Intelligent breeding decision method for new peach varieties based on multi-omics data and molecular markers

Abstract

The application discloses a peach new variety intelligent breeding decision-making method based on multi-omics data and molecular markers, and belongs to the technical field of agricultural bioinformatics. By acquiring a breeding target description text and analyzing breeding target parameters, a decision-making logic tree containing multi-level trait nodes is constructed to determine breeding complexity, and the marker association strength of each germplasm resource is obtained. The grouping granularity of the germplasm resource data of the marker combination and the marker interval are determined. A mapping relationship between the trait nodes and the peach reference genome physical position is established, the preliminary genomic segment distribution is determined, and the target segment distribution is obtained through screening. The haplotype logical recombination is performed on the germplasm resource data corresponding to the target segment, and the genomic feature map is constructed. The molecular marker data stream is extracted, and the batch data volume of single data access is adjusted according to the decision-making time sequence of the breeding task. The application realizes the collaborative optimization of the breeding decision-making logic and the multi-omics data storage access mode, and can significantly improve the decision-making efficiency of the peach new variety breeding process.

Description

Technical Field

[0001] This invention discloses an intelligent breeding decision-making method for new peach varieties based on multi-omics data and molecular markers, belonging to the field of agricultural bioinformatics technology. Background Technology

[0002] Peaches, as an important temperate fruit, have a wide economic value and consumer market globally. Traditional peach breeding relies primarily on breeders' experience, involving artificial hybridization and progeny selection based on field phenotypic traits. This method is time-consuming and inefficient, typically requiring over ten years to obtain a stable new variety. With the rapid development of molecular biology techniques, marker-assisted breeding has gradually become an important means of peach variety improvement. By utilizing molecular markers closely linked to target traits for early screening of breeding materials, the breeding cycle can be shortened to some extent. However, many important agronomic traits of peaches, such as fruit firmness, soluble solids content, and disease resistance, are controlled by multiple genes, and complex epistatic and additive effects exist between genes. Single marker-assisted selection often struggles to accurately predict the overall performance of progeny, resulting in a need to improve the accuracy and efficiency of breeding decisions.

[0003] In recent years, with the rapid development of high-throughput sequencing technology and bioinformatics, peach genomics research has achieved significant breakthroughs, accumulating massive amounts of genome resequencing data, transcriptome data, and corresponding refined phenotypic data. This multi-omics data provides an unprecedented opportunity to deeply analyze the genetic basis of important agronomic traits in peaches. However, existing data processing methods typically involve simple overlaying or correlation analysis of these multi-omics data, lacking a technical means to effectively integrate complex decision-making logic with underlying data storage and access patterns. When faced with specific breeding goals, breeders often need to repeatedly consult different databases and utilize different analytical tools to integrate information that guides breeding decisions. This discrete data processing flow is not only cumbersome but also leads to low utilization of computational resources due to the randomness and repetitiveness of data access. The data input-output process becomes a bottleneck restricting decision-making efficiency, failing to meet the demands of modern breeding for high throughput, high precision, and rapid response.

[0004] To address the aforementioned issues, a method is needed to develop a structured expression of the decision-making process of peach breeding experts, thereby optimizing the organization and access patterns of multi-omics data. This method transforms breeding objectives into a computable decision logic structure and deeply analyzes the germplasm resource utilization patterns contained in historical breeding data, enabling intelligent screening and efficient extraction of massive germplasm resource data. The method should be able to dynamically adjust the granularity of data processing according to the complexity of the breeding task, establish a bridge between decision logic and genomic physical location, and achieve synergistic optimization with decision-making timing at the data access level. This will significantly improve data processing throughput and response speed while ensuring the accuracy of breeding decisions, providing strong technical support for the targeted breeding of new peach varieties. Summary of the Invention

