolecules. These activities affect the products of gene expression, and not the genes. Moreover, a gene can encode multiple proteins, and multiple different proteins may perform similar functions. The relationship between them is not linear; thus, it is necessary to analyze them at the protein level. Protein turnover modifications (PTMs) increase the complexity of proteins, and the amino acid residues of the modified core histones regulate their interactions with nucleic acids and protein factors (Millán-Zambrano et al., 2022). These PTM-associated histones are modulated at the level of epigenetic modifications of DNA, RNA, or chromatin structures; such modifications are essentially via the recruitment of various enzymes to genomic regions. The stability of the acetylation and methylation of histone lysines varies widely (Millán-Zambrano et al., 2022). Protein phosphorylation is important for the thermal stability, overall turnover (Potel et al., 2021; Wu et al., 2021), and site-specificity of proteins and can even affect their DNA-binding properties. Multisite phosphorylation modulates the active function of proteins and attenuates protein–protein interactions (PPIs) and protein–ligand interactions. Complexes composed of PPIs are also involved in several biological processes, such as DNA transcription and replication and signaling and can even lead to genetic variations.

3.3. Metabolomics: detection of specific metabolites

Metabolites are end products expressed by the genome after interactions with the environment; thus, metabolomics can serve as the closest determinant of the phenotype. The number of metabolites in a plant is much smaller than that of genes and proteins; this not only reduces the difficulty of metabolomics analysis but also magnifies small changes in genes and proteins. Overall, metabolomics is the study of all small-molecule metabolite profiles in an organism (Huang et al., 2025). Currently, there are no analytical platforms that can assess the complete set of metabolites in a plant, owing to the availability of only a limited number of metabolites across various databases and the level of technical operators. The entire plant kingdom is estimated to produce one million metabolites; however, there are still numerous potential metabolites yet to be identified, some of which are undetectable at lower levels but have notable biological activity (Pan et al., 2023). Therefore, it is necessary to improve the accuracy and sensitivity of detection methods to enhance the overall efficiency of high-throughput sequencing.

4. Establishment of communication networks in AT

Plants, animals, and microorganisms are in a dynamic relationship with each other; predation, defense, and competition for survival resources have led to their dynamic co-evolution. Plants release a series of volatile organic compounds (VOCs), including green leaf volatiles (GLVs), terpenoids, and amino acid derivatives when subjected to mechanical injuries and abiotic stresses (Aratani et al., 2023; Arimura & Uemura, 2025; Lin et al., 2021). The VOCs, which act as signaling molecules, are received by the neighboring plants or the distal tissues of the plant, which not only breaks the temporal and spatial limitations of the transmission of the vascular bundles, but also reminds the neighboring plants beforehand to prepare themselves for adverse conditions (Bergman et al., 2025; Dobránszki et al., 2025). In addition, VOCs also undertake tasks such as attracting beneficial microorganisms and repelling herbivorous insects. Such information sharing, and subsequently, evolved communication strategies by plants, also promote the dynamic development of populations, communities, and ecosystems (Kessler et al., 2023). Given the presence of large amounts of VOCs, different parts of one AT plant or different AT plants should ideally exhibit strong and frequent communication with the outside world; the plant sizes and densities of planting may also influence the communication strategies of AT populations. However, this property of AT has not yet been emphasized, and VOC-related studies have mainly focused on their pharmacological activities (Guo et al., 2024; Liao et al., 2022), extraction methods (Liang et al., 2023; Xie et al., 2022), and constituent composition (Fu et al., 2025; Wen et al., 2025). Deciphering the communication of AT in an ecosystem and applying it rationally are important for sustainable and eco-friendly agriculture. Several pieces of evidence currently support the communication function of AT. (1) AT has significant chemosensory potential. (2) AT relies on insects for pollination. (3) AT contains approximately 45 % of VOCs, including terpenoids (He et al., 2023), with a positive correlation with elevated (E)-nerolidol, a strong VOC signal that has been widely studied.

4.1. Applicability of multi-omics technology for AT-related research

This study links the problems in the growth and development of AT and related industries with present multi-omics research, offering perspectives on how multi-omics can address the three aspects of AT: natural communication, breeding, and ST mechanisms.

4.1.1. Interaction with microorganisms

Soil contains numerous microorganisms that are involved in plant growth and development processes, with host, soil, environmental, and anthropogenic factors affecting plant-microbe interactions (Diwan et al., 2022). These microorganisms are present in the phyllosphere, interroot, and endosphere of plants and are referred to as the phyllosphere microbiota, interroot microbiota, and endosphere microbiota, respectively (Faddetta et al., 2021). Plants provide nutrients and shelter to microorganisms, while microorganisms, in turn, support their host plants by fixing nitrogen, absorbing soil nutrients (Yadav et al., 2019), inducing plant defense mechanisms, forming biofilms that act as protective physical or chemical barriers, or directly antagonizing plant pathogens (Zeng et al., 2025), which is a mutually beneficial relationship (Patel et al., 2020). Multi-omics is a molecular approach used to analyze the relationship between plants and microorganisms.

