1. From traditional breeding to intelligent design breeding: The era of big data in crop breeding
Crop genetic breeding research fundamentally aims to decipher the intrinsic relationships between genotypes and phenotypes to develop high-quality varieties that meet diverse human needs
[1]. Technological advancements have driven a revolutionary shift in crop breeding methods, transitioning from traditional phenotype-based approaches to intelligent design breeding that utilize advanced genomic technologies and big data analysis. The evolution of crop breeding can be divided into four milestone stages: domestication breeding, genetic breeding, molecular breeding, and emerging big data intelligent design breeding
[2]. Domestication breeding, which began approximately 10 000 years ago, marked a crucial point in the development of human agricultural civilization. Early agriculturists relied primarily on visual phenotypic assessments to selectively domesticate and cultivate wild plant varieties suitable for human needs
[3]. While this approach laid the foundation for modern agriculture, it lacked systematic theoretical guidance and was characterized by experience, uncertainty, and randomness. The late 19th and early 20th centuries saw crop breeding transform from an experience-driven practice to a theory-guided applied science with the emergence of Mendelian genetics and Darwinian evolution theory. This shift marked the beginning of the genetic breeding stage, significantly improving breeding efficiency. During this period, breeding practices began systematically integrating genetic theory
[4],
[5], field trial design, and statistical analysis techniques, gradually forming recordable and traceable breeding data. However, data recording in this stage was limited, with low levels of standardization and quantification. In the late 20th century, biotechnology advances drove innovations in molecular breeding theory and techniques. The widespread application of high-throughput sequencing and gene chip technologies, coupled with rapid developments in bioinformatics, has facilitated genome assembly for various crops, such as rice, corn, and wheat. These generated massive datasets and enabled the construction of comprehensive and precise genomic databases, opening new research frontiers for genetic analysis of important crop traits, precise functional gene mining, and in-depth interpretation of genetic evolution mechanisms
[6],
[7]. Since the beginning of the 21st century, the convergence of genomics, molecular biology, imaging, remote sensing informatics, big data science, and artificial intelligence (AI) has brought new opportunities for breeding technology innovation. A notable advancement in this stage is the rapid development of crop phenomics, which has overcome traditional manual phenotyping bottlenecks and generated multiscale, multidimensional high-throughput phenotypic datasets covering microscopic tissues to macroscopic organs, plants, and canopies. Simultaneously, advancements in environmental sensing technologies have enabled scientists to acquire and integrate high-precision, real-time environmental data under multiple habitat conditions, providing crucial support for precision breeding. In recent years, there has been a trend in the development of breeding data from traditional one-dimensional data to multiomics, high-dimensional data
[1]. Compared with single-omics databases, these comprehensive databases integrate multilevel biological information, establishing a more comprehensive and in-depth data foundation for accelerating crop genetic research and trait improvement (
Fig. 1).
Centuries of breeding have led to the accumulation of vast amounts of data, particularly with recent rapid advancements in genomics and phenomics technologies. These developments have propelled crop science and plant science into an era of omics research characterized by big data-driven approaches and major discoveries
[8]. Currently, high-throughput environmental sensors and advanced crop phenotyping equipment generate large volumes of semistructured and unstructured data, such as high-resolution images, precise point cloud data, and multidimensional spectral information
[9]. Moreover, massive crop genomic and multiomics datasets typically feature large scales, high dimensionality, significant noise, and strong heterogeneity. These characteristics make it challenging for traditional data analysis and expression methods to achieve efficient and accurate parsing of complex, multidimensional, and heterogeneous information. In this context, Professor Edwards Buckler, a corn genetics breeder at Cornell University and a member of the US National Academy of Sciences, proposed the concept of “Breeding 4.0”—intelligent design breeding—in 2018
[1]. Intelligently designed breeding represents a stage that integrates biotechnology with big data and AI technologies. It aims to achieve efficient, personalized breeding of new crop varieties, driving a transformative shift in breeding from “scientific” to “intelligent” approaches
[10],
[11].
The artificial selection era, from initial experience, uncertainty, and randomness, gradually formed recordable and traceable breeding data. Data recording in this stage was limited, with low levels of standardization and quantification. In the molecular breeding era, biotechnology advances drive innovations in molecular breeding theory and techniques. The widespread application of high-throughput sequencing and gene chip technologies, coupled with rapid developments in bioinformatics, has facilitated genome assembly for various crops, generated massive datasets and the construction of comprehensive and precise genomic databases. In the smart breeding era, the convergence of genomics, molecular biology, imaging, remote sensing informatics, big data science, and AI has brought new opportunities for breeding accumulation. Breeding data have shifted from traditional one-dimensional data to multiomics, high-dimensional data.
