EP4792337A1

EP4792337A1 - Digital twin of end-to-end crop breeding pipelines

Info

Publication number: EP4792337A1
Application number: EP24877990.2A
Authority: EP
Inventors: Andrew Paul BEATTY; Jhonathan PEDROSO RIGAL DOS SANTOS; Eli RODGERS-MELNICK; Alencar Xavier
Original assignee: Pioneer Hi-Bred International Inc
Current assignee: Pioneer Hi-Bred International Inc
Priority date: 2023-10-12
Filing date: 2024-10-10
Publication date: 2026-08-19
Also published as: WO2025080804A1

Abstract

Systems, methods, and apparatuses for system for developing crop varieties are described. Crop breeding pipeline data comprising crop breeding development stages, genome edited variants, and transgenic events associated with an end-to-end breeding pipeline data for crop varieties at different locations may be received. Genotype data, phenotype data, soil data, and environmental data may be acquired and used to generate an end-to-end breeding pipeline digital twin that simulates states of the crop varieties at different crop development stages of the crop breeding pipeline including for genome edited plants. A machine learning model may process the genotype data, phenotype data, soil data, environmental data, and crop breeding pipeline data and generate breeding actions to perform with respect to the crop development stages. Crop development actions include performing breeding crosses, evaluating genome edits, transgenic traits, hybrid testing, variety testing, and crop selection.

Description

DIGITAL TWIN OF END-TO-END CROP BREEDING PIPELINES

TECHNICAL FIELD OF USE

[0001] Aspects of the disclosure relate to the automated development of crop varieties in a crop breeding pipeline. More specifically, aspects of the disclosure provide for the automated monitoring and simulation of crop varieties (including genome edited and transgenic versions) with the assistance of a digital twin of an end-to-end breeding pipeline in order to select plant varieties to breed and/or plant.

BACKGROUND

[0002] A crops breeding pipeline may include a large variety of interconnected processes that may be performed over the course of years. These time intensive processes may involve a significant amount of analysis and processing of data regarding the state of the crop varieties that are being developed. Decisions to develop certain varieties may be made in an ad hoc manner or based on practices that may not have predictable outcomes. In some cases, the decision-making process is based on trial and error, which may result in breeding outcomes that are less than optimal. Numerous operational drawbacks in the current, technological processes and systems are available for improvements.

SUMMARY

[0003] The following presents a simplified summary of various aspects described herein. This summary is not an extensive overview and is not intended to identify key or critical elements or to delineate the scope of the claims. The following summary merely presents some concepts in a simplified form as an introductory prelude to the more detailed description provided below.

[0004] Aspects described herein may allow for automatic methods, systems, non-transitory machine-readable media, and/or devices for the automated development and operation of an end-to-end crop breeding pipeline. More particularly, some aspects described herein may provide a system for developing a plurality of crop varieties in a breeding pipeline. The system may comprise a crop genotyping device configured to generate genotype data comprising an indication of a genotype from a plurality of crop varieties from a population of plants comprising a plurality of parent crop varieties at a plurality of locations. The genotype or genotypic data include genome sequencing data, molecular markers and DNA/RNA and protein sequence data for crop varieties that include traditional breeding, genome editing (or genome edited plants) and transgenic plants (or transgenic traits).

[0005] The system may comprise a crop monitoring device configured to detect, based on use of one or more crop sensors, one or more phenotypic states based on one or more phenotypic traits of the plurality of crop varieties at each of the plurality of locations over a plurality of time intervals. The crop monitoring device may generate phenotype data based on the one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over the plurality of time intervals.

[0006] The system may comprise a soil monitoring device configured to detect, based on one or more soil sensors, one or more soil conditions at the plurality of locations. The soil monitoring device may generate, based on the one or more environmental conditions, soil data based on the one or more soil conditions at each of the plurality of locations over the plurality of time intervals.

[0007] The system may comprise an environmental monitoring device configured to detect, based on one or more environmental sensors, one or more environmental conditions at the plurality of locations. Further, the environmental monitoring device may generate, based on the one or more environmental conditions, environmental data based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals.

[0008] The system may comprise a computing device configured to receive crop breeding pipeline data comprising a plurality of crop breeding development stages associated with an end-to-end breeding pipeline data for the plurality of crop varieties. The crop variety can be, e.g., com and the various development stages can include e.g., inbreds and hybrids. The crop breeding pipeline data may further comprise irrigation data associated with irrigation practices in the end-to-end breeding pipeline, fertilization data associated with fertilizer practices in the end-to-end breeding pipeline, and tillage data associated with tillage practices in the end-to- end pipeline. The one or more states may comprise one or more future states of the plurality of crop varieties at one or more future crop development stages of the plurality of crop development stages. The computing device may acquire the genotype data, the phenotype data, the soil data, and the environmental data. The computing device may generate, based on the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, an end-to-end breeding pipeline digital twin comprising a simulation of one or more states of the plurality of crop varieties at one or more of the plurality of crop development stages of the crop breeding pipeline.

[0009] The computing device may generate, based on inputting the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, into one or more machine learning models, breeding decision data comprising one or more breeding actions to perform with respect to one or more of the plurality of crop development stages. The one or more breeding actions may comprise population development, performing breeding crosses, performing hybrid crosses, hybrid field testing, variety field testing, or selecting one or more combinations of the plurality of crop varieties at the plurality of locations for advancement through the pipeline. The computing device may generate, based on the breeding decision data, breeding action data comprising indications for a crop planting device to plant one or more of the plurality of crop varieties. The system may comprise a crop planting device configured to receive the breeding action data. Further, the crop planting device may plant, based on the breeding action data, one or more quantities of the one or more combinations of the plurality of crop varieties at one of the plurality of locations in order to develop the plurality of crop varieties.

[0010] In an embodiment, crop varieties herein include genome edited plant material as well as transgenic plant material. Genome edited plant material includes for example multiple (or multiplexed) genome edits that target or comprise a plurality of genes, alleles, and haplotypes in one or more chromosomes. Similarly, crop varieties herein include multiple transgenic traits introduced through direct transformation of elite inbreds or introgressed through trait introgression processes and also include molecular stacks (i.e., part of the same DNA construct) or developed through breeding stacks (i.e., breeding distinct lines carrying one or more transgenic events to create a breeding stack that has the multiple transgenic events in the same line).

[0011] According to some aspects described herein, the plurality of crop varieties may comprise one or more double haploid crops. In certain aspects, the doubled haploid crop varieties are generated by a paternal double haploid process, such as for example involving microspores and microspore embryogenesis. [0012] According to some aspects described herein, the plurality of crop varieties may comprise one or more single seed descent crops.

[0013] According to some aspects described herein, the one or more breeding actions may comprise trait integration of one or more of the plurality of crop varieties.

[0014] According to some aspects described herein, the one or more machine learning models may be configured using reinforcement learning. The reinforcement learning may comprise maximization of total reward within a Markov decision process (MDP) in which selection policies performed by automated agents within the end-to-end breeding pipeline digital twin are evaluated, and the selection policies that improve the genetic gain of the plurality of crop varieties are rewarded. The selection policies may comprise selecting one or more of the plurality of crop varieties for development or breeding crosses.

[0015] According to some aspects described herein, the one or more future states of the plurality of crop varieties may be constrained by the genotype and/or the one or more phenotypic states of the plurality of crop varieties at a preceding stage of the plurality of crop development stages.

[0016] Some aspects described herein may provide a method of operating an end-to-end crop breeding pipeline. The method may comprise acquiring, by a computing device comprising one or more processors, genotype data comprising an indication of a genotype of plurality of crop varieties at each of a plurality of locations. The method may comprise acquiring, by the computing device, phenotype data comprising one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over a plurality of time intervals. The acquired data may also include descriptive data generated by humans (e.g., textual descriptions that are written or by other audio/video sources) and other forms of data capture through human intervention. The method may comprise acquiring, by the computing device, based on the one or more environmental conditions, environmental data based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals. The method may comprise acquiring, by the computing device, crop breeding pipeline data comprising a plurality of crop development stages associated with breeding the plurality of crop varieties. The method may comprise generating, by the computing device, based on the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, a digital twin comprising a simulation of one or more states of the plurality of crop varieties at one or more of the plurality of crop development stages of the crop breeding pipeline. The one or more states may comprise one or more future states of the plurality of crop varieties at one or more future crop development stages of the plurality of crop development stages. Further, the one or more states may comprise one or more current states of the plurality of crop varieties at one or more current crop development stages of the plurality of crop development stages. The one or more states may comprise simulated genotype data based on the genotype data, simulated phenotype data based on the phenotype data, simulated environmental data based on the environmental data, and/or simulated crop breeding pipeline data based on the crop breeding pipeline data. Furthermore, the method may comprise generating, by the computing device, based on inputting the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, into one or more machine learning models, breeding decision data comprising one or more breeding actions to perform with respect to one or more of the plurality of crop development stages. The one or more breeding actions may comprise an end-to-end breeding pipeline simulation comprising population development, performing breeding crosses, hybrid or variety testing, or selecting one or more combinations of the plurality of crop varieties at the plurality of locations.

[0017] As used in herein generally, genotypic data includes a variety of sequencing data that comprise e.g., naturally occurring variations including allelic variations, targeted mutations, genome edits, transgenic inserts or events in the genome, gene expression data, metabolomics, expression profiling, marker data, epigenetics profile, and other whole-genome based biologically relevant data that can further describe the genotypic characteristics of the plant or seed or a plant part.

[0018] According to some aspects described herein, the one or more machine learning models may comprise a cross breeding model configured to determine the one or more combinations of the plurality of crop varieties to cross-breed.

[0019] According to some aspects described herein, the one or more machine learning models may comprise a convolutional neural network, a reinforcement learning model, a densely-connected feed-forward neural network, a transformer-based neural network, a normalizing flow, an attention-based neural network, a decision tree, a forest of decision trees, a support vector machine, a Gaussian process, a reproducing kernel Hilbert space model, a generalized additive model, and/or a generalized linear model. [0020] In accordance with some aspects of the optimization processes described herein, Game theory-based optimization models use game theory to solve optimization problems. Some examples of game theory-based optimization models include: Nash equilibrium; Stackelberg game; supply chain optimization; and production scheduling for cost reduction. Game theory is a framework that can be used to represent dynamic correlation among a variety of variables used in the breeding pipeline -e.g., inbred population development and hybrid performance in the field that adapt and vary based on a variety of environmental factors.