With the development of high-throughput next-generation sequencing (NGS) and third-generation sequencing (TGS) technologies, complete genomes of hundreds of plant pathogens, such as those infecting rice, are available (Jang et al., 2020; Wang & Wang, 2023; Xu et al., 2021; Zhang et al., 2023). Studies have shown that AT has a significant inhibitory effect on Staphylococcus aureus and (E)-dec-2-enal is the main antibacterial component (Guo et al., 2024). The warm and humid environment in which AT grows, together with a certain degree of shade, dead leaves, and branches retained in the forests, provide bacteria and fungi with good conditions for reproduction. Thus, although significant attention has been given to pathogenic microorganisms that cause diseases such as leaf spot and seedling blight in AT (Table 2), research on beneficial microorganisms is lacking. Some microbial markers associated with agronomic traits and drug resistance phenotypes of bacteria have been identified via GWASs, and the number of related studies has increased substantially in recent years (Demirjian et al., 2023). The VOC 2,6-dimethyl-2,4,6-octatriene in AT was positively correlated with annual precipitation and inhibited the growth of the fungus Botrytis cinerea in soil (Li, Lu, et al., 2021). This may be a way for AT to improve its defense under wet conditions that cause microbial growth. The fact that plants and fungi jointly synthesize triterpenoids compounds indicates that they share certain genes, supporting the view that they are inseparable (Scouten et al., 2025; Xi et al., 2025). Assessment of microbial gene expression can also be performed with the help of the metatranscriptome, metaproteome, transcriptome, and proteome of plants (Diwan et al., 2022). Metabolites are a bridge between plant and microbial communication, and microbiologists now pin the mining of silent and unknown microbial genes on the identification of microbial metabolites (Aharoni et al., 2023). Therefore, in addition to metagenomics, metabolomics is now the most researched topic (Mishra et al., 2022), and plant inter-root secretions influence soil nutrient activation and microbial populations, as well as their own nutrient status (Dwivedi et al., 2025). A strong correlation exists between increased crop yields, plant root systems, and microbial communities. Inter-root secretions not only recruit beneficial microorganisms (Yang et al., 2023) but also influence leaf metabolism, thereby repelling herbivorous insects (Ma et al., 2025). How interroot microbes shape AT phenotypes is currently unknown, and it is expected that the underlying molecular mechanisms will be elucidated through the association of the metabolomics and stable isotope probing (SIP) (Sun et al., 2021). In the forest agro-complex system of AT, elevated soil phosphorus, potassium, and microbial populations were observed (Liu et al., 2019). However, this system was borrowed from other species and has not been tailored to the specific environmental conditions in which the AT grows, resulting in the loss of some nutrients and organic matter. A comprehensive understanding of the molecular mechanisms underlying soil microbial interactions and the introduction of beneficial and powerful microorganisms into the soil will enhance the adaptability of AT and improve its defenses in an “unmanaged” situation.

Table 2.

Major types of pests and diseases of AT, their symptoms and control measures. The names of insects and microorganisms in the table are Latin names, where Culicidae mainly refers to the larvae of Culicidae. (Bao et al., 2022; He, 2023; Li, 2020; Li, Bai, et al., 2022; Wu et al., 2022; Yu et al., 2023).

Type	Name	Disease	Symptoms	Preventive measures
Insect	Lymantria dispar		Eat the leaves	Biofungicide; Natural enemy; Bait and kill
	Chilo suppressalis		Infestation of stem	Timely reduction of dull heart plants; Drug prevention and treatment
	Locust		Eat the leaves; Affects flowering and fruiting, or even death	Catching; Destruction of egg masses
	Zygaenidae		Nutritional deficiency; Affects flower spike emergence; Fruits lesser	Remove damaged leaves, larvae; Drug prevention and treatment
	Culicidae		Mechanical damage to roots and rhizomes	Hill up soil after irrigation; Drug prevention and treatment
Microorganism	Phytophthora cactorum	Epidemic disease	Water-soaked rot on rootstock; Spot margins yellow-brown, turning black in severe cases	Control planting density, and shade; Drug prevention and treatment
	Pyricularia grisea	Leaf blast	Leaves wilted, plant dead; Brown mold on the surface of the spots	Control planting density, and shade; Enhanced ventilation, drainage; Increase in organic fertilizers; Drug prevention and treatment
	、P. variabilis Bus-saban
	Phoma sp.、Phyllosticta zingiberi Hori	Leaf spot	Irregular gray spot with a central dot; late stage the spots were flaky
	Colletotrichum sp	Anthrax	Initial leaves show irregular illness spots; Later leaf margins withered and yellowed, illness spots appearing as small black dots
	Colletotrichum gloeosporioides
	Fusarium graminearum	Blight	Leaf withered and yellowed; In severe cases, the stem base dries out and turns red	Avoiding flooding; Drug prevention and treatment
	Rhizoctonia solani	Seedling blight	Damage to seedlings, even collapse	Pull out the diseased plants; sprinkle lime; Drug prevention and treatment
	Fusarium、Glomerella	Wilt disease	Stem drying; wilting of leaves; musty, rotten, alcoholic flavor	Drug prevention and treatment
	Giborinia camelliae	Blossom rot	Flowers wilt and rot before they open	Avoiding flooding，; Control of planting density; Drug prevention and treatment
	Fusarium	Fruit rot	Fruits give off a lees-like odor before ripening and then rot away.
		Root rot	Seedling root rot; Yellowing and wilting of leaves in later stages	Soil disinfection; Prevention of underground pests and nematodes; Drug prevention and treatment
	Botryosphaeria dothidea	Maple dry rot	Dehydrated and withered; sprinkle of black dots; The cortex around the illness spots tends to crack	Drug prevention and treatment
	Fusarium	Basal rot	Stems and leaves turn yellow, wither and die.; The underground section is soft and corrupt	Increased ventilation and drainage; The soil is disinfected with lime; Drug prevention and treatment

4.1.2. Interaction with insects

4.1.2.1. Plants' own defenses

Plants and insects have coexisted for 350 million years. Although insects have shown mutual benefits by helping plants pollinate, predatory and defensive relationships have dominated (Medina-Serrano et al., 2025; Porto et al., 2025). Each year, 45 % of crop losses are due to insects; furthermore, large quantities of insecticides not only destroy crop survival, but also kill beneficial organisms, such as parasitic wasps (Sanaei & De Roode, 2025). Therefore, plant defense mechanisms should be reasonably activated. When plants are attacked by insects, the sensory system dominated by the jasmonic acid pathway activates physical and chemical defenses, including the induction of defense proteins (to poison insects) and the release of VOCs (Liu et al., 2025). The main roles of VOCs are as follows (Kalske et al., 2019): (i) attracting natural enemies of insects for predation, (ii) deterring insects, and (iii) transmitting them to distal parts of the plant or to nearby plants to give warnings of danger. All these measures are in response to insects. In addition to upregulating the synthesis of hormones, the plant itself also upregulates flavonoid synthesis genes, clearing reactive oxygen brought by insects, or enhancing defense at the cost of reduced photosynthesis. Tea plants upregulate phenylpropanoid and flavonoid biosynthetic genes to combat green tea leafhopper infestation (Zhao et al., 2020), and tomatoes accumulate phenolamides in leaves to enhance defense when infested with Tuta absoluta (Roumani et al., 2022). This could be a different defense mechanism caused by enzymes in the saliva of the insect, as well as the microorganisms and parasites it carries, the latter of which also contribute to yield reduction. Herbivore insects deploy salivary effectors to manipulate the defense of their host plants. A latest study shows that a microRNA (miR29-b) found in the saliva of the phloem-feeding whitefly (Bemisia tabaci) can transfer into the host plant phloem during feeding and fine-tune the defense response of tobacco (Nicotiana tabacum) plants (Han et al., 2025).