2. AI-driven crop breeding engineering: Integrating AI and biological big data
The advent of AI and the big data era has demonstrated unique advantages in complex object characterization, multimodal fusion, gene mining, and phenotype prediction. Crop breeding, as a key field utilizing big data and modern genomics, both require AI and provide an ideal application scenario
[12].
2.1. High-throughput phenotyping overcomes bottlenecks in trait acquisition for breeding
Despite the advances in sequencing technologies to decode the genetic variations of crops and their significant impact on breeding efficiency over the past few decades, accessing crop growth status for high-throughput phenotypic data acquisition has been challenging. Limitations in traditional manual detection of agronomic traits for large-scale germplasm have become increasingly apparent. Crop phenomics has emerged as a crucial solution to this bottleneck. Through interdisciplinary collaboration involving biology, mechanics, graphics, and computational science, a new generation of phenotyping equipment systems based on novel physical, chemical, and biological sensors, combined with AI and Internet of Things (IoT) technologies, is providing technical support for massive breeding trial data acquisition. This breakthrough has shattered the centuries-old “ruler and scale” approach to crop trait measurement, enabling high-throughput, automated, and minimally staffed phenotypic data collection.
Crop phenomics has significantly improved the efficiency, diversity, and precision of trait acquisition for breeding (
Fig. 2(a)). High-resolution unmanned aerial vehicle (UAV) photography has found applications in the identification of agronomic traits across several crops, such as plant height, biomass, the vegetation index, ear identification and senescence quantification
[13],
[14],
[15]. Phenotyping to increase resilience to biological and abiotic stresses is critical for sustaining genetic gains in crop improvement programs. For example, UAV or unmanned vehicle phenotype platforms can be used to obtain the time of day of flowering, yield composition, and photosynthetic efficiency with high throughput. These physiological traits and processes are critical for understanding crop responses to drought and heat stress during reproductive and grain-filling periods
[16]. Moreover, a high-throughput crop phenotyping platform combined with light detection and ranging (LiDAR), hyperspectral, micro- computerized tomography (CT), and red green blue (RGB) multioptical imaging technologies can be used for continuous nondestructive testing of crop phenotypes under salt stress, drought and other adverse environments, and many image traits can be obtained. This method overcomes the shortcomings of traditional methods for screening traits related to these stresses, such as being slow, laborious, destructive, and having low precision
[17],
[18]. Emerging microphenotyping technologies, such as laser ablation tomography (LAT) and CT, combined with multiomics studies provide insights for the genetic improvement of crop yield and stress resistance
[19]. With the precise identification of microphenotypic traits, an increasing number of stress resistance genes, such as those related to metaxylem vessel number (
KNAT7 and
ZmTIP1), stele area (
OsNACs), cortical cell (
bHLH121), and stem xylem and phloem traits (
NAC91), are being identified
[20],
[21],
[22]. These advancements will further promote intelligent and precise crop breeding, providing powerful technical support to address global agricultural challenges
[9],
[23].
2.2. Multiomics databases and management systems for multidimensional crop omics big data
The rapid development of genomics, transcriptomics, proteomics, and metabolomics analysis equipment, along with high-throughput phenomics acquisition and analysis instruments and environmental information collection technologies, has led to an unprecedented accumulation of crop biological data at multiple omics and high-dimensional levels. In recent years, major crop multiomics databases and management systems have been successively constructed and released, marking the entry of crop science research into a new data-driven era. In 2020, a research team from Huazhong Agricultural University integrated multiomics data from the same maize population, including genomics, transcriptomics, phenomics, metabolomics, epigenomics, genetic variation, and genetic mapping results, to construct the ZEAMAP database. This database enables cloud-based integration, rapid retrieval, and intelligent analysis of maize multiomics data, providing a foundation for the design of molecular breeding
[24]. The team subsequently published the first multi-omics data integration network map for maize and successfully predicted a batch of important functional genes via machine learning methods, including molecular regulatory pathways controlling crucial traits such as maize flowering time
[25]. In 2022, the Beijing Academy of Agriculture and Forestry Sciences used advanced phenomics technologies to acquire phenotypic data from over 1000 germplasm resources worldwide. Combined with resequencing data, they constructed a comprehensive lettuce database, LettuceGDB, which provides rich digital resources for lettuce breeding
[26]. In 2024, Guangzhou University and other institutions integrated data from six omics fields, including soybean genomics, transcriptomics, variomics, epigenetics, and phenomics, to construct SoyMD, which is currently the most systematic and comprehensive soybean multiomics database
[27]. Additionally, in 2024, comprehensive multiomics databases for cucumber (Cucumber-DB)
[28] and foxtail millet (
Setaria-DB)
[29] were released, providing important data resources and analysis platforms for crop genetic breeding research (
Table 1 [24],
[25],
[26],
[27],
[28],
[29],
[30],
[31],
[32],
[33],
[34],
[35],
[36],
[37]).