[0021] According to some aspects described herein, the environmental data may comprise one or more states of weather at the plurality of locations. The one or more states of the weather may comprise historical precipitation for the plurality of locations, historical amounts of humidity for the plurality of locations, or historical temperatures for the plurality of locations.

[0022] According to some aspects described herein, the environmental data may comprise one or more states of soil at the plurality of locations, one or more amounts of sunlight at the plurality of locations, or one or more altitudes of the plurality of locations. Such environmental data is acquired through a variety of sources such as e.g., satellites, drones (or UAVs), in ground autonomous or semi-autonomous vehicles having a plurality of sensors, hyperspectral cameras, LIDARs, NIR sensors, and other data environmental assessment tools.

[0023] According to some aspects described herein, the phenotype data and/or the environmental data are updated in real-time.

[0024] According to some aspects described herein, the plurality of development stages may comprise a genome editing stage, a transformation stage, a trait introgression stage, a crop population development stage, a crop selection stage, a hybrid creation stage, and/or a crop yield testing stage.

[0025] According to some aspects described herein, the breeding decision data may comprise one or more indications of one or more future time intervals at which to perform the one or more breeding actions.

[0026] Some aspects described herein may provide one or more non-transitory computer readable media comprising instructions that, when executed by at least one processor, cause a computing device to perform operations. The operations may comprise acquiring, by a computing device comprising one or more processors, genotype data comprising an indication of a genotype of plurality of crop varieties at each of a plurality of locations. The operations may comprise acquiring, by the computing device, phenotype data comprising one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over a plurality of time intervals. The operations may comprise acquiring, by the computing device, based on the one or more environmental conditions, environmental data based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals. The operations may comprise acquiring, by the computing device, crop breeding pipeline data comprising a plurality of crop development stages associated with breeding the plurality of crop varieties. The operations may comprise generating, by the computing device, based on the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, a digital twin comprising a simulation of one or more states of the plurality of crop varieties at one or more of the plurality of crop development stages of the crop breeding pipeline. The one or more states may comprise one or more future states of the plurality of crop varieties at one or more future crop development stages of the plurality of crop development stages. Further, the one or more states may comprise one or more current states of the plurality of crop varieties at one or more current crop development stages of the plurality of crop development stages. The one or more states may comprise simulated genotype data based on the genotype data, simulated phenotype data based on the phenotype data, simulated environmental data based on the environmental data, and/or simulated crop breeding pipeline data based on the crop breeding pipeline data. Furthermore, the operations may comprise generating, by the computing device, based on inputting the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, into one or more machine learning models, breeding decision data comprising one or more breeding actions to perform with respect to one or more of the plurality of crop development stages. The one or more breeding actions may comprise an end-to-end breeding pipeline simulation comprising population development, performing breeding crosses, hybrid or variety testing, or selecting one or more combinations of the plurality of crop varieties at the plurality of locations.

[0027] In some aspects, predictions from simulated phenotypic and environmental data may also be used to generate virtual reality applications that simulate performance of crops in time and space and providing human-interpretable visualizations and real-time analytics. These simulations, predictions, and virtual reality applications based on the digital twin concepts described here further help improve the evaluation analytics engine for crop improvement overall, including for breeding and genome-edited pipelines. [0028] Corresponding apparatuses, devices, systems, and/or computer-readable media (e.g., non-transitory computer readable media) are also within the scope of the disclosure. By generating and implementing more effective crop breeding decisions, the benefits of an improved end-to-end crop breeding pipeline may be achieved. These features, along with many others, are discussed in greater detail below.

BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present disclosure is illustrated by way of example and not limited in the accompanying figures in which like reference numerals indicate similar elements and in which:

[0030] FIG. 1 shows an example of a computing system that may be used to implement one or more aspects of the disclosure in accordance with one or more illustrative aspects discussed herein.

[0031] FIG. 2 shows an example of a machine learning model according to one or more aspects of the disclosure.

[0032] FIG. 3A-3D show an example of an end-to-end crop breeding pipeline according to one or more aspects of the disclosure. An illustration of the corn breeding and advancement pipeline for the population development phase is shown in FIG. 3 A. The inbred state space and the advancement framework including TC0, TC2 and R1 are shown in FIG. 3B. The hybrid state space and the advancement framework including R2 and R3 along with hybrid observation, hybrid yields are shown in FIG. 3C. The reward function aspects and the impact to the advancement decisions for R4 and R5 pools are shown in FIG. 3D.

[0033] FIG. 4 shows a timing diagram that indicates a data flow of signals associated with generating and implementing breeding decision data according to one or more aspects of the disclosure.

[0034] FIG. 5 shows an example flow chart for automated crop propagation, crop monitoring, and environmental monitoring according to one or more aspects of the disclosure.

[0035] FIG. 6 shows an example flow chart for automated monitoring and operation of an end-to-end crop breeding pipeline according to one or more aspects of the disclosure. [0036] FIG. 7 shows an example flow chart for automatically planting and irrigating crops according to one or more aspects of the disclosure.

[0037] FIG. 8 shows an example of a Markov decision process used in an end-to-end breeding pipeline according to one or more aspects of the disclosure. A schematic representation of a reinforcement learning model involving an agent and an environment as part of the Markov Decision Process (MDP) for the agent taking action as a selector and generating the states for the digital twin is shown. States and rewards at any given timepoint depend on the previous timepoint’s state and agent’s action.

[0038] FIG. 9 shows a schematic representation of illustrative embodiments of what can be represented in the action space in a reinforcement learning model involving an agent and an environment as part of the Markov Decision Process (MDP) (as shown in FIG. 8). Examples of the action space include: modification of secondary trait thresholds for selection based on YIELD vs. MST (moisture) residuals; vary soft vs. hard constraints, secondary trait thresholds, diversity constraints for algorithmic selection; directly provide binary choice of select/discard for each candidate.

DETAILED DESCRIPTION

[0039] In the following description of the various embodiments, reference is made to the accompanying drawings, which form a part hereof, and in which is shown by way of illustration various embodiments in which aspects of the disclosure may be practiced. It is to be understood that other embodiments may be utilized and structural and functional modifications may be made without departing from the scope of the present disclosure. Aspects of the disclosure are capable of other embodiments and of being practiced or being carried out in various ways. Also, it is to be understood that the phraseology and terminology used herein are for the purpose of description and should not be regarded as limiting. Rather, the phrases and terms used herein are to be given their broadest interpretation and meaning. The use of “including” and “comprising” and variations thereof is meant to encompass the items listed thereafter and equivalents thereof as well as additional items and equivalents thereof.

[0040] Aspects described herein are generally directed to the development of a more effective end-to-end crop breeding pipeline that may be used to improve the genetic gain, long-term yield, and diversity of crop varieties. The disclosed technology may be used to generate a novel digital twin representation of an end-to-end crop breeding pipeline that may be used to simulate the state of the actual crop breeding pipeline upon which it is based. Further, the digital twin may be updated in real-time, thereby providing the benefits of up to date information from which to determine more effective crop selection policies. The disclosed technology may then leverage the use of a machine learning model that may be configured and/or trained based on use of reinforcement learning in which the model is optimized to maximize total reward within a Markov decision process, wherein the machine learning model is used to select crop selection policies that may more optimally be used to breed crop varieties. The crop selection policies may be performed by automated agents (e.g., software agents) that may then be used to generate breeding actions. The breeding actions may be used to perform actual actions within the actual end-to-end crop breeding pipeline on which the digital twin is based. For example, the breeding actions may include the selection of certain crop varieties for development, testing, and/or breeding. By simulating the end-to- end crop breeding pipeline and determining optimized crop selection policies in advance of actual decisions, some of the costs and time usage associated with trial and error testing on actual crops may be mitigated. Further, the improved crop selections may result in improved genetic gain and diversity for the selected crop varieties.

[0041] According to the aspects described herein, the technical effects and benefits described herein may improve crop breeding decisions in order that crop varieties may be more optimally developed, bred, tested, and selected. Further, the reinforcement learning techniques used to develop crop selection policies may result in more diverse crop varieties that allow for greater long-term improvements in crop traits including improved genetic gain, yield, disease resistance, and resiliency. An end-to-end crop breeding pipeline generally includes methods and compositions that include variety development (e.g., inbred line development) and incorporates one or more genomic modifications such as genome editing or genome edited traits, native traits (or alleles), transgenic traits, and other heritable genomic modifications that are introduced, bred with or introgressed during a breeding process.

[0042] Before discussing these concepts in greater detail, however, several examples of computing devices and/or computing systems that may be used in implementing and/or otherwise providing various aspects of the disclosure will first be discussed with respect to FIG. 1. [0043] FIG. 1 shows an example of a system in accordance with aspects of the present disclosure. In particular, FIG. 1 depicts a diagram of a computing system that may be configured to perform operations comprising the exchange and processing of signals and/or data that may be used to generate breeding decision data for use in breeding crops in an end- to-end crop breeding pipeline. As shown in FIG. 1, the computing system includes the network 102, a computing system 104, a crop genotyping device 130, a crop monitoring device 132, an environmental monitoring device 134, a crop planting device 136, and an irrigation device 138. Computing system 104 may comprise one or more processors 112, memory 114, one or more machine learning models 116, phenotype data 118, environmental data 120, crop breeding pipeline data 122, genotype data 124, and/or breeding decision data 126. System 100 may operate in a standalone environment and/or as part of a networked environment that may include other devices and/or systems. For example, system 100 may operate in conjunction with other computing systems and/or other computing devices not shown in FIG. 1. As shown in FIG. 1, various computing devices including computing system 104, crop genotyping device 130, crop monitoring device 132, environmental monitoring device 134, crop planting device 136, and/or irrigation device 138 may be interconnected via the network 102. Further, the system 100 may operate via one or more networks not including the network 102 shown in FIG. 1.

[0044] Network 102 may be used to communicate (e.g., send and/or receive) signals, information, and/or data. For example, network 102 may be used to communicate environmental data 120 that may be sent from environmental monitoring device 134 to computing system 104. Network 102 may include any combination of wired and/or wireless networks and may carry any type of signal or communication including communications and/or signals using one or more communication protocols (e.g., TCP/IP, HTTP, and/or HTTPS). Further, network 102 may include any combination of a local area network (LAN), an intranet, a wide area network (WAN), and/or the Internet. Furthermore, network 102 may be configured or arranged according to any known topology and/or architecture.