4.1.2.2. Defense of human intervention

The use of genome-editing technology to modify the genetic information of plants by adding, deleting, and modifying specific DNA sequences is a common tool in breeding; this technology is now being applied to insect defense and control (Smith, 2021). However, plants and insects co-evolve in their interactions, and as plant defenses are enhanced, insects reduce their susceptibility to plant resistance genes and evolve into stronger insects. Different populations of Ostrinia furnacalis, Chilo suppressalis, and Cnaphalocrocis medinalis exhibited wide variation in susceptibility to Cry1Ab or Cry1Ac toxins, suggesting that the genetic diversity necessary for evolution of insect resistance does exist (Li et al., 2020). This might be related to the epigenetic memory of insects, which forms genomic imprints during resistance (Lai & Wang, 2025). Therefore, editing sensitive genes to enhance plant resistance or knocking out insect olfactory receptor genes to hinder insect recognition by plants have also been proposed (Sai Reddy et al., 2022). AT is frequently infested by a variety of insect pests, such as Lymantria dispar, locusts, and Zygaenidae suppressalis (Table 2) (Yang et al., 2020), and large areas of AT are threatened every year. While the variety of AT increases the diversity of pests, the long-term use of single pesticides by farmers causes pests and microorganisms to produce antibodies, which makes pest control difficult (Li, 2020). Control measures for AT pests and diseases have been proposed based on three aspects: analysis of pest species, management by governmental departments, and establishment of pest control systems (Li, 2020). After the outbreak of a large-scale L. dispar disaster a few years ago, a series of control measures from the egg stage to the adult stage and their life patterns was proposed (He, 2023). This group found that the area infested by the old leaves of AT during the L. dispar outbreak was larger than that of the new leaves, which could be a measure of self-preservation by the plant. Although biological methods to identify the natural enemies of pests have also been attempted, the current biological control technology for AT is immature and ineffective (Li, 2020). Multiple insect-resistant transgenic AT are a “once-and-for-all” solution to these diverse insects. Until insects develop genomic imprints, other types of insects cannot evolve because the ecological balance of AT is disturbed. However, this process is lengthy. The main task was to determine the range of defense mechanisms and VOCs released by AT after various insect infestations and adjust the relationship between AT and insects.

4.1.3. Interaction with plants

4.1.3.1. VOCs as mediators of interaction

The plants mentioned here included three cases: distal or different parts of the same AT, different AT in the same population, and other plants in the vicinity of the AT. Microorganisms and insects do not entirely rely on VOCs to communicate with plants; they are in spatial contact with plants (Medina-Serrano et al., 2025; Zeng et al., 2025). By contrast, solidly growing plants seem to rely on the spatial transmission of VOC to “eavesdrop” on each other. In the current “natural forest understory” model for planting AT, the vegetation around AT is often cleared, leaving taller vegetation to create shade. Although this seems to create a high-quality environment for the AT, but to a certain extent reduces the AT and other vegetation communication strategy. Therefore, large-scale planting areas focus on different parts of the AT and communication between the AT. The consensus among growers is to control the planting density of AT.

4.1.3.2. Transmission and sensing of VOCs

VOCs cause changes in membrane potential, thereby generating electrical signals and cytotoxicity repair responses, altering downstream transcription and metabolism (Kessler et al., 2023); some GLVs are directly metabolized into defense compounds (Hu, Tian, et al., 2025). Scientists have made some achievements in VOCs production, release, delivery, and sensing, but it is still difficult to apply the role of plant–plant intercommunication in practice (Liu, Wang, et al., 2023). First, VOCs in the air are diluted or oxidized in actual production, and the concentrations are much smaller than those in the experiments. Second, VOCs transfer in the air is susceptible to meteorological factors, such as wind speed and direction and temperature (Brilli et al., 2019). This uncertainty is amplified at high altitudes, and may explain why VOCs released from AT have not received much research attention. (E)-Nerolidol, a volatile sesquiterpene, is also an important VOC signal that activates defense mechanisms in neighboring plants or uninjured tissues, a mechanism that is currently well-studied in tea tree. (E)-Nerolidol initiates defense mechanisms primarily by enhancing the expression of defense-related genes and increasing the levels of hormones, such as jasmonic acid. (E)-Nerolidol has also been found in AT. However, it has been established that its content is proportional to altitude (Li, Tian, et al., 2021), and the role (E)-nerolidol plays in the network of defense mechanisms in AT remains unknown. Theoretically, released VOCs are picked up more readily by conspecifics. Because the release and reception processes of VOCs need to pass through the cuticle, cell wall, cell membrane, and cytoplasm (Kessler et al., 2023) and be transported by carrier proteins, these are very similar in the same plant species, as from one “home” to another (Fig. 3). Crops are mostly cultivated in monoculture patterns, with similar types of VOCs released into the air at high concentrations. VOCs in such environments are readily received, and the defense mechanisms of adjacent crops are more likely to be activated, thereby providing marked environmental adaptability, while also containing VOCs that are inhibitory (Arimura & Uemura, 2025). Thus, theoretically, the VOCs released between neighboring AT may interact with each other. The second factor is shade, where the planting density of AT changes the shade. Concurrently, this change is captured by the photosensitive pigments, causing a series of changes in the molecular network, as shown in Section 4.3.2.

Fig. 3.