[0005] According to a first aspect of the present invention, the present invention claims protection for an intelligent breeding decision-making method for new peach varieties based on multi-omics data and molecular markers, comprising: S1, obtain the description text of the breeding target trait corresponding to the target breeding task, and parse the breeding target trait description text to obtain the breeding target parameters, and obtain the historical call records of germplasm resource data associated with the breeding target parameters in the peach germplasm resource bank within the historical breeding cycle; S2, construct a decision logic tree containing multiple trait nodes according to the breeding target parameters, extract the node topology features of each trait node from the tree structure of the decision logic tree, determine the breeding complexity corresponding to the breeding target parameters, and use a hidden Markov model to perform state transition analysis on the historical call records to obtain the label association strength corresponding to each germplasm resource data in the peach germplasm resource bank. S3, based on the breeding complexity and the marker association strength, determine the grouping granularity for tagging all germplasm resource data in the peach germplasm resource bank and the marker spacing between adjacent groups; S4, establish a mapping relationship between each trait node in the decision logic tree and the physical location of the peach reference genome, determine the preliminary genome segment distribution based on the mapping relationship, and filter the preliminary genome segment distribution based on the marker association strength to obtain the target genome segment distribution; S5, perform haplotype logical recombination on the germplasm resource data corresponding to the distribution of the target genome segment according to the grouping granularity, and construct a genome feature map that matches the topology of the decision logic tree; S6. Based on the genome feature map and the marker spacing, the corresponding molecular marker data stream is extracted in parallel from the peach germplasm resource bank, and the batch processing data volume of a single data access is adjusted according to the decision sequence of the target breeding task in the intelligent breeding process, so as to complete the accelerated decision-making of the peach new variety breeding process.

[0006] Further, S1 includes: The system receives target trait description text for the selection and breeding of new peach varieties from breeding experts, identifies trait keywords in the target trait description text through natural language processing technology, and performs semantic matching of the trait keywords with a preset trait ontology library to obtain breeding target parameters used to characterize specific breeding directions. Based on the breeding target parameters, a search is conducted in the metadata index of the peach germplasm resource bank to determine the unique identifier of at least one germplasm resource data that is associated with the breeding target parameters; Based on the unique identifier, a backtracking query is performed in the access log system of the peach germplasm resource bank to extract the timestamp, call frequency, and breeding project type associated with each germplasm resource data in the historical breeding cycle. The timestamp, call frequency, and breeding project type are aggregated and encapsulated to form a historical call record corresponding to each germplasm resource data.

[0007] Further, S2 includes: The breeding target parameters are used as the root node of the decision logic tree. The first-level trait sub-nodes under the root node are determined according to the trait determination relationship in the peach variety breeding process. For each first-level trait sub-node, the second-level trait sub-nodes are obtained by recursively expanding according to the epistatic effect and additive effect relationship in the peach genetic law, until all sub-nodes can correspond to specific molecular markers or germplasm resource entities, forming a multi-layer tree structure containing root nodes, intermediate nodes and leaf nodes, which serves as the decision logic tree. For each trait node in the decision logic tree, the level depth of the trait node in the tree structure and the number of lower branches directly connected to the trait node are identified. The level depth and the number of branches are combined to form the node topology feature of the trait node. The node topology features of all trait nodes are weighted and aggregated to obtain the breeding complexity, which is used to characterize the total amount of information dimensions that the breeding target parameters need to be processed in the breeding decision process. Extract the time series data of each germplasm resource data being called from the historical call records. Define each call event in the time series data as an observable state in the Hidden Markov Model (HMM), and define the genetic similarity relationship between germplasm resource data as a latent state in the HMM. By calculating the transition probability matrix between latent states and the output probability matrix of the observable state to the latent state, perform forward-backward iterative training on the time series data to obtain converged HMM parameters. Use the converged HMM parameters to calculate the posterior probability of each germplasm resource data being selected again under the current breeding objective, and use the posterior probability as the label association strength corresponding to that germplasm resource data.