[0045] Computing system 104 may, in some embodiments, implement one or more aspects of the present disclosure by accessing and/or executing instructions; and/or performing one or more operations based at least in part on the instructions. For example, computing system 104 may generate instructions (e.g., breeding decision data 126) that may comprise breeding action data that may be used by the crop planting device 136. In some embodiments, the system 100 may be incorporated into and/or include a computing device (e.g., a computing device with one or more processors, one or more memory devices, one or more input devices, and/or one or more output devices). For example, computing system 104 may be incorporated into and/or include a desktop computer, a computer server, a computer client, a mobile device (e.g., a laptop computer, a tablet computer, a smart phone, and/or a smart watch), an agricultural device (e.g., tractor and/or harvester), and/or any other type of processing device.

[0046] Computing system 104 may include one or more interconnects for communication between different components of the computing device. Computing system 104 may also include a network interface via which computing system 104 may exchange one or more signals including information and/or data with other computing systems and/or computing devices. For example, computing system 104 may send information and/or data (e.g., the breeding decision data 126) to crop planting device 136 via the network 102. By way of further example, computing system 104 may receive information and/or data, via the network 102, from crop monitoring device 132 and/or environmental monitoring device 134 and may acknowledge that the information and/or data was received. Further, computing system 104 may include one or more input devices (e.g., a keyboard, mouse, touch screen, stylus, and/or microphone) and/or one or more output devices (e.g., a display device and/or audio output devices including loudspeakers).

[0047] Computing system 104 may include one or more computing devices. Further, as seen in FIG. 1, computing system 104 may include one or more processors 112 and a memory 114. The one or more processors 112 may include any combination of processing devices (e.g., one or more computer processing units (CPUs), one or more graphics processing units (GPUs), one or more processor cores, one or more microprocessors, one or more application specific integrated circuits (ASICs), one or more field programmable gate arrays (FPGAs), and/or one or more controllers). By way of example, computing system 104 may comprise the one or more processors 112 and memory 114 that may store instructions that, when executed by the one or more processors 112, causes computing system 104 to perform operations which may include the operations described herein. The one or more processors 112 may execute instructions including instructions stored in the memory 114. Further, the one or more processors 112 may be arranged in various configurations including any combination of one or more serial processors and/or one or more parallel processors. [0048] The memory 114 may include one or more computer-readable media (e.g., non- transitory computer-readable media) and may be configured to store data and/or instructions including one or more machine learning models 116 and/or phenotype data 118. Further, the memory 114 may include one or more memory devices including random access memory (RAM), read only memory (ROM), programmable read only memory (PROM), erasable programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), solid state drives (SSDs), hard disk drives (HDDs), and/or hybrid memory devices that use a combination of different types of memory technologies.

[0049] As shown in FIG. 1, the memory may be used to store data which may comprise one or more machine learning models 116, phenotype data 118, environmental data 120, crop breeding pipeline data 122, and/or genotype data 124. The genotype data also include genome edits and transgenes. One or more machine learning models 116 may include instructions that may be used perform operations based on data comprising phenotype data 118, environmental data 120, crop breeding pipeline data 122, and/or genotype data 124. As described herein, one or more machine learning models 116 may be configured to generate breeding decision data for a crop based on phenotype data 118, environmental data 120, crop breeding pipeline data 122, and/or genotype data 124, associated with a crop. Further, one or more machine learning models 116 may be configured to generate output comprising breeding decision data 126, which may be used to generate breeding action data that may be used by crop planting device 136 to plant seeds of a crop variety.

[0050] Phenotype data 118 may include information associated with traits of a plurality of crop varieties. The phenotype data 118 may comprise information associated with crop traits comprising yield, height, biomass, root length, and/or leaf shape. The phenotype data 118 may be acquired and/or generated by crop monitoring device 132. Further, phenotype data 118 may comprise information indicating the location of the plurality of crops (e.g., latitude, longitude, and/or altitude). Phenotype data 118 may include the information and/or features of phenotype data 204 described herein with respect to FIG. 2.

[0051] Environmental data 120 may include information associated with the current, past, and/or predicted state (e.g., precipitation, temperature, humidity, and/or barometric pressure) of the environment at a location of a crop. Environmental data 120 may be based on information acquired by the environmental monitoring device 134. Further, environmental data 120 may be used in the generation of breeding decision data and/or breeding decision data 126 as described herein. Environmental data 120 may include the information and/or features of environmental data 206 described herein with respect to FIG. 2.

[0052] The crop breeding pipeline data 122 may include information associated with an end-to-end breeding pipeline digital twin that is a simulation of the states of crop varieties at crop development stages of the crop breeding pipeline. For example, the crop breeding pipeline data 122 may indicate the state of crops at the crop population development stage, crop selection stage, hybrid creation stage, or crop yield testing stage. Further, the crop breeding pipeline data 122 may indicate the geographic location of crops in the crop breeding pipeline. Crop breeding pipeline data 122 may be used in the generation of breeding decision data 126 as described herein. Crop breeding pipeline data 122 may include the information and/or features of crop breeding pipeline data 208 described herein with respect to FIG. 2.

[0053] Genotype data 124 may include information associated with the genotype of a crop. Genotype data 124 may be associated with a crop grown in a location from which the crop monitoring device 132 uses sensors to detect the state of a crop. For example, image sensors may be used to determine the height of crops. Further, genotype data 124 may be used in the generation of breeding decision data as described herein. Genotype data 124 may include the information and/or features of genotype data 210 described herein with respect to FIG. 2.

[0054] Breeding decision data 126 may include information that indicates one or more crop breeding actions that may be performed, which includes which genome edits are introduced during the breeding population development and which genome edits are considered as part of the advancement process or decision-making steps. For example, breeding decision data 126 may be used to develop a genetic variant (which includes genome edited variants and/or transgenic events) of a crop, select certain crop varieties, evaluate traits of crop varieties, and/or distribute and multiply seeds of a crop variety. The breeding decision data 126 may be used to generate breeding action data which uses a subset of the breeding decision data 126 and may comprise instructions that may be used by devices at a particular location to perform breeding actions on a set of crops indicated in the breeding action data.

[0055] Soil data 128 may include information associated with the current and/or past state of soil sampled from a plurality of locations in which a plurality of crop varieties are grown. Soil data 128 may be based on information acquired by a soil sampling device. Further, soil data 128 may be used in the generation of breeding decision data and/or breeding decision data 126 as described herein. Soil data 128 may include the information and/or features of soil data 211 described herein with respect to FIG. 2.

[0056] The crop genotyping device 130 may include a mechanical and/or electronic device that may be configured to acquire crop samples from various locations and determine the genotype of each crop sample. For example, DNA sequencing may be used to determine the genotype of a crop. The crop genotyping device 130 may for example, determine the genotype of various crops including rice, canola, sunflower, sorghum, wheat, potatoes, barley, corn, soybeans, and/or cotton. The crop genotyping device 130 may generate genotype data 124 based on the crop samples acquired and/or analyzed by the crop genotyping device 130. The crop genotyping device 130 may include any of the features and/or components of computing system 104. For example, a crop genotyping device 130 may include one or more processors, a memory, one or more input devices, and/or one or more output devices. Further, a crop genotyping device 130 may have different or similar configurations and/or architectures to that of computing system 104.

[0057] The crop monitoring device 132 may include a mechanized and/or electronic device that may be configured to acquire and/or analyze samples of crops. In some embodiments, the crop monitoring device 132 may determine the phenotype of crops using various sensors including image sensors (e.g., image sensors that may be used to determine the height and/or yield of a crop) and/or weight sensors (e.g., scales that may be used to determine the weight of a crop).. The crop monitoring device 132 may generate phenotype data 118 based on the crop samples acquired and/or analyzed by the crop monitoring device 132. The crop monitoring device 132 may include any of the features and/or components of computing system 104. For example, a crop monitoring device 132 may include one or more processors, a memory, one or more input devices, and/or one or more output devices. Further, a crop monitoring device 132 may have different or similar configurations and/or architectures to that of computing system 104.

[0058] Environmental monitoring device 134 may comprise a mechanized and/or electronic device that may be configured to determine the weather conditions at a location. For example, the environmental monitoring device 134 may comprise one or more sensors (e.g., thermometers, barometers, humidity sensors, rain gauges, anemometers, and/or aerial cameras) that may be used to determine environmental conditions comprising temperature (e.g., air temperature at a location of a crop), humidity, radiation, barometric pressure, rainfall, wind direction, and/or wind speed. Further, the environmental monitoring device 134 may be operated by and/or receive information from a third-party entity that generates environmental data for one or more locations including a location of a crop that is being monitored. The environmental monitoring device 134 may include any of the features and/or components of computing system 104. For example, an environmental monitoring device 134 may include one or more processors, a memory, one or more input devices, and/or one or more output devices. Further, the environmental monitoring device 134 may have different or similar configurations and/or architectures to that of computing system 104.

[0059] Crop planting device 136 may include a mechanized and/or electronic device that may be configured to perform operations comprising planting seeds of a crop. In some embodiments, crop planting device 136 may plant seeds of one or more crops comprising rice, wheat, potatoes, barley, corn, soybeans, and/or cotton. The crop planting device 136 may be configured to receive one or more signals which may comprise breeding action data based on the breeding decision data 126. Further, crop planting device 136 may be configured to plant one or more seeds of a crop by processing breeding action data based on the breeding decision data 126. For example, crop planting device 136 may plant cottonseeds in a location and quantity and at a time indicated in the breeding action data. The crop planting device 136 may include any of the features and/or components of computing system 104. For example, a crop planting device 136 may include one or more processors, a memory, one or more input devices, and/or one or more output devices. Further, a crop planting device 136 may have different or similar configurations and/or architectures to that of computing system 104.