Release and reception of VOCs from AT. A, G: represent the internal environment of the receiving and releasing cells, respectively. B, F: represent the cell wall of the receiving and releasing cells, respectively. C, E: represent the cuticle of the receiving and releasing cells, respectively. D: spatial environment of VOCs transport between neighboring AT. Genes are present in chloroplasts, mitochondria, and nucleus of cell, and all three may generate VOCs. The generated VOCs is released through the exclusive transport proteins on the phospholipid bilayer, and then through the cell wall and the cuticle. Upon entering the air environment the VOCs may be oxidized, decomposed, or blown to other areas by natural winds, resulting in dilution of the VOCs. The VOCs, after being captured by the receiving neighboring AT, will be transported through the same cuticle, cell wall, and phospholipid bilayer and by the exclusive transporter protein as the releasing AT. This process may affect gene expression in chloroplasts, mitochondria and nuclei in two ways. One is the alteration of Ca²⁺ concentration, which leads to a change in membrane potential, thus generating an electrical signal that affects gene expression. Secondly, the captured VOCs may also directly affect gene expression. Ultimately, the purpose of “eavesdropping” is achieved, and the defense mechanism is activated in advance.

4.2. Molecular breeding of AT

4.2.1. The need for molecular breeding

AT growth is disturbed by abiotic stresses, such as droughts, frosts, and floods, and biotic stresses, such as pests and diseases, and is currently not being managed efficiently, given the lack of effective interventions, resulting in low fruiting and survival rates and serious yield losses (Yang et al., 2019; Yang et al., 2020). To cope with this present situation, it is necessary to improve farmers' awareness and theoretical knowledge of cultivation and management and screen and breed high-quality AT varieties. Currently, AT is still in the first stage of breeding and artificial selection, also known as domestication. Farmers domesticate wild AT, select the desired phenotype of their choice, treat the selected seeds, and sow them. This process usually ignores genetic complexity and crudely divides the different germplasms in an AT population into a whole, and some critical but poorly phenotypically expressed genes are likely to be lost in the process. Currently, the only economic value present in AT is the fruit; however, AT does not bear fruit until 3 years of cultivation, which is a long growth cycle. This also increases the breeding cycle for high-quality AT and therefore requires rapid breeding. In addition, as environmental problems, such as rising global temperatures, irregular rainfall scales, increasing biotic stresses, and more complex highland climates, will exacerbate in the future, improved breeding of AT will be an effective way to combat extreme stresses. Multiomics technology is an effective breeding aid. Molecular assisted breeding has also been used to improve soybean yield, latitude adaptation, and seed oil production (Bisht et al., 2023; Ma, Zhang, Meng, et al., 2020). In maize, the breeding time has been greatly reduced by rapid breeding, genome editing, and high-throughput phenotyping (HTP) to introduce tolerance genes and improve adaptation to stress (Farooqi et al., 2022). Recently, by combining metabolic, evolutionary, and spatiotemporal transcriptomics to construct the allicin metabolic pathway in Allium-specific a variety of bulb-expansion-promoting genes was identified (including cytoskeletal rearrangement), which provides a theoretical basis for the molecular breeding of Allium-specific genes (Hao et al., 2023). There are also many other crops, such as pepper and sesame, which have realized the cultivation, screening, and improvement of high-quality germplasm resources, markedly improving their adaptability to stressful environments. For AT breeding, the following steps were performed (Fig. 4).

Fig. 4.

Breeding process for AT. Firstly, AT genes were localized by GWAS for traits such as “fruit type”, “high yield”, “flowering time”, “disease resistance” and “volatile oil” and other traits and form a pan-genome. On this basis, genes controlling excellent traits in the close relatives such as Amomum paratsao-ko (A), Amomum koenigiit (B), Amomum villosum (C), and Alpinia katsumadai (D) were added to the pan-genome of AT to construct a super pan-genome. Phenotypes and genes are then linked by GWAS to localize trait-controlling genes. Finally, by means of CRISPR/Cas9 gene editing, high-quality AT that are resistant to insects and pathogenic microorganisms and have high and excellent production were produced.

4.2.2. Constructing a high-quality genome

The first step in the construction of a high-quality genome involves mapping the pan-genome of AT, followed by construction of a high-quality reference genome. The next steps involve linking genes to phenotypes and dissecting the genetic mechanisms underlying the complex phenotypes of AT. These are efforts are the cornerstone of breeding programs. For instance, quantitative trait locus (QTLs) that affect pericarp color and tannin content were identified in sorghum germplasm based on whole-genome sequencing (Zhang et al., 2023); By constructing high-quality reference genomes Researchers also found an overview of cell-type categorization and gene expression changes associated with spongy mesophyll cell expansion during onion bulb formation, thus indicating the functional roles of bulb formation genes (Hao et al., 2023). After a long period of domestication and cultivation, significant geographic differences in quantitative traits, such as fruiting rate, dry-fruit weight, soil elements, and seeds, emerged in AT (Li, 2021; Liu, Zhang, et al., 2023; Ma, Zhang, Zhu, et al., 2020). Much of this variation is related to structural variations (SVs) induced by the growing environment. This suggests that AT has a rich genetic variation resource that can provide a reference for molecular breeding. Currently, two genome assemblies have been created for AT, with before and after sizes of 2.08 GB and 2.70 GB, respectively; the true genome size of AT is bound to exceed these due to the diversity of genetic variation and the limitation of not having the complete coverage of the measured subjects (Li, Bai, et al., 2022; Oksuz et al., 2023). In another study, high-quality chromosome-level AT genomes identified genes involved in fruit type regulation and terpene biosynthesis, and multi-omics data revealed the mechanism of quality trait formation (Zhang et al., 2025). Wild crop relatives are frequently used in breeding programs, contributing to features such as crop adversity resistance (Kapazoglou et al., 2023) and quality (Saini et al., 2023). However, it is important to note that gene introgression in wild relatives may result in large-scale genetic recombination, which can be hampered by large SVs and is not conducive to the flexible selection of specific genes during hybridization (Tao et al., 2019). The chloroplast genomes of Amomum are highly conserved, and there are similarities in their medicinal constituents, especially in Amomum villosum and Alpinia katsumadai, which are aromatic herbs with AT (Ma, Wang, & Lu, 2021). Further, several species of Alpinia, which are the closest relatives of Amomum, share the same breeding system, meiotic system, and chromosomes as AT. Currently, cultivated AT can be categorized into two species, i.e., yellow-flowered and white-flowered (Amomum paratsao-ko), which are likely to share a common ancestor; however, there is no evidence of hybridization between the two. However, with the loss of low-frequency alleles during long-term domestication, these alleles have become distantly genotypically related (Li, Lu, et al., 2021; Liu, Zhang, et al., 2023). Nonetheless, compared with the other species, A. paratsao-ko is the closest to AT. The genetic diversity of A. paratsao-ko was higher than that of AT, and the sources of genetic variation were mainly between populations (Li, Lu, et al., 2021). Therefore, it is necessary to construct a super pan-genome to search for genes lost during the domestication and evolution of AT to provide more genetic resources, which may be the key to solving the breeding bottlenecks.