Currently, multiomics big data databases and management systems for major crops, such as rice, wheat, maize, and soybean, have been established. Compared with traditional genomic databases, these multiomics big data databases provide more comprehensive and rich genetic variation information, as well as their associations with multiple omics data. Additionally, these databases offer more user-friendly interfaces with enhanced functionality. The systematic integration of multiomics information can significantly improve researchers’ efficiency in mining candidate variations and genes, laying a solid foundation for accelerating crop genetic research and improving superior agronomic traits.
2.3. AI-based integrated multi-omics analysis: A key to accurately decoding the functional genome of crops
Owing to the introduction of multiomics and spatiotemporal omics information, modern breeding big data present several specific and important attributes, referred to as “the 9 Vs” (volume, velocity, variety, veracity, variability, validity, visibility, value, and vexing), requiring the use of AI technology for data analysis, processing, mining, and prediction
[38]. As AI technologies advance and multiomics data accumulate, scientists are seeking new methods to decipher complex crop traits. These traits, such as crop height, yield, disease resistance, or stress tolerance, are typically regulated by multiple genes forming intricate genetic regulatory networks. The vast number of genes and variables involved in these networks has long challenged scientists in identifying, confirming, and elucidating their interactions. Although classical genetic approaches have cloned and analyzed a number of important functional genes, progress remains slow. Remarkably, even nearly 20 years after the proposal of functional genomics, fewer than 10% of the total genes in rice and maize have been cloned and functionally characterized
[18]. The advent of AI and the big data era has demonstrated unique advantages in complex object characterization, multimodal fusion, gene mining, and network analysis. This has made it possible to systematically interpret biological complexity at multiple omics levels
[9],
[39]. In 2023, a research team from Huazhong Agricultural University achieved a major breakthrough in this field by successfully constructing a first-generation multiomics integrated network map for maize. This map encompasses 2 million network relationships across multiple genetic hierarchies, including genomics, transcriptomics, the translatome, and the protein interactome. Using advanced machine learning algorithms, the team successfully predicted a batch of important functional genes and identified molecular regulatory pathways controlling crucial traits such as maize flowering time. Notably, 62 out of the 63 known cloned genes regulating maize kernel development were accurately located on this integrated map, demonstrating its reliability and accuracy. Furthermore, the team successfully predicted and validated a pentatricopeptide repeat (PPR) protein, confirming its significant impact on maize kernel shrinkage. By deeply mining this integrated network map, researchers have predicted 2 651 candidate flowering time genes, divided them into eight subnetwork pathways based on functional similarity, and functionally validated 20 of these genes. This study provides a new means for systematically analyzing gene functions and the genetic mechanisms of trait variation, to some extent revolutionizing the paradigm of classical genetic research
[25]. Moreover, studies have validated the superiority of multitrait models based on deep learning or transfer learning over single-trait prediction models via statistical analysis
[40],
[41],
[42].
AI-driven multi-omics analysis not only accelerates the process of gene function discovery but also provides powerful tools for constructing precise regulatory network models, truly becoming the “key” to unlocking the genetic code of crops (
Fig. 2(a)).