[0060] Irrigation device 138 may include a mechanized device, electronic device, and/or structures that may be configured to irrigate crops. In some embodiments, the irrigation device 138 may irrigate crops using water and/or other substances that may be used to promote or inhibit the growth of crops. The irrigation device 138 may perform surface irrigation (e.g., furrow irrigation) that irrigates crops through use of water channels, localized irrigation in which crops may be irrigated using pipes, drip irrigation in which hoses may be configured to drip water onto crops, and/or sprinkler irrigation in which crops are irrigated by being sprayed with water. Irrigation device 138 may include any of the features and/or components of computing system 104. For example, an irrigation device 138 may include one or more processors, a memory, one or more input devices, and/or one or more output devices. Further, an irrigation device 138 may have different or similar configurations and/or architectures to that of computing system 104.

[0061] One or more aspects described herein may be embodied in computer-usable, computer-readable data, and/or computer-executable instructions, which may be stored as data and/or instructions in one or more memory devices and/or executed by one or more computing devices and/or other devices described herein. Data and/or instructions may include software applications and/or computer programs that may be used to perform the operations described herein (e.g., generating one or more breeding actions) when executed by one or more processors in a computing device or other device. The data and/or instructions be written in source code that is compiled for execution by a computing device. The computer executable instructions may be stored on a computer readable medium (e.g., a non-transitory computer- readable medium) such as a hard disk, solid state drive, optical disk, removable storage media, solid state memory, and/or RAM. The functionality of the computing applications described herein may be combined or distributed in various embodiments. Further, the functionality of the computing applications described herein may be partly or wholly embodied in firmware or hardware equivalents including integrated circuits and/or field programmable gate arrays (FPGA).

[0062] FIG. 2 shows an example of a machine learning model according to one or more aspects of the disclosure. One or more machine learning models 212 may be implemented on the system 100 illustrated in FIG. 1, according to an embodiment of the invention.

[0063] Input 202 may comprise various data and/or information that may be used singly or in combination as an input to one or more machine learning models 212. As shown in FIG. 2, input 202 comprises phenotype data 204, environmental data 206, crop breeding pipeline data 208, and genotype data 210. Input 202 may be based on data received from a variety of sources (e.g., the system 100 described with respect to FIG. 1 and/or a remote computing system that provides data) and one or more portions of data may be added to or removed from input 202.

[0064] Input 202 may comprise phenotype data 204 which may comprise data and/or information associated with traits of a crop in a geographic location. Phenotype data 204 may be based on analysis of crops detected at a geographic location (e.g., images of crops acquired by the crop monitoring device 132 described with respect to FIG. 1). Further, phenotype data 204 may comprise information from phenotype data 118 that is described with respect to FIG.

1.

[0065] Input 202 may comprise environmental data 206 which may comprise information associated with the state of an environment in a location (e.g., daily temperatures at a location over time). Environmental data 206 may be based on weather conditions of a geographic location (e.g., weather conditions determined and/or acquired by environmental monitoring device 134 described with respect to FIG. 1). Further, environmental data 206 may comprise information from environmental data 120 that is described with respect to FIG. 1.

[0066] Input 202 may comprise crop breeding pipeline data 208 which may comprise information associated with an end-to-end breeding pipeline digital twin that simulates the states of crop varieties at crop development stages of the crop breeding pipeline. Further, environmental data 206 may comprise information from crop breeding pipeline data 122 that is described with respect to FIG. 1.

[0067] Input 202 may comprise genotype data 210 which may comprise information associated with the genotype of a crop in a location. Genotype data 210 may indicate the unique DNA sequence. Further, environmental data 206 may comprise information from environmental data 120 that is described with respect to FIG. 1. Soil data 211 may comprise information from soil data 128 that is described with respect to FIG. 1.

[0068] One or more machine learning models 212 may be configured and/or trained to generate breeding decision data 214. Determination of breeding decision data 214 may be based on one or more machine learning models 212 that are configured and/or trained to receive input 202 which may comprise phenotype data 204, environmental data 206, crop breeding pipeline data 208, genotype data 210, and/or soil data 211. The one or more machine learning models 212 may process the input 202 and generate output comprising breeding decision data 214. Breeding decision data 214 may comprise a plurality of breeding actions. The plurality of breeding actions may comprise indications of crops on which to perform the breeding actions, locations at which to perform the plurality of breeding actions, and/or a type of breeding action to perform (e.g., crop selection, hybrid testing, and/or breeding crosses). The plurality of breeding action may comprise an indication of a stage of crop development at which to perform a breeding action. Further, the plurality of breeding actions may be associated with a corresponding plurality of time intervals in which to perform the breeding actions. [0069] One or more machine learning models 212 may be configured and/or trained to analyze inputs (e.g., input 202) and generate breeding decision data based on the inputs. One or more machine learning models 212 may, for example, comprise one or more neural networks (e.g., one or more recurrent neural networks (RNNs), one or more multi-layer perceptrons (MLPs), and/or one or more convolutional neural networks (CNNs), one or more random forest models, one or more support vector machines (SVMs), and/or Bayesian hierarchical models. Other neural network architectures such as transformers are also readily applicable to the digital twin framework. Transformers are a type of neural network architecture that transforms an input sequence of vectors into an equally-sized output sequence of vectors by learning contextual relationships between sequence components through an attention-based process. See e.g., Vaswani et al., (2017). "Attention is All you Need". Advances in Neural Information Processing Systems. 30. Curran Associates, Inc. They are foundational to modern natural language processing (NLP) tasks, in addition to finding extensive use in computer vision, biological sequence, and other task domains benefitting from modelling relationships between data features. Large language models (LLM), such as GPT (Generative Pre-trained Transformer) and BERT (Bidirectional Encoder Representations from Transformers) are some examples of transformer-based models that can learn complex sequence representations. Other attention-based network structures, such as structured state space models like Structured State Space sequence (S4) model and MAMBA a deep learning architecture that incorporates sequence modeling including S4, may also be applied to deep learning tasks within the digital twin end-to-end crop breeding framework. See e.g., Gu, Albert; Dao, Tri (2023). "Mamba: Linear-Time Sequence Modeling with Selective State Spaces". arXiv:2312.00752. The term machine learning model may be construed as meaning one or more machine learning models any of which may operate singularly or in combination with one or more other machine learning models to perform the operations described herein.

[0070] Further, one or more machine learning models 212 may be configured and/or trained using various training techniques including supervised learning, unsupervised learning, semi -supervised learning, and/or reinforcement learning. One or more machine learning models 212 may, for example, comprise parameters that have adjustable weights and fixed biases. As part of the process of configuring and/or training the one or more machine learning models 212, values associated with each of the weights of the one or more machine learning models 212 may be modified based on the extent to which each of the parameters contributes to increasing or decreasing the accuracy of output generated by the one or more machine learning models 212. For example, parameters of one or more machine learning models 212 may correspond to various features of inputs (e.g., inputs based on phenotypic features, genotypic features, environmental features, soil features, and/or breeding pipeline features). Over a plurality of iterations and based on inputting training data to one or more machine learning models 212, the weighting of each of the parameters may be adjusted based on the extent to which each of the parameters contributes to accurately generating breeding decision data at sample locations.

[0071] Training the one or more machine learning models 212 may comprise the use of a cost function that may be used to minimize the error between output of the one or more machine learning models 212 and a ground-truth value. For example, one or more machine learning models 212 may receive input comprising training data comprising phenotype training data similar to phenotype data 204, environmental training data similar to environmental data 206, crop breeding pipeline training data similar to crop breeding pipeline data 208, and/or genotype training data similar to genotype data 210. The training data may comprise historical breeding decision data that indicates historical states of a crop breeding pipeline. Accurate output by the one or more machine learning models 212 may include generating breeding decision data that closely matches (e.g., 95% similarity) ground-truth states of the historical breeding decision data. For example, ground-truth breeding decision data may indicate that specific combinations of crop varieties were selected. If the breeding decision data generated as output by the one or more machine learning models comprises is very similar to the ground-truth breeding data, and indicates similar combinations of the same crop varieties as the ground-truth breeding decision data then the output may be determined to be highly accurate. If the breeding decision data generated as output by the one or more machine learning models comprises is very different from the ground-truth breeding data and indicates mostly different combinations of crop varieties then the output may be determined to be highly inaccurate. Inaccurate output by one or more machine learning models 212 may include determining breeding decision data that does not match and/or is very dissimilar to (e.g., outside of a predetermined range of similarity) the ground-truth yield. Over a plurality of training iterations, the weighting of the parameters of one or more machine learning models 212 may be adjusted until the accuracy of the machine learning model’s output reaches some threshold accuracy level (e.g., 96% accuracy).

[0072] FIG. 3A-3D show an example of an end-to-end crop breeding pipeline according to one or more aspects of the disclosure. Operations of an end-to-end crop breeding pipeline may be performed by the systems and devices described herein including the system 100 shown in FIG. 1.

[0073] FIG. 3A shows a digital twin representation of an end-to-end breeding pipeline 300. The end-to-end breeding pipeline 300 may be based on an actual end-to-end breeding pipeline that is used to develop crop varieties. Further, the end-to-end breeding pipeline 300 may comprise different stages that may correspond to different time periods. In this example, the end-to-end breeding pipeline 300 may comprise a plurality of stages that occur over the course of a five-year period. Different end-to-end pipelines may comprise varying numbers of stages and may occur over varying time periods that may be shorter or longer than the five- year period shown in this example.

[0074] The end-to-end breeding pipeline 300 may comprise a parent pool (e.g., parent crop varieties having an associated indication of genotype data) from which various crop varieties are taken. Further, the end-to-end breeding pipeline 300 may comprise a population development stage in which different crops may be developed. The population development stage may comprise the development of single seed descent crops that are homozygous over a plurality of filial generations (e.g., F1-F5). Further, the population development stage may comprise the development of double haploid crops that are homozygous by crossing heterozygous crops with an inducer line to produce haploid embryos.

[0075] In an illustrated embodiment shown in FIG. 3 A for the end-to-end corn breeding pipeline, the state and the action space of the advancement pipeline along with the associated process are shown e.g., parent pool selection step, cross-selection or allocation step, generation of double haploid populations (including maternally derived DHs and paternally - microspore- derived DHs), single seed descents (SSDs). The use molecular markers to select, identify and/or predict the performance using GEBVs is shown as an example. The pool of available genotypes for selection by the reinforcement learning (RL) agent and the action space of the RL agent are shown. The various “existing” process states and the “simulated” process states are shown by clear lines or hatched lines in FIG. 3A population development panel section. Selected populations are advanced to the next stage and those that did not get selected are discarded at the TC0 stage.