4.2.3. Linking genes and traits

The construction of a high-quality reference genome is a long and cumbersome process. Although the AT genome has been assembled (Sun et al., 2022), the annotated genes are not sufficient to be used as a selection tool for breeding, and no linkage has been made to related traits. In AT, the genetic differences between populations are minimal, as evidenced by the low coefficient of differentiation between populations; however, each population has high genetic uniformity and genetic variation with strong gene flow (Ma et al., 2022; Ma, Zhang, Meng, Wang, et al., 2021; Ma, Zhang, Meng, et al., 2020; Ma, Zhang, Zhu, et al., 2020). This may be due to the heterogeneous pollination methods used for AT, which block the spatial exchange between populations and expand the genetic advantage within the population. This is an important reason underlying phenotypic diversity, which is a significant factor in volatile oil differences in AT. Factors, such as geographic conditions, origin, and soil affect the content and type of volatile oils present in AT (Li, 2021; Li et al., 2023; Yang et al., 2022), thus making it possible to determine their origin by qualitative and quantitative analysis of volatile oils (Li, Tian, et al., 2021). Fresh weight, dry weight, and dry kernel weight of AT were positively correlated with altitude, whereas the contents of 1,8-cineole, trans-citral, (Li, Lu, et al., 2021) and the pungent compound were negatively correlated with altitude above 2000 m (Li, Zhang, et al., 2022); the latter has been shown to be associated with gene expression. Some of these traits may also involve the co-inheritance of genes, i.e., linkage disequilibrium. Fruit length has been found to affect the content of certain compounds (nine in total), including (E)-2-hexenal, octanal, and α-phellandrene, as well as the weight of fresh fruit and seed kernels (Li, Tian, et al., 2021; Xu et al., 2021). In addition, cob length exhibited a highly significant positive correlation with cob width, number of fruits per cob, total weight per cob, cob weight, and fruit weight per cob (Xu et al., 2021). Overall, AT traits are complex and diverse; however, their linkages to genes have not been evaluated.

GWASs can detect associations between single nucleotide polymorphisms (SNPs) and traits as well as the genomic regions of related genes, answering a range of questions from genes to phenotypes to evolution (Guo et al., 2025). In rice, DNA marker technologies, such as SNP identification, GWASs, and QTL localization, have been used to improve agronomic traits, with transgenic “golden rice” being the most typical representative of such improvements (Iqbal et al., 2021; Peng et al., 2020). Additionally, these efforts have led to the identification of numerous stress-fighting genes, proteins, and metabolites. Phenotypic maps of AT that will help bridge the gaps between phenotypes and genes are yet to be generated using GWASs. Molecular markers have been used in the genetic analysis of AT; they not only reflect the genetic variation in AT at the individual and interspecies levels, but also lay the foundation for the rapid detection of target trait genes.

4.2.4. Construction of transgenic systems

The final step in the molecular breeding of AT is selection of the desired genes (drought resistance, cold resistance, high yield, etc.) from a large number of genetic resources obtained to breed the desired AT germplasm AT via transgenic means (e.g., CRISPR/Cas9-based gene editing) (Hu, Zhang, et al., 2025). Selection for phenotypic and adversity resistance is also influenced by mRNAs, ncRNAs, proteins, and metabolites. While conducting in-depth studies, GWASs can be integrated with epigenome-wide association studies, transcriptome-wide association studies, proteome-wide association studies, and metabolite-wide association studies to obtain a comprehensive picture of a series of molecular networks that are active under conditions of environmental disturbance.

4.3. Deciphering the mechanisms underlying ST in AT

4.3.1. The influence of lighting on shade avoidance (SA) and ST strategies

Light is essential for the growth of all plants, and a high-level canopy allows for light gradients in the plant community, including those for different parts of the same plant and those at different growth periods. Plants growing under a canopy experience changes in the quality and quantity of light they receive. In such environments, plants develop both SA and ST strategies. The former helps plants adapt to shaded environments through hypocotyl elongation, branching reduction, and apical dominance, which are collectively referred to as SA syndromes (SASs) (Martinez-Garcia & Rodriguez-Concepcion, 2023). The latter is a more conservative strategy for long-lasting adaptation to the shade environment; it occurs via the adjustment of respiration rate and light utilization (Martinez-Garcia & Rodriguez-Concepcion, 2023). Both strategies rely on photosensitive pigment receptors to sense changes in red: far-red (R:FR) and blue light (BL) (Wang et al., 2020), and both exhibit increased specific leaf area (SLA) and photosystem II:I, as well as reduced chlorophyll a:b ratios (Wang & Wang, 2023). The ST of AT has been observed in the last century; however, there have only been a few studies on the molecular genetic mechanisms underlying ST. AT is a shade-tolerant plant and its growth and development are restricted under strong light, with leaf scorching, plant dwarfing, and failure to flower and fruit (Yang et al., 2019). This is the second factor that needs to be considered for planting density: the light intensity required for the normal development of AT ranges from 1000 to 10,000 lx (optimal, 4000–8000 lx), and the shade should be 60–70 % at a young age (Yang et al., 2019). This is in line with the prediction of the carbon gain hypothesis, i.e., the growth rate under high light conditions and survival under low light conditions are negatively correlated across species (Bison & Michaletz, 2024). Therefore, most farmers plant AT directly in the forest understory to create a natural shade environment.