[0008] Further, S3 includes: The breeding complexity is mapped to a threshold for the number of blocks when performing haplotype block segmentation on germplasm resource data. The maximum number of molecular markers that each haplotype block can contain is determined based on the threshold for the number of blocks. The maximum number of molecular markers is used as the grouping granularity for combining markers in germplasm resource data. The grouping granularity determines the upper limit of the number of markers processed in each combination operation when extracting molecular marker data from the peach germplasm resource bank. All germplasm resource data in the peach germplasm resource bank are sorted in descending order according to the marker association strength to obtain a germplasm resource sequence sorted from high to low marker association strength. The intensity transition positions where the marker association strength changes significantly are identified from the germplasm resource sequence. The germplasm resource sequence is divided into at least two intensity intervals according to the intensity transition positions. The physical distance or genetic distance between adjacent intensity intervals is measured and the physical distance or genetic distance is used as the marker spacing between adjacent groups. The germplasm resource data within each intensity interval is divided into haplotype blocks according to the grouping granularity, such that the number of molecular markers in each haplotype block does not exceed the maximum number of molecular markers, and the address offset between adjacent haplotype blocks is determined according to the marker spacing to ensure that data in different haplotype blocks will not conflict during parallel access.

[0009] Further, S4 includes: For each trait node in the decision logic tree, the annotation file of the peach reference genome is queried to determine the starting and ending physical coordinates of the functional gene associated with the trait corresponding to the trait node on the peach reference genome. The continuous interval between the starting and ending physical coordinates is taken as the candidate genome segment associated with the trait node, and the correspondence between the trait node and the candidate genome segment is taken as the mapping relationship. Collect all candidate genome regions corresponding to all trait nodes in the decision logic tree, merge and remove duplicates of the physical locations of all candidate genome regions on the peach reference genome, and obtain a preliminary genome region distribution composed of non-overlapping genome regions. For each genomic region in the preliminary genomic region distribution, the average value of the marker association strength of all molecular markers falling within the genomic region is calculated. This average value is used as the region association score of the genomic region. The region association score is compared with a preset screening threshold. Genomic regions with region association scores greater than or equal to the preset screening threshold are retained, while genomic regions with region association scores less than the preset screening threshold are removed. All the retained genomic regions are used as the target genomic region distribution.

[0010] Further, S5 includes: S51, determine the maximum number of markers to group all molecular markers within the target genome segment distribution according to the grouping granularity, divide the target genome segment distribution into multiple consecutive haplotype blocks according to the maximum number of markers, and assign a unique block index to each haplotype block; S52, extract the logical arrangement order of all leaf nodes in the decision logic tree. The logical arrangement order is determined by the breadth-first traversal order from the root node to the leaf node. Reorder the multiple haplotype blocks according to the logical arrangement order so that the reordered haplotype block sequence corresponds one-to-one with the leaf node sequence of the decision logic tree. S53, for each reordered haplotype block, extract allelic variation information of all molecular markers within the haplotype block, and arrange the allelic variation information according to the position of the haplotype block in the haplotype block sequence to form a two-dimensional matrix structure, wherein the rows of the matrix correspond to different germplasm resources and the columns of the matrix correspond to molecular markers within the haplotype block, and the two-dimensional matrix structure is the genome feature map.

[0011] Further, S53 includes: For each haplotype block after reordering, the corresponding physical interval is located in the distribution of the target genome segment according to the block index of the haplotype block. The genotype data of all germplasm resource materials corresponding to the physical interval are read from the peach germplasm resource bank. The allelic variation type of each molecular marker locus is extracted from the genotype data and the allelic variation type is converted into binary code or character code. Based on the positional order of the haplotype blocks in the haplotype block sequence, the column start position and column end position of the haplotype block in the two-dimensional matrix structure are determined, and the allelic variant codes of all extracted molecular marker sites are sequentially filled into the corresponding columns of the two-dimensional matrix structure according to the column start position and column end position. Traverse all the reordered haplotype blocks and fill the allelic variant codes of each haplotype block into the corresponding positions of the two-dimensional matrix structure until all haplotype blocks have been processed, forming a two-dimensional matrix with the number of rows equal to the number of germplasm resources and the number of columns equal to the total number of molecular markers in all haplotype blocks. Use this two-dimensional matrix as the genome feature map.