[0076] In an illustrated embodiment shown in FIG. 3B for the end-to-end com breeding pipeline, the inbred state space is further described. The per se testing record and hybrid testing record are updated according to simulation processes of the digital twin based on the plurality of testing locations simulated over time. Inbred line rankings are updated based on selection actions, where the outcome is directly updated according to the actions of the RL agent. Rankings and testing records may vary by region and over time, while the genotype - including its pedigree, trait package, and any potential genome edits - are constant. The marker profile may vary with time, based on simulated ascertainment from different genotyping platforms. As shown in FIG. 3B, the TCO pool is further evaluated based on inbred obs data producing per se simulated phenotypes at experimental locations. The TC2 pool is evaluated based on simulated hybrid obs and yield trials, given locations and testing configurations. Simulated phenotypes are processed into BLUPs and GEBVs. Inbred Obs and their associated per se values are also considered for the TC2 population development. For the R1 pool selection, the evaluation criteria are similar to the TC2 population pool, with location and testing configurations producing simulated hybrid obs and hybrid yield trials from which BLUPs and GEBVs may be calculated. Likewise, inbred obs trials provide per se values for additional inbred characterization.

[0077] In an illustrated embodiment shown in FIG. 3C for the end-to-end com breeding pipeline, the hybrid state space is further described. Followed by the ranking step in the previous process, the Code Pool of inbreds are used to generate the Hybrid Cross Allocation step that is performed by the RL agent. Similar to the R1 selection, the R2 pool is selected based on the evaluation criteria including BLUPs and GEBVs arising from the simulated hybrid obs and hybrid yield trials run for specified locations and testing configurations. As shown in FIGs. 3B and 3C, trait introgression (whether transgenic traits or genome-edited traits) is performed during the R1 and R2 stages. These can also be performed in earlier stages. For example, genome-edited variants can enter the inbred state space at TCO, TCI, TC2, R1 and/or R2 stages.

[0078] In an illustrated embodiment shown in FIG. 3D for the end-to-end corn breeding pipeline, the reward function for reinforcement learning processes is further described. In the illustrated example, some of the positive attributes for the reward function include product concepts, genetic gain by stage, and remaining diversity (all emanating from selection processes conducted by RL-trained agents) and some of the negative attributes include phenotyping costs, population development costs, trait introgression (TI) costs, and genotyping costs. Seed increase is another function that is shown in FIG. 3D indicating the later stages of the end-to-end breeding pipeline. The R4 and R5 pool indicates high-performing elite inbred class that have advanced based on the hybrid performance.

[0079] Molecular markers in the population of crops may be used to detect polymorphisms within the population of crops. Detection of molecular markers may be based on DNA sequencing, PCR-based sequence specific amplification methods, detection of restriction fragment length polymorphisms (RFLP), detection of isozyme markers, detection of polynucleotide polymorphisms by allele specific hybridization (ASH), detection of amplified variable sequences of the plant genome, detection of self-sustained sequence replication, detection of simple sequence repeats (SSRs), detection of single nucleotide polymorphisms (SNPs), and/or detection of amplified fragment length polymorphisms (AFLPs).

[0080] CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) genome editing is a powerful tool used to modify plant DNA with high precision. It can be used to edit multiple genes simultaneously, enhancing the ability to develop crops that are more resilient to environmental stresses and have improved potential for higher yield. Endonucleases are enzymes that cleave the phosphodiester bond within a polynucleotide chain. Examples of endonucleases include restriction endonucleases, meganucleases, TAL effector nucleases (TALENs), zinc finger nucleases, and Cas (CRISPR-associated) effector endonucleases. In some aspects of the methods and compositions described herein, a genome editing system comprises a Cas endonuclease, one or more guide polynucleotides, and optionally donor DNA, and editing a target polynucleotide sequence comprises nonhomologous end-joining (NHEJ) or homologous recombination (HR) following a Cas endonuclease-mediated double-strand break. In some aspects of the methods and compositions described herein, a genome editing system comprises a base editing agent and a plurality of guide polynucleotides and editing a target polynucleotide sequence comprises introducing a plurality of nucleobase edits in the target polynucleotide sequence resulting in a variant nucleotide sequence. One or more nucleobases of a target polynucleotide can be chemically altered, in some cases to change the base from one type to another, for example from a Cytosine to a Thymine, or an Adenine to a Guanine. In some aspects of the methods and compositions described herein, a genome editing system comprises a prime editing agent and a guide polynucleotide and editing a target nucleotide sequence comprises introducing one or more insertions, deletions, or nucleobase swaps in a target nucleotide sequence without generating a double-stranded DNA break. In some aspects, the prime editing agent is a Cas polypeptide fused to a reverse transcriptase, wherein the Cas polypeptide is modified to nick DNA rather than generating double-strand break. This Cas-polypeptide-reverse transcriptase fusion can also be referred to as a "prime editor" or "PE".

[0081] The end-to-end breeding pipeline 300 may comprise a plurality of selection stages. The plurality of selection stages may comprise sub-stages in which inbred crop phenotypes, hybrid crop phenotypes, hybrid special phenotypes, and hybrid yield phenotypes may be tested. The testing stage of the end-to-end breeding pipeline 300 may comprise hybrid creation, yield testing, and/or purity testing. Selection of crops may be based on the use of genomic estimated breeding values (GEBVs) that may be determined for crops and which may indicate a measurable degree to which one or more, polymorphic markers, haplotypes and/or genotypes heritably affect the expression of a phenotype associated with a trait.

[0082] Further, the end-to-end breeding pipeline 300 may comprise a trait integration stage and/or a seed integration stage. The trait integration stage may be used to introduce traits of interest (e.g., traits that increase yield) into a crop.

[0083] The end-to-end breeding pipeline 300 may be monitored by a plurality of sensors and other devices that may be used to generate the phenotype data, environmental data, crop breeding pipeline data, and/or genotype data described herein. The phenotype data, environmental data, crop breeding pipeline data, and/or genotype data may be used to generate a digital twin of the end-to-end breeding pipeline 300. The digital twin of the crop end-to-end breeding pipeline 300 may comprise a simulation of the actual crop breeding pipeline that may be updated in real-time or near real-time based on changes in the actual crop breeding pipeline. Further, the digital twin of the crop breeding pipeline may be used to perform simulations of the crop breeding pipeline that may be used to predict future states of the crop breeding pipeline based on current states of the crop breeding pipeline. The digital twin of the crop end-to-end breeding pipeline may be used as an environment within a Markov decision process. Further, selection policies may be performed by automated agents (e.g., software agents) within the environment (e.g., the digital twin of the crop end-to-end breeding pipeline).

[0084] The actions performed by the automated agents may govern the processes of selection (e.g. advancing crop genotypes/varieties through stages of the pipeline) and crossing (e.g. creating new genotypes/varieties from those existing). Based on an action by the automated agents (e.g., actions comprising a set of crop variety selections, crop breed crossings, and/or changes to a parameterized selector), the digital twin environment may generate a new state (e.g., genotypes, phenotypes, and/or ranks), and rewards (genetic gain). The new states may include extant and/or potential (e.g., non-existent or hypothetical) genotypes. The selection policy may be parameterized as a type of machine learning model (e.g., a neural network, decision tree, random forest, and/or linear model) or as a random selector. A selection policy may be expressed as 7t, an action by an automated agent at time t may be expressed as At, and a state of the environment may be expressed as St. The probability of an agent performing an action at time t may be expressed as 7t(A_t|St). Further, the probability of an agent performing an action at time t may be parameterized by some 9 (theta) which may specify specific selection policy instantiations. Selection policy evaluation may comprise determining the value (i.e., expected reward over multiple generations) for a selection policy. For example, the value function may output a total amount of reward (e.g., a reward at time t expressed as Rt) that is expected to accumulate when automated agents perform a selection policy starting from a given state. This may provide a way of using the digital twin prior to performing reinforcement learning.

[0085] Selection policy optimization may comprise operations to improve a selection policy. The selection policy may be optimized by having the automated agents act greedily with respect to a value function (e.g., R) or by directly optimizing the parameters (e.g. 9) of the machine learning model representing the policy (e.g. 7t(A_t|St)) with respect to the expected value. For example, selection policies may be optimized to include actions (e.g., A) that increase the reward value (e.g., R). An optimized selection policy may then be deployed into an active breeding program, selecting genotypes to advance (e.g., crop genotypes to be tested in a subsequent round of experiments) and/or parents to be crossed for generation of seed.

[0086] In some embodiments, predictions of future states of the crop breeding pipeline may be generated through use of a machine learning model that is trained using reinforcement learning. Reinforcement learning of the machine learning model may occur within a Markov decision process. The Markov decision process may generate states and rewards that may be used optimize breeding decisions (e.g., make more favorable crop breeding decisions that result in crops with greater genetic gain, greater yield, greater resistance to disease, and/or greater heat tolerance versus less favorable crop breeding decisions that result in crops with lower genetic gain, yield, resistance to disease, and/or heat tolerance). A state of the Markov decision process may be based on the current state of the digital twin. An action state of the Markov decision process may comprise constraints on the types of actions that automated agents (e.g., software agents) may perform (e.g., the type of breeding decisions that are possible). The transition process of the Markov process may comprise changes to the state of the digital twin based on selection policies comprising breeding actions that are performed by the automated agents based on the breeding decisions. Further, the Markov decision process may comprise reward functions to reward or penalize certain selection policies (and the subsequent breeding actions) based on the results of the breeding decisions (e.g., reward breeding decisions that result in a greater genetic gain and penalize crop breeding decisions that result in a lowering of genetic gain). As shown in FIGS. 8 and 9, a schematic representation of a reinforcement learning model involving an agent and an environment as part of the Markov Decision Process (MDP) for the agent taking action as a selector and generating the states for the digital twin representation. States and rewards at any given timepoint depend on the previous timepoint's state and agent's action. Examples of the action space include: modification of secondary trait thresholds for selection based on YIELD vs. MST (moisture) residuals; vary soft vs. hard constraints, secondary trait thresholds, diversity constraints for algorithmic selection; directly provide binary choice of select/discard for each candidate.

[0087] In some embodiments, a best linear unbiased prediction (BLUP) may be used to analyze phenotype data. Further, the BLUP may be used to generate a predicted phenotype for a crop variety based on spatial variability and/or temporal variability to generate a predicted phenotype of the variety.