2.4. AI-powered breeding software tools based on biological big data accelerate crop improvement
The rapid development of AI technologies has provided new tools for modern crop breeding iterations, demonstrating enormous application potential. By combining rich crop big data with AI, researchers can use computers to perform precise calculations and model genomic, phenotypic, and other big data, achieving efficient simulated breeding and accelerating crop genetic improvement. Recent advances in genomics and phenomics have made it possible to collect unprecedented genetic and phenotypic information, driving the transition from traditional experience-based and conventional breeding to intelligent genomic selection. By integrating genotype, phenotype, and environmental data to construct suitable big data models, breeders can accurately predict the genetic potential of future phenotypic individuals and optimize hybridization schemes accordingly
[38]. This approach not only significantly shortens breeding cycles but also markedly improves breeding selection accuracy
[43]. To accelerate breeding decisions, research teams both domestically and internationally have successively released intelligent design breeding software tools and platforms based on crop big data and AI technologies. These platforms integrate machine learning, deep learning, and high-performance cloud computing technologies, achieving full-process intelligence from information collection, processing, machine learning, and deep learning to model establishment, deployment, and practical application (
Table 2). Notably, genomic selection breeding technology has entered the industrialization stage or preindustrialization preparation phase in developed countries such as the United States, showing tremendous potential. Researchers continuously innovate by integrating multiomics data into genomic selection models and using AI models to accurately predict relevant genetic loci, providing reliable targets for precise crop design. For example, Monsanto, a global seed industry giant, has taken the lead in collecting phenotypic, genotypic, and environmental data at tens of thousands of test sites across hundreds of global experimental stations. Based on breeding trial big data and molecular marker data, they utilized AI technology to create breeding prediction algorithms driven by multidimensional big data of genotype–phenotype–environment interactions. This approach allows them to screen the most promising combinations for field trials, greatly shortening their breeding cycles and improving their breeding efficiency. Furthermore, Monsanto employs machine learning and predictive modeling techniques to tailor digital solutions for specific planting environments and cultivation measures, fully demonstrating the broad application prospects of AI technology in modern agriculture. Recently, intelligent breeding platforms based on biological big data and AI technology have also been released in China. For example, the Institute of Crop Sciences, Chinese Academy of Agricultural Sciences and Alibaba DAMO Academy jointly developed a full-process smart breeding platform, achieving the integration of breeding data management and analysis, computational acceleration, and AI prediction of parental lines and superior varieties
[44]. Research results show that this platform accelerates variant site calculations in gene sequencing data 110-fold and population genetics analysis more than 1000-fold. The accuracy of genomic selection algorithms has improved by an average of more than 15%, laying a solid foundation for the future precision and efficiency of breeding work.
The deep integration of biological big data and AI technologies has made it possible to generate “ideal phenotypes” through massive computations and combinations. This ultimately allows for the precise design of optimal combinations of natural variations and their realization through the most efficient hybridization methods, achieving a new breeding model that is digital, customized, and intelligent
[10],
[39],
[45] (
Fig. 2(a)).
2.5. Intelligent-driven crop engineering: Large-scale gene identification, efficient gene manipulation, engineered variety design, and systematized biobreeding
The era of crop Breeding 4.0 is rapidly transitioning from theory to practice, evolving toward digital, informationized, and intelligent design-driven breeding. Leveraging new physical, chemical, and biological sensors in combination with AI and IoT technologies has enabled high-throughput acquisition and intelligent analysis of crop phenotypes. The utilization of next-generation AI technologies for integrated multiomics research has led to the identification of an increasing number of functional genes and the elucidation of gene expression regulatory pathways, while predictions of plant phenotypes and protein structures have become more precise. Newly developed genome editing technologies allow for accurate modification of genomic sequences, enabling the mutation or editing of virtually any gene of interest, thus realizing engineered design breeding. In the future, the deep integration of biotechnology and information technology (BT + IT) will drive intelligent crop design breeding with two key characteristics: intelligent hybrid breeding based on big data and breeding models and intelligent biological breeding utilizing AI and synthetic evolution technologies. Ultimately, future breeding will evolve into intelligent-driven crop engineering, achieving scaled-up gene mining, enhanced efficiency of gene manipulation, engineering of variety design, and systematization of biological breeding, thus establishing a comprehensive system for intelligent crop improvement
[45] (
Fig. 2(b)).
3. Current status and analysis of China’s seed industry technology development
3.1. Analysis of the current state of China’s seed industry technology development
3.1.1. Rapid progress in the precise identification of China’s crop germplasm resources
Germplasm resources are the “chips” of agriculture, and high-quality germplasm resources are crucial for breeding superior varieties. China’s work on crop germplasm resources dates back to the 1950s. In 1986, the establishment of the National Crop Germplasm Bank marked the formal organizat ion and management of germplasm resources. By the end of 2002, China had established a crop germplasm resource preservation system centered on long-term storage, supported by duplicate banks, medium-term storage, germplasm nurseries, and integration with the national gene bank
[46],
[47]. According to the
Catalog of Available Agricultural Crop Germplasm Resources published in 2023, National Crop Germplasm Bank has strategically preserved over 530 000 germplasm resources, ranking second globally after the United States. In 2021, the Ministry of Agriculture and Rural Affairs of the People's Republic of China launched a comprehensive initiative for the precise identification of crop germplasm resources. As of February 2023, genotype and phenotype identification of the first batch of 76 000 resources has been completed. This work focused on important target traits, such as high yield, quality, environmental friendliness, disease and pest resistance, salt-alkali tolerance, and stress resistance. Through phenotypic (multienvironment identification over three consecutive years) and genotypic (resequencing and single-nucleotide polymorphism array) identification, a number of superior germplasms and genes with high yield, nutrient efficiency, disease and pest resistance, and stress tolerance have been discovered and applied to breeding innovation.