[0012] Further, S6 includes: Based on the molecular marker sites corresponding to each column in the genome feature map, the physical address of the molecular marker site on the peach reference genome is determined. The address difference between two adjacent molecular marker sites is calculated based on the marker spacing. The address difference is used as the address jump step size when continuously reading molecular marker data from the peach germplasm resource bank. Based on the germplasm resource material corresponding to each row in the genome feature map, a data reading instruction corresponding to that germplasm resource material is generated. The data reading instruction contains the physical addresses of all molecular marker sites corresponding to the germplasm resource material and the address jump step size between adjacent physical addresses. The data reading instructions for different germplasm resource materials are sent in parallel to the storage controller of the peach germplasm resource bank. The storage controller reads the molecular marker data of multiple germplasm resource materials in parallel from the storage medium according to the address jump step size and merges them to form the molecular marker data stream. The decision sequence of the target breeding task in the intelligent breeding process is analyzed. The decision sequence includes decision stages and the start and end times of each decision stage. The available decision duration is calculated based on the time difference between the end time and the start time of the current decision stage. The maximum amount of data that can be transmitted from the peach germplasm resource bank within the current decision stage is calculated based on the data output bandwidth of the storage controller and the available decision duration. The molecular marker data stream is divided into multiple data blocks according to the row structure of the genome feature map. Each data block corresponds to a complete set of molecular marker information for a germplasm resource material. The number of data blocks that can be included in a single data access is determined based on the maximum data volume. The number of data blocks is used as the batch processing data volume. The storage controller transmits the molecular marker data stream to the intelligent breeding decision module in batches according to the batch processing data volume.

[0013] Furthermore, following S6, it also includes: The molecular marker data stream is input into a pre-configured peach variety breeding decision engine. The peach variety breeding decision engine performs parallel decoding of the molecular marker data stream according to the marker combination relationship defined by the genome feature map, and restores the haplotype combination of each germplasm resource material on the target genome segment. The peach variety breeding decision engine compares the restored haplotype combination with the preset ideal haplotype template, calculates the genetic similarity between each germplasm resource material and the ideal haplotype template, and sorts all germplasm resource materials according to the genetic similarity to generate a list of recommended parent materials. The recommended parent material list is linked with the field phenotypic data corresponding to each germplasm resource material, so that breeding experts can determine the final hybridization combination scheme based on the recommended parent material list and the field phenotypic data.

[0014] Furthermore, following S3, it also includes: The grouping granularity and the tag spacing are encoded into metadata tags, and the metadata tags are associated with the breeding target parameters and stored to establish a correspondence library between the breeding target parameters and the data access mode; When a new breeding target task is received, the corresponding relation database is searched to see if there is a historical grouping granularity and historical marker spacing that match the new breeding target parameters. If so, the historical grouping granularity and historical marker spacing are directly called to construct a genome feature map to accelerate the response speed to the new breeding target task.