[0088] FIG. 4 shows a timing diagram that indicates a data flow of signals associated with generating and implementing breeding decision data according to one or more aspects of the disclosure. Computing devices and/or systems associated with data flow 400 may include any of the features and/or components of computing system 104, crop genotyping device 130, crop monitoring device 132, environmental monitoring device 134, and/or crop planting device 136, which are shown in FIG. 1. Further, the operations performed as part of the data flow 400 may be performed by the computing systems and/or computing devices that are described herein.

[0089] Crop genotyping device 130 may acquire and/or process samples of crops in order to determine their genotype. The crop genotyping device 130 may then generate genotype data and send signal 410 comprising the genotype data to computing system 104. [0090] Crop monitoring device 132 may detect the state of crops and generate phenotype data based on the detected traits of the crops. Crop monitoring device 132 may then send a signal 414 that includes the phenotype data to computing system 104.

[0091] Environmental monitoring device 134 monitor the state of the environment and/or soil conditions around a crop. Environmental monitoring device 134 may then generate environmental data and send a signal 414 that includes the environmental data to computing system 104.

[0092] Computing system 104 may be configured to receive the signal 412, the signal 414, and/or the signal 416. In some embodiments, computing system 104 may be configured to send signals comprising a request for data (e.g., phenotype data, environmental data, and/or genotype data) to crop genotyping device 130, crop monitoring device 132, and/or environmental monitoring device 134. In response to the request for data from computing system 104, crop genotyping device 130 may send genotype data, crop monitoring device 132 may send phenotype data, and/or environmental monitoring device 134 may send environmental data.

[0093] Further, computing system 104 may receive crop breeding pipeline data which may indicate the configuration of a crop breeding pipeline for crops associated with the phenotype data, environmental data, and/or genotype data). The computing system 104 may use the phenotype data, environmental data, genotype data, and/or crop breeding pipeline data to generate a digital twin of an end-to-end crop breeding pipeline. The digital twin of the end-to- end crop breeding pipeline may comprise a simulation of an actual crop breeding pipeline (e.g., the crop breeding pipeline for the crop varieties from which the phenotype data and genotype data are derived).

[0094] In this example, computing system 104 may generate breeding decision data based on the signals 410-414 as well as the crop breeding pipeline data that comprises the digital twin of the crop breeding pipeline. Further, computing system 104 may generate breeding action data based on the breeding decision data. The breeding action data may comprise one or more breeding actions that may be performed by a crop cultivation device (e.g., a crop planting device and/or an irrigation device). Computing system 104 may then send a signal 416 (e.g., a signal including one or more breeding actions) to crop planting device 136. [0095] Based on receiving signal 416, crop planting device 136 may send signal 418 (e.g., a signal confirming receipt of the one or more breeding actions encoded in the breeding action data of signal 416) to computing system 104. Crop planting device 136 may use instructions included in the signal 416 to plant seeds of the crop variety specified in the breeding action data.

[0096] FIG. 5 shows an example flow chart for automated crop propagation, crop monitoring, and environmental monitoring according to one or more aspects of the disclosure. One or more aspects of the disclosure may be implemented by the devices described herein (e.g., the system 100 shown in FIG. 1). One or more of the steps described with respect to FIG. 5 may be omitted, performed in a different order, and/or modified. Further, one or more additional steps (e.g., the steps described with respect to FIGS. 6 and/or 7) may be added to the steps described with respect to FIG. 5.

[0097] At step 502, a plurality of crop varieties from a population of plants comprising a plurality of parent crop varieties may be acquired. The plurality of crop varieties may comprise double haploid crops and/or single seed descent crops. The plurality of crop varieties may include genome edited crops. The crop varieties may be acquired from a plurality of locations comprising a particular variety of crop (e.g., wheat). For example, a crop genotyping device (e.g., the crop monitoring device 132 described with respect to FIG. 1) may acquire crops from a plant nursery that propagates crops.

[0098] At step 504, genotype data may be generated. The genotype data may comprise an indication of a genotype of the plurality of crop varieties at each of the plurality of locations. The plurality of crop varieties may be associated with genotype data comprising an indication of a genotype of the plurality of crop varieties at each of the plurality of locations. For example, the crop genotyping device 130 may sequence the DNA of a wheat crop.

[0099] At step 504, based on use of one or more crop sensors, one or more phenotypic states based on one or more phenotypic traits of the plurality of crop varieties at each of the plurality of locations over a plurality of time intervals may be detected. For example, the crop monitoring device 132 may detect changes in the height of a com crop.

[00100] At step 506, phenotype data may be generated. The phenotype data may comprise one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over the plurality of time intervals. For example, the crop monitoring device 132 may generate phenotype data indicating changes in the height of a com crop over the course of a year.

[00101] At step 508, based on one or more environmental sensors, one or more environmental conditions at the plurality of locations. The one or more environmental conditions in a location may comprise air temperature, wind speed, wind direction, humidity, and/or precipitation (e.g., snow and/or rainfall). For example, the environmental monitoring device 134 may monitor environmental conditions such as precipitation and temperature in a location using one or more sensors.

[00102] At step 510, environmental data may be generated. The environmental data may be generated based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals. For example, the environmental monitoring device 134 may generate environmental data indicating changes in the temperature and/or precipitation in a location over a six-month period.

[00103] FIG. 6 shows an example flow chart for automated monitoring and operation of an end-to-end crop breeding pipeline according to one or more aspects of the disclosure. One or more aspects of the disclosure may be implemented by the devices described herein (e.g., the system 100 shown in FIG. 1). One or more of the steps described with respect to FIG. 6 may be omitted, performed in a different order, and/or modified. Further, one or more additional steps (e.g., the steps described with respect to FIGS. 5 and/or 7) may be added to the steps described with respect to FIG. 6.

[00104] At step 602, genotype data may be acquired. Genotype data may comprise an indication of a genotype of the plurality of crop varieties at each of the plurality of locations. For example, a computing system (e.g., computing system 104) may acquire genotype data from another computing system or computing device (e.g., a computing system used by an entity responsible for breeding crops at a location) via a network (e.g., the network 102). Further, the genotype data may comprise the information described in genotype data 124 and/or genotype data 210. For example, the genotype data may include information associated with the DNA sequence of a crop and may be used to determine genes that may be relevant to traits of the crop (e.g., yield traits). The genotype data represented herein includes genetic data associated with one or more genome edits and/or transgenic events. [00105] At step 604, phenotype data may be acquired. Phenotype data may be based on based on the one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over the plurality of time intervals. For example, the one or more phenotypic states of the plurality of crop varieties may be based on traits of the plurality of crop varieties comprising crop yield, height, biomass, root length, and/or leaf shape. For example, a computing system (e.g., computing system 104) may acquire phenotype data from a multispectral scanner (e.g., the crop monitoring device 132 described with respect to FIG. 1) via a network (e.g., the network 102). Further, the phenotype data may comprise the information and/or features described in phenotype data 118 and/or phenotype data 204.

[00106] At step 606, environmental data may be acquired. Environmental data may be based on one or more environmental conditions at each of the plurality of locations over the plurality of time intervals. For example, a computing system (e.g., computing system 104) may acquire environmental data from an environmental monitoring device (e.g., the environmental monitoring device 134 described with respect to FIG. 1) via a network (e.g., the network 102). Further, the environmental data may comprise the information described in environmental data 120 and/or environmental data 206. For example, the environmental data may include information associated with a temperature and/or precipitation in a location at which a crop is being bred.

[00107] At step 608, crop breeding pipeline data may be acquired. The crop breeding pipeline data may comprise information associated with a plurality of crop development stages associated with breeding the plurality of crop varieties. For example, a computing system (e.g., computing system 104) may acquire crop breeding pipeline data (which includes data associated with one or more genome edits and/or transgenic events) from another computing system or computing device (e.g., a computing system used by an entity responsible for breeding crops at a location) via a network (e.g., the network 102). Further, the crop breeding pipeline data may comprise the information described in crop breeding pipeline data 122 and/or crop breeding pipeline data 208. For example, the crop breeding pipeline data may include information associated with crop testing, selection, and/or breeding.

[00108] At step 610, an end-to-end breeding pipeline digital twin comprising a simulation of one or more states of the plurality of crop varieties at one or more of the plurality of crop development stages of the crop breeding pipeline may be generated. The end-to-end breeding pipeline digital twin may be generated based on the genotype data, the phenotype data, the environmental data, and/or the crop breeding pipeline data. In some embodiments, the end-to- end breeding pipeline digital twin may be updated based on changes to the actual crop breeding pipeline on which the digital twin is based. Further, the one or more states may comprise one or more future states of the plurality of crop varieties at one or more future crop development stages of the plurality of crop development stages. The one or more future states of the plurality of crop varieties may be constrained by the genotype and/or the one or more phenotypic states of the plurality of crop varieties at a preceding stage of the plurality of crop development stages.

[00109] This disclosure includes specific algorithms (e.g., the Markov decision process used to generate selection policies that is described with respect to FIG. 3) for measuring tailored parameters (e.g., the parameters of the selection policies described with respect to FIG. 3) that are incorporated into a novel digital twin simulating an end-to-end breeding pipeline. The novel combination of one or more of genotype data, phenotype data, environmental data, and/or crop breeding pipeline data (e.g., configurations of a crop breeding pipeline) are used to generate an end-to-end breeding pipeline digital twin. The digital twin may simulate one or more current states of the plurality of crop varieties at one or more of the plurality of crop development stages of the end-to-end breeding pipeline. The digital twin may execute in a computing system 104 and generate an input into a machine learning model 212 that outputs breeding decision data 214 that drives breeding actions. Some examples of breeding actions including but are not limited to: population development, performing breeding crosses, hybrid or variety testing, and/or selecting one or more combinations of the plurality of crop varieties. Moreover, the digital twin may comprise a simulation that may be updated in real-time or near real-time based on changes in the actual crop breeding pipeline on which the digital twin is based. Furthermore, the digital twin may be configured to predict future states of the crop breeding pipeline based on current states of the crop breeding pipeline. For example, the digital twin may comprise simulated genotype data based on the genotype data, simulated phenotype data based on the phenotype data, simulated environmental data based on the environmental data, or simulated crop breeding pipeline data based on the crop breeding pipeline data, and/or GEBV prediction models based on the simulated data in the digital twin.