3.1.2. Digital transformation of crop breeding
Crop breeding is undergoing a profound digital transformation, evolving from traditional hybrid breeding to molecular breeding and then to smart breeding. In the hybrid breeding stage, China has significant advantages in terms of breeding technologies for hybrid rice, wheat, and some unique varieties. Using traditional techniques such as heterosis, distant hybridization, plant chromosome hybridization, and mutation, new varieties of cruciferous vegetables and grain and oil plants have been successfully bred, with some varieties, such as super rice, high-yield hybrid corn, and high-quality special wheat, achieving industrial application
[45],
[47]. These technologies rely mainly on the observation and selection of phenotypic traits, which are characterized by significant uncertainty and long cycles. With the rapid development of molecular biology and genomics, breeding technology has entered the molecular breeding stage. Over the past decade, Chinese scientists have identified and utilized key genes controlling important traits, such as yield, plant type, disease resistance, and drought resistance, through quantitative trait locus (QTL) mapping, gene cloning, and functional verification. They have revealed molecular genetic regulatory mechanisms and provided a batch of gene resources with independent intellectual property rights. Additionally, they have created a crop genome single-base editing technology system, improved technologies such as maize haploid breeding and intelligent male sterility in major crops, and they have established a complete transgenic breeding technology system, significantly enhancing the genetic improvement efficiency of major food crops such as rice, wheat, corn, and soybeans
[47].
Currently, breeding technology is advancing toward the smart breeding stage, characterized by cutting-edge technologies such as gene editing (e.g., clustered regularly interspaced short palindromic repeats and CRISPR-associated protein 9 (CRISPR/Cas9)), genomic selection, and AI-designed breeding. Smart breeding utilizes multiomics data, including high-throughput genome sequencing, phenomics, and metabolomics data, to predict and select optimal breeding combinations through big data analysis and machine learning algorithms. Although modern smart breeding technology is still in its early stages, its core lies in the deep integration of biological data and information technology, aiming to build an efficient, precise, and intelligent seed industry innovation system. Despite the emergence of large amounts of biological data, big data analysis, informatization, and software system development are still being continuously improved, and the seed industry innovation system has not yet fully developed. The establishment of this system will greatly shorten breeding cycles, improve breeding efficiency, and provide strong support for addressing global climate change and food security challenges.
3.2. Analysis of the gap between China’s seed industry technology development and global frontiers
The global seed industry is experiencing rapid technological advancements, characterized by increasing precision, efficiency, and intelligence in breeding processes. This section provides a comprehensive analysis of the current state of China’s seed industry in comparison with that of international leaders, with a focus on key aspects such as scientific innovation and core technologies (
Fig. 3(a)), intelligent breeding systems (
Fig. 3(b)), germplasm resource utilization (
Fig. 3(c)), and market competitiveness (
Fig. 3(d)).
3.2.1. Scientific and technological innovation capability and core technologies
A significant disparity exists between China and leading international countries in terms of the scientific and technological innovation of the seed industry. The United States, for example, maintains a dominant position in biological breeding, accounting for 80% of global core patents and producing a substantially greater number of highly cited papers. Although China has made strides in overall publication output, it lags in high-impact research and core patent ownership. As of early 2021, China had published 403 highly cited research articles in molecular breeding, approximately half the number produced by the United States
[48],
[49]. With respect to core breeding technologies, China predominantly occupies a “following” or “parallel” position. China largely relies on foreign technical principles and methodologies in critical areas, such as genome sequencing and gene editing
[47],
[50]. While China has recently increased investments in emerging fields such as crop phenomics, it still trails behind international frontiers in terms of original breakthroughs and industrial applications. Developed nations, particularly the United States, have successfully transitioned technologies such as whole-genome selection and AI-driven design breeding into industrial application phases, whereas China remains in the experimental research and development stage
[47].