[0015] This invention discloses an intelligent decision-making method for peach breeding based on multi-omics data and molecular markers, belonging to the field of agricultural bioinformatics. By acquiring the breeding target description text and parsing the breeding target parameters, a decision logic tree containing multi-level trait nodes is constructed to determine the breeding complexity and obtain the marker association strength for each germplasm resource. The grouping granularity and marker spacing of the germplasm resource data for marker combinations are determined. A mapping relationship between trait nodes and the physical location of the peach reference genome is established to determine the preliminary genomic segment distribution and screen to obtain the target segment distribution. Haplotype logical recombination is performed on the germplasm resource data corresponding to the target segments to construct a genome feature map. Molecular marker data streams are extracted, and the batch processing data volume for a single data access is adjusted according to the decision-making timeline of the breeding task. This invention achieves synergistic optimization of the breeding decision logic and the multi-omics data storage and access mode, significantly improving the decision-making efficiency of the peach breeding process. Attached Figure Description

[0016] Figure 1 This is a flowchart of the intelligent breeding decision-making method for new peach varieties based on multi-omics data and molecular markers, as claimed in the embodiments of the present invention. Figure 2 This is a second flowchart of the intelligent breeding decision-making method for new peach varieties based on multi-omics data and molecular markers, as claimed in the embodiments of the present invention. Figure 3 The third flowchart of a smart breeding decision-making method for new peach varieties based on multi-omics data and molecular markers, which is claimed in the embodiments of the present invention, is shown. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] The technical solutions disclosed in the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0019] According to the first embodiment of the present invention, referring to Figure 1 This invention claims protection for an intelligent breeding decision-making method for new peach varieties based on multi-omics data and molecular markers, comprising: S1, obtain the description text of the breeding target traits corresponding to the target breeding task, and parse the breeding target parameters based on the description text of the breeding target traits, and obtain the historical call records of germplasm resource data related to the breeding target parameters in the peach germplasm resource bank within the historical breeding cycle; S2. Construct a decision logic tree containing multiple trait nodes based on the breeding target parameters, and extract the node topology features of each trait node from the tree structure of the decision logic tree to determine the breeding complexity corresponding to the breeding target parameters. Use a hidden Markov model to perform state transition analysis on the historical call records to obtain the label association strength corresponding to each germplasm resource data in the peach germplasm resource bank. S3. Based on breeding complexity and marker association strength, determine the grouping granularity for tagging all germplasm resource data in the peach germplasm resource bank and the marker spacing between adjacent groups. S4. Establish the mapping relationship between each trait node in the decision logic tree and the physical location of the peach reference genome. Determine the preliminary genome segment distribution based on the mapping relationship, and filter the preliminary genome segment distribution based on the marker association strength to obtain the target genome segment distribution. S5. According to the grouping granularity, perform haplotype logical recombination on the germplasm resource data corresponding to the distribution of the target genome segments to construct a genome feature map that matches the topology of the decision logic tree; S6, based on genome feature maps and marker spacing, extracts corresponding molecular marker data streams in parallel from the peach germplasm resource bank, and adjusts the batch data volume of a single data access according to the decision sequence of the target breeding task in the intelligent breeding process, so as to accelerate the decision-making process of new peach variety breeding.

[0020] In this embodiment, the system obtains the description text of the breeding target trait corresponding to the target breeding task. This description text is input by a breeding expert through a human-computer interaction interface. The text content describes the trait characteristics that the new peach variety to be bred should possess in natural language form, such as high fruit firmness, high soluble solids content, and resistance to brown rot. After receiving the text, the system performs word segmentation processing on the text using a word segmentation tool in natural language processing technology to extract keywords from the text. These keywords are then semantically matched with a preset trait ontology database. The trait ontology database pre-stores standardized trait names and their hierarchical relationships in the field of peach breeding. Through semantic similarity calculation, the target trait is selected from the target trait. The system matches the closest standard trait terms from the germplasm bank as breeding target parameters. Simultaneously, it retrieves access logs from the peach germplasm resource bank to obtain access records of germplasm resource data associated with these breeding target parameters within historical breeding cycles. Specifically, the system searches the metadata index of the germplasm resource bank based on the breeding target parameters to determine the unique identifiers of the germplasm resources associated with these parameters. Then, it backtracks to query the occurrence of these identifiers in the historical access logs, extracting information such as the timestamp of each access, the identity of the accessor, and the name of the breeding project associated with the access. This information is then aggregated according to the germplasm resource identifier to form the historical access record for each germplasm resource.