[00110] At step 612, breeding decision data for the crop in the location over a plurality of time intervals may be generated. The breeding decision data may be based on inputting the genotype data, the phenotype data, the environmental data, and/or the crop breeding pipeline data into one or more machine learning models. The one or more machine-learning models (e.g., the one or more machine learning models 212 described with respect to FIG. 2) may be configured and/or trained to determine and/or generate breeding decision data based on input comprising the genotype data, the phenotype data, the environmental data, and/or the crop breeding pipeline. The breeding decision data may comprise one or more breeding actions that may be performed in an end-to-end breeding pipeline. For example, the one or more breeding actions may comprise population development, performing breeding crosses, hybrid testing, variety testing, and/or selecting one or more combinations of the plurality of crop varieties.

[00111] In some embodiments, the breeding decision data may be based on processing the genotype data, the phenotype data, the environmental data, and/or the crop breeding pipeline. For example, the genotype data, the phenotype data, the environmental data, and/or the crop breeding pipeline may be used as inputs to breeding decision data algorithm that performs operations on the inputs and outputs the breeding decision data. Some inputs may be weighted based on the genotype data. For example, the genotype and/or the phenotype of a crop may be used to determine the relative weighting of the environmental data and/or the crop breeding pipeline, which may then be used to determine breeding decision data. The one or more breeding actions may comprise a plurality of time intervals at which to develop a genetic variation of a crop, select certain crop varieties, evaluate traits of crop varieties, and/or distribute and multiply seeds of a crop variety.

[00112] At step 614, breeding action data may be generated. The breeding action data may comprise indications for crop cultivation devices (e.g., a crop planting device and/or an irrigation device) to cultivate one or more of the plurality of crop varieties. For example, a computing system (e.g., computing system 104 described with respect to FIG. 1) may send wireless signals to a crop planting device (e.g., the crop planting device 136 described with respect to FIG. 1) via a network (e.g., the network 102). Further, the crop planting device may be configured to plant one or more quantities of seeds of the crop variety indicated in the breeding action data.

[00113] FIG. 7 shows an example flow chart for automatically planting and irrigating crops according to one or more aspects of the disclosure. One or more aspects of the disclosure may be implemented by the devices described herein (e.g., the system 100 described with respect to FIG. 1). One or more of the steps described with respect to FIG. 7 may be omitted, performed in a different order, and/or modified. Further, one or more additional steps (e.g., the steps described with respect to FIGS. 5 and/or 6) may be added to the steps described with respect to FIG. 7.

[00114] At step 702, breeding action data may be received by a crop planting device. A planting device (e.g., the crop planting device 136 described with respect to FIG. 1) may comprise a wireless receiver that is configured to receive signals comprising breeding action data. For example, the crop planting device 136 may receive breeding action data indicating specific amounts of a particular genotype of wheat to plant at a specific location at a specific time and/or date. In some embodiments, the breeding action data may be encrypted and the crop planting device 136 may decrypt the breeding action data and then authenticate the sender of the breeding action data.

[00115] At step 704, one or more quantities of the one or more combinations of the plurality of crop varieties at one of the plurality of locations may be planted in order to develop the plurality of crop varieties. Planting of the seeds may be based on the breeding action data. For example, based on breeding action data received by the crop planting device 136, specific amounts of particular genotypes of wheat seeds may be planted at a location. In some alternate embodiments, the reference to the location of the planting may be interchanged with environmental conditions reproduced in a nursery or greenhouse where each of a plurality of locations in the nursery/greenhouse are controlled and regulated to simulate different geographic locations. For example, one compartmentalized section of a nursery may recreate the environment in a hot, humid region of the country, while another section may emulate a hot, dry region with a specific amount of rainfall at specific intervals of time. For purposes of this disclosure, any reference to a geographic location or plurality of locations should be understood to be interchangeable with an emulated environment in a controlled setting such as a nursery.

[00116] At step 706, breeding action data may be received by an irrigation device. An irrigation device (e.g., the irrigation device 138 described with respect to FIG. 1) may comprise a wireless receiver that is configured to receive signals comprising breeding action data. For example, irrigation device 138 may receive breeding action data comprising an amount of water to use to irrigate crops at one or more times.

[00117] At step 708, the plurality of crop varieties at the plurality of locations may be irrigated using quantities of water based on the breeding decision data. For example, irrigation device 138 may irrigate a plurality of crops at different locations using different quantities of water based on the type of crop indicated in the breeding action data.

[00118] FIG. 8 shows an example of a Markov decision process used in an end-to-end breeding pipeline according to one or more aspects of the disclosure. Operations of a Markov decision process used in an end-to-end crop breeding pipeline may be performed by the systems and devices described herein including the system 100 shown in FIG. 1.

[00119] The Markov decision process 800 may be used to determine the process of selecting crop genotypes/vari eties through different stages of an end-to-end breeding pipeline (e.g., the end-to-end breeding pipeline described with respect to FIG. 3). The Markov decision process may comprise an agent 802 (e.g., an automated agent that selects an action to perform) that performs an action 804 (e.g., an action comprising selecting a genotype/variety) within an environment 806 (e.g., a digital twin simulation of an end-to-end crop breeding pipeline). The action 804 may be performed to transition from a state St to a state St+1. The state St may comprise an actual state (e.g., a current state of a genotype/variety) and the state St+1 may comprise a potential state (e.g., a future genotype/variety). The action such that value of the reward 810 increases. For example, the agent 802 may perform an action 804 that may increase the reward 810 which may be associated with increasing the yield of genotype/variety within the environment 806.

[00120] FIG. 9 shows another example of a Markov decision process used in an end-to-end breeding pipeline according to one or more aspects of the disclosure. Operations of a Markov decision process used in an end-to-end crop breeding pipeline may be performed by the systems and devices described herein including the system 100 shown in FIG. 1.

[00121] The Markov decision process 900 may be used to determine the process of selecting crop genotypes/vari eties through different stages of an end-to-end breeding pipeline (e.g., the end-to-end breeding pipeline described with respect to FIG. 3). The Markov decision process may comprise an agent 902 that performs an action 904. The action 904 may be expressed as At and the probability of the agent 902 performing the action 904 at time t may be expressed as 7t(At|St). The action 902 may selected from an action space that comprises the possible actions that may be performed by the agent 902. The action space may comprise actions to select a genotype/variety based on modification of secondary trait thresholds. The secondary trait thresholds may be based on yield or MST residuals. Further, the action space may comprise actions based on soft constraints, hard constraints, secondary trait thresholds, and/or diversity constraints for algorithmic selection. Further, the action 904 may comprise a binary choice to select or discard each of the candidate genotypes/varieties. The action 904 may be performed to change the state 908. For example, state 908 may transition from state St to a state St+1 such that value of the reward 910 increases from Rt to Rt+1. For example, the agent 902 (e.g., an autonomous agent) may perform an action 904 (e.g., selecting a genotype/variety) in an environment 906 (e.g., a simulation of an end-to-end breeding pipeline) that may transition the state 908 and increase the reward 910 which may be associated with increasing the disease resistance of a genotype/variety within the environment 906.

[00122] Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims. The steps of the methods described herein are described as being performed in a particular order for the purposes of discussion. A person having ordinary skill in the art will understand that the steps of any methods discussed herein may be performed in any order and that any of the steps may be omitted, combined, and/or expanded without deviating from the scope of the present disclosure. Furthermore, the methods described herein may be performed using any manner of device, system, and/or apparatus including the computing devices, computing systems, and/or computing apparatuses that are described herein.

Claims

What is claimed is:

1. A system for developing a plurality of crop varieties in an end-to-end breeding pipeline, the system comprising: a crop genotyping device configured to: generate genotype data comprising an indication of a plurality of genotypes from a plurality of crop varieties from a population of plants comprising a plurality of parent crop varieties at a plurality of locations; a crop monitoring device configured to: detect, based on use of one or more crop sensors, one or more phenotypic states based on one or more phenotypic traits of the plurality of crop varieties at each of the plurality of locations over a plurality of time intervals; and generate phenotype data based on the one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over the plurality of time intervals; a computing device configured to generate, based on the one or more soil conditions, soil data based on the one or more soil conditions at each of the plurality of locations over the plurality of time intervals; an environmental monitoring device configured to: detect, based on one or more environmental sensors, one or more environmental conditions at the plurality of locations; and generate, based on the one or more environmental conditions, environmental data based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals; a computing device configured to: receive crop breeding pipeline data comprising a plurality of crop breeding development stages associated with the end-to-end breeding pipeline for the plurality of crop varieties, wherein the crop breeding pipeline data further comprises irrigation data associated with irrigation practices in the end-to-end breeding pipeline, fertilization data associated with fertilizer practices in the end-to-end breeding pipeline, and tillage data associated with tillage practices in the end-to-end pipeline; acquire the genotype data, the phenotype data, the soil and/or location data, and the environmental data; generate, based on the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, an end-to-end breeding pipeline digital twin comprising a simulation of one or more states of the plurality of crop varieties at one or more of the plurality of crop development stages of the end-to-end breeding pipeline, wherein the one or more states comprise one or more future states of the plurality of crop varieties at one or more future crop development stages of the plurality of crop development stages, , wherein the one or more states comprise one or more current states of the plurality of crop varieties at one or more current crop development stages of the plurality of crop development stages, and wherein the one or more simulated states comprise simulated genotype data based on the genotype data, simulated phenotype data based on the phenotype data, simulated environmental data based on the environmental data, or simulated crop breeding pipeline data based on the crop breeding pipeline data; and generate, based on inputting the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, into one or more machine learning models, breeding decision data comprising one or more breeding actions to perform with respect to one or more of the plurality of crop development stages, wherein the one or more breeding actions comprise population development, performing breeding crosses, hybrid testing, variety testing, or selecting one or more combinations of the plurality of crop varieties for advancement in the pipeline; generate, for each of the plurality of locations, based on the breeding decision data, breeding action data comprising indications for crop cultivation devices to cultivate one or more of the plurality of crop varieties, wherein the crop cultivation devices comprise a crop planting device configured to: receive the breeding action data; and plant, based on the breeding action data, one or more quantities of the one or more combinations of the plurality of crop varieties at one of the plurality of locations in order to develop the plurality of crop varieties.