3.2.2. Intelligent breeding systems and big data applications
A considerable gap exists between China and leading international companies in the development and implementation of intelligent breeding systems and big data applications. Over the past 20 years, major European and American seed corporations, including Corteva and Syngenta, have accomplished large-scale breeding data accumulation and digital transformation of breeding processes. These companies have established commercialized intelligent breeding systems through strategic collaborations with specialized genomic big data firms (e.g., NRGene) and plant phenotyping companies (e.g., LemnaTec). This has enabled data-driven intelligent breeding decisions, significantly reducing breeding cycles and costs (NRGene, 2019). In contrast, China presents notable deficiencies in the development of big data management systems, data analysis platforms, and core algorithms. The country has yet to develop key breeding decision models with independent intellectual property rights, such as genotype-to-phenotype prediction and whole-genome selection models. These limitations significantly impede China’s progress in precision and efficient breeding practices.
3.2.3. Utilization of germplasm resources and precise identification
While China possesses a rich quantity of germplasm resources, there is a discernible gap with international standards in terms of high-quality resource proportions and utilization efficiency. The United States maintains a germplasm collection of 720 000 accessions, with 80% being foreign resources and an effective utilization rate exceeding 10%. In comparison, China’s germplasm resources exceed 560 000 accessions, but the proportion of high-quality foreign resources is less than 25%, and only a small fraction has undergone comprehensive, precise identification
[47]. China also lags significantly in terms of precise identification technologies for germplasm resources. For example, in maize germplasm resources, only approximately 10% have undergone comprehensive characterization of key agronomic traits
[12],
[39]. In contrast, countries such as the United States have implemented more extensive and in-depth applications of technologies such as high-throughput phenotyping and marker-assisted selection.
3.2.4. Industrial structure and market competitiveness
The structure of China’s seed industry remains fragmented and relatively weak compared with that of the international market. The global seed market is characterized by high concentration, dominated by a small number of large multinational corporations. For example, the US company Corteva commanded 17% of the global seed market share in 2020 and owned 27.5% of global molecular breeding patents in the maize sector. In contrast, China has over 7 000 crop seed companies, which are generally small in scale and lack competitiveness. The combined annual sales of the top 10 seed companies in China amount to only approximately 10 billion CNY (approximately 1.5 billion USD). More critically, 80% of Chinese seed companies do not possess new plant variety rights, and only 2.3% of companies hold more than ten new variety rights. This industrial structure significantly constrains the competitiveness and global influence of Chinese seed companies.
In conclusion, while significant advancements have been made in recent years, China’s seed industry still lags behind internationally advanced levels in several critical areas. These gaps are particularly pronounced in areas such as scientific and technological innovation capabilities, the mastery of core technologies, intelligent breeding systems and big data applications, the utilization and precise identification of germplasm resources, and industrial structure and market competitiveness. Developed countries and regions, such as the United States and the European Union, maintain leadership positions in policy support, funding allocation, technological innovation, and industrial application.
3.3. Analysis of future technologies widely applied by breeders and local breeding units
In recent years, BT and IT have been deeply integrated into crop breeding
[50]. In the future, the use of cutting-edge breeding technologies such as high-throughput phenotype technology, multiomics analysis technology based on big data, and agricultural large model technology will accelerate China’s transition from traditional breeding to intelligent design breeding stages and cultivate new crop varieties faster, better and more efficiently. These three technologies are also key directions that can be widely applied by breeders and local breeding units in breeding practice.
(1) High-throughput and intelligent crop phenotype acquisition has long been a “bottleneck problem”. Owing to the complexity of phenotype identification and its high sensitivity to the environment, large-scale germplasm screening and genetic improvement efficiency are greatly limited. The latest high-throughput phenotyping technology can be used to measure the internal and external states of crops, such as their morphological structure, physiological function, and environmental response, and to carry out real-time monitoring and accurate identification of crop phenotype data throughout the entire growth period
[51]. This enables breeding to move from “mass selection” to “selection,” promoting traditional breeding toward an intelligent design breeding stage that is “perceptible, quantifiable, computable, controllable, and predictable.”
(2) Research on crop gene identification in China still has a situation of “three more and three less”; that is, there are more predicted genes in crops and fewer genes for localization and cloning, more genes have been discovered in rice than in other crops, more genes have been identified, and few genes with important application value have been identified
[52]. By integrating multiomics big data and environmental variable big data through the combination of multiomics, systems biology, and synthetic biology, we aim to elucidate the intrinsic connections and interaction mechanisms of the gene–phenotype–environment system and comprehensively analyze the regulatory network and environmental adaptability molecular basis formed by important traits such as yield, disease resistance, and stress resistance of crops such as maize and wheat under complex environmental conditions
[53]. Finally, through the development of design breeding technology with intelligent decision support, breakthrough crop varieties with synergistic improvements in multiple traits can be cultivated faster and more efficiently.