[0021] Based on the breeding target parameters, a decision logic tree containing multiple trait nodes is constructed. The system first uses the breeding target parameters as the root node. Then, according to the known trait determination relationships in peach genetics and breeding, such as certain traits being determined by combinations of other traits, the root node is decomposed into several first-level trait sub-nodes. For example, fruit quality is decomposed into firmness, sugar content, acidity, etc. For each first-level trait sub-node, the system further consults a pre-constructed trait-gene association knowledge base. Based on the epistatic and additive effects of genes, the sub-node is recursively decomposed into more fine-grained second-level trait sub-nodes until all sub-nodes can be directly mapped to specific molecular markers or germplasm resource entities. Finally, a multi-layered tree structure with root nodes, intermediate nodes, and leaf nodes is formed, i.e., the decision logic tree. The system extracts the node topology features of each trait node from the tree structure, including the level depth of the node and the number of lower-level branches directly connected to the node. These two values ​​are combined as the topology feature vector of the node. The topological feature vectors of all nodes are weighted and aggregated, with the weights preset according to the importance of the nodes in the tree. The result of the weighted sum is the breeding complexity, which reflects the total amount of information dimensions that the breeding target parameters need to process in the breeding decision-making process.

[0022] Simultaneously, a Hidden Markov Model (HMM) is used to perform state transition analysis on historical call records. The system transforms the historical call records of each germplasm resource data into a time series, with each call event considered as an observable state. The genetic similarity relationship between germplasm resources is defined as a latent state, and the number of latent states is predetermined through cluster analysis. The Baum-Welch algorithm is used to iteratively train the parameters of the HMM, calculating the transition probability matrix between latent states and the output probability matrix of observable states to latent states. After training convergence, for each germplasm resource, based on its state in the most recent call event, the Viterbi algorithm is used to infer its most likely current latent state. Then, based on the output probability matrix, the probability that the germplasm resource will be observed again in the current latent state is calculated. This probability value is the label association strength of the germplasm resource.

[0023] Based on breeding complexity and marker association strength, the grouping granularity for marker combination of all germplasm resource data in the peach germplasm resource bank and the marker spacing between adjacent groups are determined. The breeding complexity value is converted into a block number threshold through a preset mapping table. This threshold specifies the maximum number of molecular markers that each block can contain when performing haplotype block segmentation; this maximum number is the grouping granularity. All germplasm resources are sorted from high to low according to marker association strength to form sequences. The difference in marker association strength between adjacent germplasm resources in the sequence is calculated. When the difference exceeds a preset transition threshold, it is identified as a strength transition position. Based on these transition positions, the sequence is divided into multiple strength intervals, and the distance between adjacent strength intervals on the genetic or physical map is measured; this distance is used as the marker spacing.

[0024] A mapping relationship is established between each trait node in the decision logic tree and its physical location in the peach reference genome. For each trait node in the decision logic tree, the system queries the annotation file of the peach reference genome for functional genes or QTL segments associated with that trait, obtains its start and end physical coordinates, and uses this coordinate interval as the associated genomic segment of that node. The associated segments of all nodes are merged and deduplicated to obtain a preliminary genomic segment distribution. For each segment, the average value of the marker association strength of all molecular markers falling within that segment is calculated as the interval association score for that segment. The interval association score is compared with a preset screening threshold, and segments with scores higher than the threshold are retained, while segments with scores lower than the threshold are removed, thus obtaining the target genomic segment distribution.