2. The system of claim 1, wherein the plurality of crop varieties are doubled haploid or comprise at least one parent that is a double haploid.

3. The system of claim 1, wherein the plurality of crop varieties are of single seed descent.

4. The system of claim 1, wherein the one or more breeding actions comprise trait integration of one or more of the plurality of crop varieties.

5. The system of claim 1, wherein the one or more machine learning models are configured using reinforcement learning, and wherein the reinforcement learning comprises maximization of total reward within a Markov decision process (MDP) in which selection policies performed by automated agents acting within the end-to-end breeding pipeline digital twin are evaluated, and the selection policies that improve the genetic gain of the plurality of crop varieties are rewarded.

6. The system of claim 5, wherein the selection policies comprise selecting one or more of the plurality of crop varieties for development or breeding crosses.

7. The system of claim 1, wherein the one or more future states of the plurality of crop varieties are constrained by the genotype or the one or more phenotypic states of the plurality of crop varieties at a preceding stage of the plurality of crop development stages.

8. The system of claim 1, wherein the plurality of crop varieties comprise a genome modification selected from the group consisting of one or more genome edits, one or more transgenes, one or more native allelic variation, and a combination thereof.

9. The system of claim 1, wherein the genotype data includes genetic data associated with one or more genome edits, one or more transgenes, one or more native allelic variation, or a combination thereof.

10. The system of claim 1, wherein the one or more machine learning models are configured using a deep learning model, and wherein the deep learning comprises maximization of total reward based on a transformer neural network.

11. A method of operating an end-to-end crop breeding pipeline, the method comprising: acquiring, by a computing device comprising one or more processors, genotype data comprising an indication of a genotype of plurality of crop varieties at each of a plurality of locations; acquiring, by the computing device, phenotype data comprising one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over a plurality of time intervals; acquiring, by the computing device, based on one or more soil conditions, soil data based on one or more soil conditions at each of the plurality of locations over the plurality of time intervals; acquiring, by the computing device, based on one or more environmental conditions, environmental data based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals; acquiring, by the computing device, crop breeding pipeline data comprising a plurality of crop development stages associated with breeding the plurality of crop varieties; generating, by the computing device, based on the genotype data, the phenotype data, the soil data, the environmental data, and the crop breeding pipeline data, a digital twin comprising a simulation of one or more states of the plurality of crop varieties at one or more of the plurality of crop development stages of the end-to-end crop breeding pipeline, wherein the one or more states comprise one or more future states of the plurality of crop varieties at one or more future crop development stages of the plurality of crop development stages, wherein the one or more states comprise one or more current states of the plurality of crop varieties at one or more current crop development stages of the plurality of crop development stages, and wherein the one or more states comprise simulated genotype data based on the genotype data, simulated phenotype data based on the phenotype data, simulated environmental data based on the environmental data, or simulated crop breeding pipeline data based on the crop breeding pipeline data; and generating, by the computing device, based on inputting the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, into one or more machine learning models, breeding decision data comprising one or more breeding actions to perform with respect to one or more of the plurality of crop development stages, wherein the one or more breeding actions comprise population development, performing breeding crosses, hybrid testing, variety testing, or selecting one or more combinations of the plurality of crop varieties.

12. The method of claim 11, wherein the one or more machine learning models comprise a cross breeding model configured to determine the one or more combinations of the plurality of crop varieties to cross-breed at each of the plurality of locations.

13. The method of claim 11, wherein the one or more machine learning models comprise a convolutional neural network, a reinforcement learning model, a densely-connected feedforward neural network, a transformer-based neural network, a normalizing flow, an attention-based neural network, a decision tree, a forest of decision trees, a support vector machine, a Gaussian process, a reproducing kernel Hilbert space model, a generalized additive model, or a generalized linear model.

14. The method of claim 11, wherein the environmental data comprises one or more states of weather at the plurality of locations, wherein the one or more states of the weather comprise historical precipitation for the plurality of locations, historical amounts of humidity for the plurality of locations, or historical temperatures for the plurality of locations.

15. The method of claim 11, wherein the environmental data comprise one or more states of soil at the plurality of locations, one or more amounts of sunlight at the plurality of locations, or one or more altitudes of the plurality of locations.

16. The method of claim 11, wherein the phenotype data and the environmental data are updated in real-time.

17. The method of claim 11, wherein the plurality of development stages comprise a genome editing stage, a transformation stage, trait introgression stage, a crop population development stage, a crop selection stage, a hybrid creation stage, and a crop yield testing stage.

18. The method of claim 11, wherein the breeding decision data comprises one or more indications of one or more future time intervals at which to perform the one or more breeding actions.

19. One or more non-transitory computer readable media comprising instructions that, when executed by at least one processor, cause a computing device to perform operations comprising: acquiring genotype data comprising an indication of a genotype of plurality of crop varieties at each of a plurality of locations; acquiring phenotype data comprising one or more phenotypic states of the plurality of crop varieties detected at each of the plurality of locations over a plurality of time intervals; acquiring, based on the one or more environmental conditions, environmental data based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals; acquiring crop breeding pipeline data comprising a plurality of crop development stages associated with breeding the plurality of crop varieties; generating, based on the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, a digital twin comprising a simulation of one or more states of the plurality of crop varieties at one or more of the plurality of crop development stages of the crop breeding pipeline, wherein the one or more states comprise one or more future states of the plurality of crop varieties at one or more future crop development stages of the plurality of crop development stages, wherein the one or more states comprise one or more current states of the plurality of crop varieties at one or more current crop development stages of the plurality of crop development stages, and wherein the one or more states comprise simulated genotype data based on the genotype data, simulated phenotype data based on the phenotype data, simulated environmental data based on the environmental data, or simulated crop breeding pipeline data based on the crop breeding pipeline data; and generating, based on inputting the genotype data, the phenotype data, the environmental data, and the crop breeding pipeline data, into one or more machine learning models, breeding decision data comprising one or more breeding actions to perform with respect to one or more of the plurality of crop development stages, wherein the one or more breeding actions comprise population development, performing breeding crosses, hybrid or variety testing, or selecting one or more combinations of the plurality of crop varieties.

20. The one or more non-transitory computer readable media of claim 19, wherein the one or more machine learning models are configured using reinforcement learning, and wherein the reinforcement learning comprises maximization of total reward within a Markov decision process (MDP) in which selection policies performed by automated agents acting within the end-to-end breeding pipeline digital twin are evaluated, and the selection policies that improve the genetic gain of the plurality of crop varieties are rewarded.

21. The one or more non-transitory computer readable media of claim 19, wherein the one or more machine learning models comprise a cross breeding model configured to determine the one or more combinations of the plurality of crop varieties to cross-breed at each of the plurality of locations.

22. The one or more non-transitory computer readable media of claim 19, wherein the plurality of development stages comprise a crop population development stage, a crop selection stage, a hybrid creation stage, and a crop yield testing stage.

23. The one or more non-transitory computer readable media of claim 19, wherein the breeding decision data comprise one or more indications of one or more future time intervals at which to perform the one or more breeding actions.

24. A method of operating an end-to-end com breeding pipeline, the method comprising: acquiring, by a computing device comprising one or more processors, genotype data comprising an indication of a genotype of plurality of corn varieties comprising inbreds and/or hybrids at each of a plurality of locations; acquiring, by the computing device, phenotype data comprising one or more phenotypic states of the plurality of corn varieties detected at each of the plurality of locations over a plurality of time intervals; acquiring, by the computing device, location data at each of the plurality of locations over the plurality of time intervals, wherein the location data includes soil type; acquiring, by the computing device, based on one or more environmental conditions, environmental data based on the one or more environmental conditions at each of the plurality of locations over the plurality of time intervals; acquiring, by the computing device, com breeding pipeline data comprising a plurality of corn development stages associated with breeding the plurality of corn varieties; generating, by the computing device, based on the genotype data, the phenotype data, the location data, the environmental data, and the crop breeding pipeline data, a digital twin comprising a simulation of one or more states of the plurality of com inbreds and/or hybrids at one or more of the plurality of com development stages of the end-to-end com breeding pipeline, wherein the one or more states comprise one or more future states of the plurality of corn varieties at one or more future com development stages of the plurality of corn development stages, wherein the one or more states comprise one or more current states of the plurality of com varieties at one or more current crop development stages of the plurality of corn development stages, and wherein the one or more states comprise simulated genotype data based on the genotype data, simulated phenotype data based on the phenotype data, simulated environmental data based on the environmental data, or simulated corn breeding pipeline data based on the com breeding pipeline data; and generating, by the computing device, based on inputting the genotype data, the phenotype data, the environmental data, and the corn breeding pipeline data, into one or more machine learning models, breeding decision data comprising one or more breeding actions to perform with respect to one or more of the plurality of com development stages, wherein the one or more breeding actions comprise population development, performing breeding crosses, hybrid testing, inbred testing, or selecting one or more combinations of the plurality of corn varieties.

25. The method of claim 24, wherein the one or more machine learning models comprise a cross breeding model configured to determine the one or more combinations of the plurality of corn inbred populations to cross-breed at each of the plurality of locations.

26. The method of claim 24, wherein the one or more machine learning models comprise a convolutional neural network, a reinforcement learning model, a densely-connected feedforward neural network, a transformer-based neural network, a normalizing flow, an attention-based neural network, a decision tree, a forest of decision trees, a support vector machine, a Gaussian process, a reproducing kernel Hilbert space model, a generalized additive model, or a generalized linear model.

27. The method of claim 24, wherein the environmental data comprises one or more states of weather at the plurality of locations, wherein the one or more states of the weather comprise historical precipitation for the plurality of locations, historical amounts of humidity for the plurality of locations, or historical temperatures for the plurality of locations.

28. The method of claim 24, wherein the environmental data comprise one or more states of soil at the plurality of locations, one or more amounts of sunlight at the plurality of locations, or one or more altitudes of the plurality of locations.

29. The method of claim 24, wherein the phenotype data and the environmental data are updated in real-time.

30. The method of claim 24, wherein the plurality of development stages comprise a genome editing stage, a transformation stage, trait introgression stage, a com population development stage, an inbred selection stage, a hybrid creation stage, and a crop yield testing stage.

31. The method of claim 24, wherein the breeding decision data comprises one or more indications of one or more future time intervals at which to perform the one or more breeding actions.