(3) Given that data from laboratory research and breeding production have always been disconnected from each other, increasing the application of data science and information technology in breeding research and promoting the role of agricultural big model technology in crop breeding production decision-making, management, and operations are crucial. By constructing complex system models with billions of parameters and utilizing multisource big data and breeding expert knowledge in the breeding field for vertical domain training, it is possible to achieve complex scenario environments such as breeding production control, climate change, and market fluctuations; optimize crop planting, soil management, water resource utilization, and pest control; and promote the transformation of breeding toward data-driven and intelligent approaches.
4. Development recommendations for next-generation AI and big data intelligent design breeding in China
The global seed industry has entered a new phase of intensive innovation and rapid industrial transformation. Big data and AI technologies have become the digital foundation for building a strong seed industry
[54]. Crop science is increasingly integrated with data science and computational science, driving breeding toward intelligent development
[55],
[56]. In this context, the deep fusion of AI, information technology with conventional breeding and biotechnology is expected to address bottleneck issues such as precise phenotypic identification of crop germplasm resources and multiomics big data integration analysis models. This integration will increase the utilization of germplasm resources and accelerate the construction of a large-scale, systematic, and collaborative intelligent design breeding innovation system
[51]. These advancements are crucial for cultivating strategic new varieties, achieving breakthroughs in seed industry technology, and promoting the rapid development of China’s seed industry
[47],
[48],
[49],
[56].
4.1. Development approach and objectives
To address key issues such as high-throughput and precise phenotypic identification of crops, key gene mining, and molecular pyramiding breeding technology, which are currently limited by external factors, we propose focusing on developing automated intelligent crop phenotype acquisition technology for multiple environments, advancing multisource heterogeneous information fusion mechanisms and analysis techniques, and creating omics big data analysis and intelligent design breeding algorithm model technologies. These efforts, which are based on China’s germplasm resource bank, aim to establish large breeding models for typical crops, efficiently analyze genetic regulatory networks of important agronomic traits, and construct a precision breeding decision system driven by multidimensional big data on genotype–phenotype–environment interactions. This will help overcome bottlenecks in China’s biological seed industry and achieve a turnaround in the seed sector.
By 2040, we aim to develop frontier core technologies in data acquisition, including automatic control, intelligent path planning, efficient remote transmission, and data self-inspection, addressing the multiscale, multidimensional, multimodal, and multienvironment characteristics of phenotypic data. We will create integrated sensor arrays and crop phenotype imaging units to break international monopolies, advancing China’s next-generation phenotyping platform technology from following to leading globally
[57]. In phenomics big data intelligent analysis, we will achieve breakthroughs in multimodal data fusion and intelligent analysis of crop morphological–structural–physiological phenotypic information, constructing crop multidimensional, multiscale, and multimodal phenotypic cognitive maps. By combining mechanism models and machine learning, we will achieve intelligent phenotypic cognition and form an autonomous crop phenotype analysis technology system. By utilizing deep learning, data fusion, and hybrid intelligence technologies, we will develop a series of proprietary model algorithms for multiomics big data integration, mining, functional analysis, and breeding decisions
[50]. These advancements will enable efficient and precise analysis of genetic regulatory networks for important agronomic traits in major crops, forming a crop breeding system characterized by large-scale gene mining, efficient gene manipulation, engineered variety design, and systematic biological breeding.