[0025] The system performs haplotype logical recombination on the germplasm resource data corresponding to the distribution of target genomic regions according to the grouping granularity. The system divides the target genomic region distribution into multiple consecutive haplotype blocks based on the grouping granularity, i.e., the maximum number of markers in each haplotype block. Each block is assigned a unique index, and the logical order of all leaf nodes in the decision logic tree is extracted. This order is determined by a breadth-first traversal from the root node to the leaf node. All haplotype blocks are rearranged according to the order of the leaf nodes, ensuring a one-to-one correspondence between the recombinated block sequences and the leaf node sequences. For each rearranged block, allelic variation information of all molecular markers within that block is extracted. This information is arranged into a two-dimensional matrix structure according to the block position. The rows of the matrix correspond to different germplasm resource materials, and the columns correspond to molecular marker loci within the block. This two-dimensional matrix is ​​the genomic feature map.

[0026] Based on the genome feature map and marker spacing, the system extracts corresponding molecular marker data streams in parallel from the peach germplasm resource bank. The system calculates the jump step size between adjacent addresses when continuously reading data from the germplasm resource bank, based on the physical addresses of the molecular marker loci on the reference genome for each column of the genome feature map and the marker spacing. For each row of germplasm resources in the map, the system generates a data reading instruction containing the physical addresses of all marker loci and the jump step size for that material, and sends the instructions for different materials in parallel to the storage controller. The storage controller reads the data in parallel according to the instructions, merges them to form a molecular marker data stream, and the system analyzes the decision-making timeline of the target breeding task in the intelligent breeding process to obtain the start and end times of the current decision stage and calculate the available decision time. Based on the data output bandwidth of the storage controller and the available decision time, the system calculates the maximum amount of data that can be transmitted in the current stage. The molecular marker data stream is divided into multiple data blocks according to the row structure of the map. Each data block corresponds to the complete marker information of a germplasm resource. The number of data blocks that can be included in a single data access is determined based on the maximum data volume. The storage controller transmits the data stream to the decision module in batches according to this batch data volume to accelerate the decision-making process for the breeding of new peach varieties.

[0027] Furthermore, referring to Figure 2 S1 includes: The system receives text describing the target traits for the breeding of new peach varieties from breeding experts, identifies the trait keywords in the text using natural language processing technology, and performs semantic matching between the trait keywords and a pre-set trait ontology database to obtain breeding target parameters that characterize the specific breeding direction. Based on the breeding target parameters, a search is conducted in the metadata index of the peach germplasm resource bank to determine the unique identifier of at least one germplasm resource data that is associated with the breeding target parameters; Based on the unique identifier, a backtracking query is performed in the access log system of the peach germplasm resource bank to extract the timestamp, call frequency, and breeding project type associated with each germplasm resource data in the historical breeding cycle. The timestamp, call frequency, and breeding project type are aggregated and encapsulated to form a historical call record corresponding to each germplasm resource data.

[0028] In this embodiment, the system receives text describing the target traits for new peach varieties, input by breeding experts. This text can be input via a text box in a graphical user interface. Experts describe the desired characteristics of the peach variety in natural language, such as early fruit ripening, vibrant fruit color, non-melting flesh, and resistance to peach aphids. The system has a built-in natural language processing module. This module first preprocesses the input text, including removing meaningless punctuation and converting uppercase letters to lowercase. Then, it uses a dictionary-based word segmentation algorithm to segment the text, dividing continuous strings into independent word units. During word segmentation, the system calls a specialized dictionary built specifically for peach breeding, which includes all standardized trait names, common aliases, and variety names to ensure accuracy. After word segmentation, a series of candidate keywords are obtained. Next, the system performs semantic matching: comparing each candidate keyword with concepts in a pre-defined trait ontology database. The trait ontology is a hierarchical knowledge system. For example, fruit traits are divided into external quality and internal quality, and external quality is further divided into fruit shape, fruit color, and fuzz density. The semantic matching process is achieved by calculating the semantic similarity between candidate keywords and ontology concepts. Similarity calculation is based on methods such as word vectors or edit distance. For each candidate keyword, the system selects the ontology concept with the highest similarity as the matching result. If the similarity is lower than a preset threshold, the keyword is considered an unrecognized word and is manually mapped by experts. All successfully matched ontology concepts constitute the breeding target parameters, and each parameter corresponds to a standardized trait identifier.