4.3.2. Regulation of ST strategies by the bHLH family

It is not clear how plants select between SA and ST strategies in shaded environments, but most views support gene expression, especially within the chloroplasts. The upregulation of photosynthesis and chlorophyll biosynthesis genes to increase light utilization is an adaptation of Populus tomentosa seedlings to shade (An et al., 2022). Shade-tolerant rice varieties are also affected by the positive selection of chloroplast-related genes (Gao et al., 2019). In contrast, none of the six positively selected genes detected in the chloroplast genes of AT were related to photosynthesis (Ma, Zhang, Meng, Wang, et al., 2021); this may be due to the fact that the understory shade environment itself was positively selected or that the positive selection of photosynthesis-related genes occurred a long time ago. Currently, there is no systematic explanation for the ST mechanism in plants; in addition to the analysis of the expression of shade-tolerant genes, comprehensively deciphering the molecular mechanism whereby SA paves the way for ST is a current research strategy. In addition to having the same photoreceptor system and partially overlapping shade responses, SA and ST have opposing responses to low R:FR and low photosynthetically active radiation (PAR) (Morelli et al., 2021); further, in shade-tolerant plants, SA is strongly inhibited. The basic-helix-loop-helix (bHLH) is a central player here, and its interaction with photoreceptors activates hormonal pathways, photomorphology, and other shade-tolerant responses (Li, Wang, et al., 2025). Here, we discuss the phytochrome-interacting factors (PIFs), the most studied member of the bHLH family, to speculate on the possible mechanisms underlying ST (Fig. 5).

Fig. 5.

Molecular network of shade tolerance mechanisms.

4.3.2.1. The regulation of hormones by the bHLH family

PIF1, PIF3, PIF4, PIF5, and PIF7 positively regulate SA and promote hypocotyl elongation, with the latter three playing a central role, whereas PIF3-LIKE 1 and LONG HYPOCOTYL IN FAR-RED 1 form dimers to repress PIF (Wang et al., 2020). Low R:FR causes PIF-mediated changes in the expression of genes associated with the hypocotyls or petioles over a short period of time (Morelli et al., 2021). These genes are usually associated with the synthesis of hormones, such as abscisic acid, brassinosteroids, auxin, ethylene, and gibberellin (GA), and cell expansion (Saud et al., 2022). Both PIF4 and PIF5 bind directly to the promoters of auxin synthesis genes in response to shade under conditions of low R:FR or PAR (Ma & Li, 2019). A study shows that light-activated GmCRY1s increase the abundance of the bZIP transcription factors STF1 and STF2, which directly upregulates the expression of gene encoding GA₂ oxidases to deactivate GA₁ and repress soybean stem elongation (Lyu et al., 2021). DELLAs are able to interact with PIFs and prevent them from binding to their target genes and can induce the degradation of PIF proteins，and ultimately promote hypocotyl elongation (Han et al., 2024). GA can also guide soybean SAs in response to blue light reduction, under the regulation of the cryptochrome GmCRY1s (Lyu et al., 2021). This suggests that there is a response in GA to utilize SA strategies and then, a bias towards a shade-tolerant phenotype when the GA level declines.

4.3.2.2. The regulation of phytochrome (Phy) by the bHLH family

Phytochrome B (PhyB) triggers the phosphorylation of PIF and degrades it to lose its transcriptional capacity. However, under conditions of low R:FR, phyB is inactivated, and PIF is dephosphorylated, which in turn, restores its ability to bind to target genes (Han et al., 2024). In this scenario, PIF3, PIF4, and PIF5 degrade rapidly, and PIF7, which is the most stable, does not degrade rapidly and is dephosphorylated rapidly under shade conditions (Wang et al., 2020). PIF7 promotes the elongation of hypocotyls by interacting with the histone methyltransferases MRG1/MRG2 and promoting the modification of H3K4me3/H3K36me3 in downstream genes (Fang et al., 2025; Krahmer & Fankhauser, 2024). In contrast, PhyB enhances the modification of H3K27me3 and represses the transcription of growth-related genes (Kim et al., 2021). To a certain extent, phytochrome A (PhyA) accumulation inhibits the inactivation of PhyB, acting as an indirect SA repressor (Molina-Contreras et al., 2019); at the same time, PhyA can also directly act as a repressor of SA. These occurrences have already been demonstrated in crops such as rice and potatoes. Phy also affects the synthesis of plant hormones, thereby inhibiting SA (Lyu et al., 2021), which is usually associated with the regulation of PIF expression. Thus, both PhyA and PhyB inhibit SA, and this inhibition is negated when their activity decreases. This may be because when Phy activity decreases, the utilization of light decreases, and the plant adapts to the low-light environment under shade through phenotypic changes such as hypocotyl elongation, which is an SA strategy. In contrast, when the Phy activity is sufficient, plants can effectively utilize light energy under shade and adapt to a shaded environment without the need for SA, which represents a shade-tolerant strategy. Possibly preventing hypocotyl over-elongation under shade, which leads to collapse due to homeostatic disorders or excessive energy expenditure on elongation, low R:FR conditions also promote the accumulation of PIF antagonists, such as HYPOCOTYL ELONGATED 5, PIL1 and HFR1 (Han et al., 2024; Martinez-Garcia & Rodriguez-Concepcion, 2023).

4.3.3. Planting strategies based on SA and ST

In summary, SA and ST occur simultaneously and inhibit each other. They are in opposing dynamic regulation, in that one side will show dominance in a shaded environment and the other side will simply be suppressed and not disappear; Phy and PIF are in control of this balance (Fig. 5). Based on the current situation, natural forest understory planting of AT is not the best choice. This type of planting destroys natural forest resources, reduces species diversity, and interferes with ecological functions to a certain extent, causing serious soil erosion and markedly lowering the organic matter content in the soil. Other shade-tolerant plants, such as Tetrastigma hemsleyanum and Panax notoginseng, have been produced using artificial shade for farmland planting; this is convenient for cultivation management and helpful for the expansion of planting scale. AT plants are large, and a high planting density can easily cause shade in neighboring plants. Currently, AT plants cultivated in greenhouses are mostly used for scientific research and have not yet been used for large-scale practical production. The light signaling system, plant hormone signaling pathway, and other ST responses of AT remain unknown; future research should focus on deciphering the mechanism underlying ST in AT plants, the regulatory network of bHLH, and screening of ST-associated genes; such information will provide more theoretical support for the optimal and sustainable cultivation of AT in farmlands.