4.2. Development recommendations
4.2.1. Multidisciplinary integration
New physical, chemical, biological and physiological sensors have accelerated the construction of large-scale scientific facilities and projects for plant phenomics, such as the Shennong large-scale facility, through the multidisciplinary intersection of biology, mechanics, and graphic imaging
[51]. These facilities have the ability to collect and analyze crop biological big data, serving as “accelerators” for intelligent and efficient crop variety selection and breeding. By 2040, we envision the comprehensive application of automatic control, robotics, and next-generation communication technologies. By integrating various new sensors, we aim to create high-throughput acquisition systems and equipment for crop phenotypic data across multiple scales (microscopic tissue/organ, mesoscopic individual plant, and macroscopic population levels), environments, and modalities throughout the entire growth period
[57]. Deep learning, edge computing, and hybrid intelligence will be paramount in developing algorithms for crop phenotype multimodal data fusion analysis and cooperative optimization of computing systems
[58]. These advancements will enable real-time and rapid analysis of crop phenotypic data throughout the entire life cycle. Additionally, it is crucial to build a comprehensive and multifaceted phenotype cognition map for crops, encompassing multiple dimensions, scales, and modalities
[59]. By combining mechanism models with machine learning techniques, intelligent cognition of phenotypes can be achieved
[60]. Finally, it will achieve a major transformation of “replacing manpower with machines” and “replacing imports with independent technology.” However, realizing this vision faces key challenges, including real-time integration of multiscale, multimodal data and ensuring sensor stability in complex field environments. Breakthrough areas include the development of high-precision, low-cost sensors; edge computing algorithms for massive real-time data processing; and the integration of plant physiology with computer science
[19],
[61],
[62]. Overcoming these hurdles is crucial for achieving the transformation of “replacing manpower with machines” and “replacing imports with independent technology.”
4.2.2. Data-driven precision breeding
With the core design concept of “data-driven, precise analysis, and intelligent decision-making,” we propose developing intelligent design breeding core algorithm models driven by crop multiomics big data
[54]. This will involve constructing breeding decision model libraries with independent intellectual property rights for genotype–phenotype prediction and whole-genome selection
[63]. These advancements will break the long-standing dominance of foreign products in the breeding software market, fill the gap in China’s lack of breeding software and decision models with independent intellectual property rights, and ensure autonomous control of China’s crop breeding data resources. By 2040, we aim to develop algorithms for multiomics big data integration, mining, and analysis via deep learning, data fusion, and hybrid intelligence technologies
[50]. This will enable in-depth mining and analysis of genetic regulatory networks for important agronomic traits, such as crop plant type, yield, and resistance. We will develop original intelligent genome design and whole-genome selection prediction models and algorithms, constructing breeding decision models driven by multidimensional big data of “genotype–phenotype–environment” interactions. Further integration with gene editing and synthetic biology technologies will achieve large-scale gene mining, efficient gene manipulation, engineered variety design, and systematic biological breeding
[64]. Despite the promising outlook, significant challenges remain in integrating heterogeneous big data, accurately predicting complex genotype–phenotype relationships, and balancing data sharing with intellectual property protection
[65]. Key areas for technological breakthroughs include developing efficient multiomics data integration algorithms, improving the precision of genome editing techniques, and constructing large-scale computational models capable of simulating complex biological systems.
4.2.3. Collaborative innovation platform construction
The new paradigm of intelligent design breeding, which combines next-generation AI and big data, requires a collaborative, multidisciplinary approach
[64]. There is an urgent need to construct a platform-based research strategy characterized by large-scale unions, systems, and collaboration
[66]. This necessitates a top-level design at the national level, proposing mechanisms for collaborative innovation, open sharing, and incentive constraints. By 2040, we aim to support and cultivate leading, flagship innovation groups that integrate research institutions and breeding enterprises, intensifying efforts to overcome key core technological challenges. This will address the fragmentation and disconnection between basic applied research and commercial breeding, establishing a standardized, programmed, informative, and large-scale intelligent breeding system. Through concentrated research efforts, we will cultivate a batch of new high-yield, high-quality, multiresistant, and efficient agricultural varieties, ensuring national food security and enhancing agricultural international competitiveness
[67]. Nevertheless, building such a comprehensive collaborative innovation platform is challenging. Major difficulties include coordinating interests between different institutions and enterprises, establishing effective interdisciplinary cooperation mechanisms, and ensuring long-term stable funding
[55]. Critical areas for advancement include developing information management systems that support large-scale collaboration, establishing scientific and effective mechanisms for intellectual property protection and sharing, and cultivating high-level talent with interdisciplinary backgrounds.
Acknowledgments
This work was partially supported by the Construction of Collaborative Innovation Center of Beijing Academy of Agricultural and Forestry Sciences (KJCX20240406), the Beijing Natural Science Foundation (JQ24037), the National Natural Science Foundation of China (32330075), and the Earmarked Fund for China Agriculture Research System (CARS-02 and CARS-54).
Compliance with ethics guidelines
Ying Zhang, Guanmin Huang, Yanxin Zhao, Xianju Lu, Yanru Wang, Chuanyu Wang, Xinyu Guo, and Chunjiang Zhao declare that they have no conflict of interest or financial conflicts to disclose.
PDF
(2208KB)