5. Plant genome databases

5.1. The current situation of the plant genome database

Currently, there are no genomic records of AT in the NCBI (https://www.ncbi.nlm.nih.gov/), Ensembl (http://plants.ensembl.org), or Phytozome (http://www.phytozome.net) databases; however, AT has been documented in previous literature. According to the current findings, AT possesses 24 chromosomes, with 95.4 % of the encoded proteins and GO functions, such as metabolic functions and catalytic activity, and KEGG metabolic pathways, such as metabolism and environmental information processing, being annotated (Li, Bai, et al., 2022; Ma et al., 2022; Sun et al., 2022). For AT, two issues must be urgently addressed: establishing a linkage between the genome and phenome, and identifying the specific expression of genes as well as downstream products at specific times and spaces, such as the expression of defense system-related genes and products during insect infestation. The former is necessary obtain high-quality germplasm resources for AT, while the latter is necessary to ensure the normal growth and development of AT. Databases can store large amounts of data and facilitate data sharing, which undoubtedly accelerates the progress of related research. Plant genome databases store a wealth of gene-related information such as gene sequences, gene structures, proteins, exons, and introns (Table 3). These databases provide researchers with a solid theoretical foundation and research background, providing additional data support for experimental studies. However, with the advancement of ML algorithms and the further exploration of genes, plant genome databases inevitably encounter challenges.

Table 3.

10 Plant genome databases and their intent to create, related information and websites.

Database	Intent	Content	Website
PlantGDB	Sequencing molecular sequences and integrating some bioinformatics tools that facilitate gene prediction and cross-species comparisons. Provides a basis for solving functional and evolutionary genomics problems	12 online analysis tools, 187 sequences of common plants, 29 plant genomes,	(https://goblinp.luddy.indiana.edu/)
PlantTFDB	Systematically authenticate and annotate TFS. As a useful resource for studying the function and evolution of TFs	165 species predicted 320,011 TFs (including 58 families)	(http://planttfdb.cbi.pku.edu.cn)
GreenPhylDB	facilitate comparative functional genomics in Oryza sativa and Arabidopsis thaliana genomes. Assign O. sativa and A. thaliana sequences to gene families using a semi-automatic clustering procedure and to create ‘orthologous’ groups using a phylogenomic approach.	46 species, 19 pangenomes, 27 reference genomes	(http://greenphyl.cirad.fr)
PLANEX	Functional characterization of genes, analysis of correlations between genes.understanding expression similarity and functional enrichment of input genes.	Functional annotation, comparison, and clustering of co-expressed genes in eight species	(http://planex.plantbioinformatics.org)
Phytozome	Provides the evolutionary history of each plant gene at the level of sequence, gene structure, gene family, and genome organization, along with sequence and functional annotations for a growing number of complete plant genomes	399 assembled and annotated genomes, from 173 Archaeplastida species, and contains the 54 Brachypodium distachyon lines from the BrachyPan pan-genome study,	(http:// www.phytozome.net)
BarleyBase	design for plants without a fully sequenced genome. Addresses the needs of plant biologists to analyze gene expression data and to place expression data in the context of functional genomics by using controlled genes and plant ontology to describe experimental conditions.	11 plant microarray platforms, including the 22 K Affymetrix Barley1 and Arabidopsis, 2119 hybridizations from 72 experiments	(www.barleybase.org)
Plant Genome Editing Database (PGED)	Providing information about plants generated using CRISPR/Cas9 technology. Driving a new revolution in gene editing and plant breeding	Transformation experiments, names of transformed plant varieties, DNA constructs used, including guide RNA sequences and primers used to characterize the resulting mutations, and details about the mutant plant lines	(http://plantcrispr.org)
DPVweb	Comprehensive source of high-quality sequence and taxonomic information for all fully sequenced viral, virus-like and satellite genes of plants, fungi and protozoa	rovides a central source of information about viruses, viroids and satellites of plants, fungi and protozoa, with some additional data on animal viruses and phages with RNA or ssDNA genomes.	(http://www.dpvweb.net/)
Ensembl Plants	Providing genome-scale information on sequenced plant genomes	Genome-wide comparison of 153 plant-related species, Variation databaser, Regulation database	(http://plants.ensembl.org)
LncPheDB	A functional database for genome-wide studies of lncRNA-regulated phenotypes	203,391 known and predicted lncRNA sequences in 9 species, 2000 phenotypic traits and 120,271 significant association loci manually screened from 421 high quality references	(https://lncphedb.com/)

5.2. Speed of database updates

First, the database is updated in “real time.” Using “plant genome” as the keyword, we entered the Web of Science database to observe the literature release information. Since the beginning of the 21st century, the number of publications on plant genomes has increased annually, and since 2020, tens of thousands of publications have been published annually. The increase in literature indicates that research results on plant genomes are also increasing, and researchers are using the database more frequently. Database updates not only refer to new content in the context of the original work, such as new gene sequences, but also to the emergence of new concepts, such as coding lncRNAs (CodLncRNAs). Traditionally, lncRNAs do not have a coding ability and are only involved in some basic biological processes, such as signaling, post-transcriptional modification, and gene expression regulation (Ferrer & Dimitrova, 2024). However, recent studies have found that certain lncRNAs contain open reading frames, which are recognized by the translation machinery to generate short peptides or proteins under specific spatial and temporal conditions (Jung et al., 2025). However, the emergence of CodLncRNAs has not only blurred the distinction between coding and non-coding RNAs, but also impacted the emergence of CodLncRNAs, which not only blurred the distinction between coding and non-coding RNAs, but also impacted the only channel for mRNAs to encode proteins; this, in turn, has led researchers to ponder whether other ncRNAs also possess coding ability. To address this, the CodLncScape database (https://cellknowledge.com.cn/codingLncRNA), which incorporates 353 entries of CodLncRNAs that were mainly associated with pan-cancer and spermatogenesis, was recently established (Liu et al., 2024).

5.3. Applications of ML algorithms

The second challenge is the introduction of ML algorithms with better performance. Parts of the database information must be uploaded by the researcher or manually screened against a large amount of literature; this makes updating the database information time-consuming, inefficient, and passive. Therefore, it is necessary to make full use of the prediction and classification performance of ML models to mine text. Examples of such include natural language processing, RF, support vector machine (SVM), logistic regression, bioformers, extremely randomized trees, and multilayer perceptrons (Hosseini & Leonenko, 2023; Liu et al., 2024).

5.4. Absence of epigenetic databases

The third challenge is the lack of relevant databases on plant epigenetics. The function of plant genomes depends on epigenetic modifications, such as DNA methylation and covalent modification of histones, under the regulation of which gene expression is often altered to change a range of physiological activities (Fang et al., 2025; Krahmer & Fankhauser, 2024). As important elements of the molecular mechanisms underlying plant memory, epigenetic modifications respond to environmental stimuli and preserve stress-related memories (Gallusci et al., 2023). However, these memories are often inherited after cell division and even over multiple generations. In addition, whether the regulation of genes by epigenetic modifications is associated with specific sites on the gene sequence remains uncertain, and descriptions of methylation are vaguely denoted, i.e., as “hypermethylated” or “hypomethylated.” For example, hypomethylation in specific pericentromeric regions can quantitatively confer disease resistance (Furci et al., 2019). When a sufficient number of entries are included in an epigenetic database, methylation sites can be predicted using ML models, revealing the relationship between gene expression and specific methylation sites.

6. Conclusions and future prospects

This paper summarizes the significance of multi-omics for the study of plant molecular mechanisms and using AT as an example, illustrates how multi-omics technology can solve the practical problems of non-model plants. Multi-omics technology based on high-throughput sequencing has become a trend in plant research, and other non-model plants like AT can accomplish their own “molecular revolution” in this era.

Rigorous data-processing methods provide strong support for the conclusions of multi-omics studies. At present, nonlinear functions are the focus and difficulty in the field of computers, and the nonlinear models obtained from nonlinear functions are much more complicated, and at the same time, can solve more problems, than linear models. In systems biology, DNA, RNA, proteins, and metabolites often do not have simple linear relationships, and the upstream and downstream relationships between them are not absolute, but can be converted to each other. In addition, the research field of systems biology continues to extend to new sources of proteins or peptides–CodLncRNAs. Further, scientists are shifting their focus on what they study in systems biology. The influence of a single factor on the molecular network has gradually shifted to the joint analysis of multiple factors, such as ncRNAs interacting with each other and TFs to regulate the expression of genes, which in turn, causes a series of downstream pathway changes. At the same time, scientists are trying to link the molecular structure and phenotype of plants, which is of great significance for the molecular breeding of plants. However, this requires not only a large amount of sample data but also an executable algorithm to run the obtained data and link genetic data to phenotypic features. These changes necessitate new data-processing methods, and in this context, nonlinear data DR is a valuable product.

Various physiological mechanisms in model plants, such as Arabidopsis thaliana and rice, are currently being widely explored. Non-model plants, which have long growth cycles and complex in vivo mechanisms, are often not the first choices for plant research. However, many non-model plants, such as AT, are associated with environment-related problems. In this review, we listed these problems and the mechanisms underlying natural communication, breeding, and ST in AT; these characteristics are also prevalent in other plants. As an example of ST mechanisms, it is difficult to grasp the specific degree of shade given to AT by high-rise stands or black screens in actual production; therefore, a specific method for ascertaining the degree of shade is also needed. We believe that we can refer to the plant's own shade perception system, i.e., the intensities of R:FR and BL, and thus link these values to evaluate shade as a growth-influencing parameter. Steep slopes and dense forests in the AT production area are important constraints associated with the implementation of management measures; if we want to achieve standardized AT cultivation in greenhouses in the future, shade is the first problem that needs to be solved. Genetic instability is one of the main difficulties associated with the molecular breeding of all plants. In this regard, we have a bold idea, i.e., considering gene editing at the epigenetic level. Epigenetic modifications are passed on to offspring when cells divide, and this contemporary and transgenerational inheritance is more stable in plants than in animals. Further, drought re-responses to a significantly higher expression level of the gene locus H3K4me3, which allows the inheritance of certain DNA methylation sites to the next generation through the RdDM pathway. Overall, our study lays a strong foundation for future research regarding the use of multi-omics techniques for plant-based research; the findings highlighted here can be translated to the broader research of the molecular mechanisms underlying various physiological process in non-model plants. Ultimately, our research will be useful in terms of enhancing germplasm resources, the efficiency of molecular breeding protocols, and ultimately, high-quality plant cultivation.

Author contribution

All authors contributed to the preparation of the manuscript. Dengke Fu is the lead author of the abstract, introduction, machine learning, systems biology, and conclusion. Yuanzhong Wang designed the outline of the review and is the lead author of the Plant Genome Database summary. Jinyu Zhang is the lead author of the chapter on advances in Amomum tsaoko multi-omics research and development potentials. All authors reviewed and revised the final manuscript.

CRediT authorship contribution statement

Dengke Fu: Writing – original draft, Software, Methodology, Data curation. Yuanzhong Wang: Writing – review & editing, Visualization, Conceptualization. Jinyu Zhang: Writing – review & editing, Investigation, Funding acquisition.

Consent for publication

Not applicable.

Ethics approval and consent to participate

Not applicable.

Funding

Our sincere thanks to Yunnan Province's major science and technology special plan project “Nujiang Amomum tsaoko Industry Science and Technology Innovation and Application Research” (202304BI090032–17); Innovative Research on the Seed Industry of Polygonatum kingianum, Amomum villosum and Amomum tsaoko (202502AS100009–03) and “Xing Dian Ying Cai” Talent Program (Grant number: XDYC-CYCX-2022-0027).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

Data will be made available on request.