[0029] The system searches the metadata index of the peach germplasm resource bank based on breeding target parameters. The peach germplasm resource bank is a database storing thousands of peach germplasm resources, each associated with detailed metadata, including germplasm name, origin, conservation institution, and phenotypic data obtained through multi-year, multi-location observations. The metadata index is a fast lookup structure built upon this metadata, such as an inverted index. The system uses breeding target parameters as query conditions to search for germplasm resources with these phenotypic traits in the metadata index. For example, if the breeding target parameter includes high flesh firmness, the index will return unique identifiers for all germplasm resources that record high firmness for this trait. The search result is a list of germplasm resource identifiers containing all germplasm resources that are directly or indirectly associated with the breeding target parameter.

[0030] A retrospective query is performed on the access log system of the peach germplasm resource bank based on unique identifiers. The access log system records every access operation of all users to the germplasm resource bank. Each log entry contains an access timestamp, the visitor's user ID, the accessed germplasm resource identifier, and the name or ID of the breeding project associated with the access. Based on the list of germplasm resource identifiers obtained in the first step, the system performs a query operation in the access log database to filter out all log records involving these identifiers within the historical breeding cycle, such as the past ten years. For each germplasm resource identifier, the system sorts all its corresponding log records in chronological order to form the historical call sequence of that germplasm resource. The system also counts the total number of times each germplasm resource is called within the historical breeding cycle, i.e., the call frequency. At the same time, the breeding project name is extracted from each log record, and the project name is associated with the specific type of project, such as conventional hybridization breeding, molecular marker-assisted breeding, and resistance screening breeding. This information, including the timestamp sequence, call frequency, and associated project type, is structured and encapsulated to form a data object, which is the historical call record of that germplasm resource data. All historical access records of relevant germplasm resources are compiled into a dataset for use in subsequent steps.

[0031] Furthermore, S2 includes: The breeding target parameters are used as the root node of the decision logic tree. Based on the trait determination relationship in the peach variety breeding process, the first-level trait sub-nodes under the root node are determined. For each first-level trait sub-node, based on the epistatic effect and additive effect relationship in the peach genetic law, the second-level trait sub-nodes are recursively expanded until all sub-nodes can correspond to specific molecular markers or germplasm resource entities, forming a multi-layer tree structure containing root nodes, intermediate nodes and leaf nodes, which serves as the decision logic tree. For each trait node in the decision logic tree, identify the level depth of the trait node in the tree structure and the number of lower branches directly connected to the trait node. Combine the level depth and the number of branches to form the node topology feature of the trait node. Then, perform weighted aggregation on the node topology features of all trait nodes to obtain the breeding complexity, which is used to characterize the total amount of information dimensions that the breeding target parameters need to be processed in the breeding decision process. The time series data of each germplasm resource data being called is extracted from the historical call records. Each call event in the time series data is defined as an observable state in the Hidden Markov Model (HMM), and the genetic similarity relationship between germplasm resource data is defined as a latent state in the HMM. By calculating the transition probability matrix between latent states and the output probability matrix of the observable state to the latent state, forward-backward iterative training is performed on the time series data to obtain the converged HMM parameters. The posterior probability of each germplasm resource data being selected again under the current breeding objective is calculated using the converged HMM parameters. The posterior probability is used as the label association strength corresponding to that germplasm resource data.

[0032] In this embodiment, the breeding target parameters are used as the root node of the decision logic tree. These parameters are one or more standardized trait identifiers, such as fruit firmness and soluble solids content. The system first creates a root node, representing the final goal of the entire breeding decision. Then, the system accesses a built-in peach trait decomposition knowledge base, which stores the genetic determination relationships between traits in the form of rules or graphs. For example, the knowledge base records that fruit firmness is determined by three secondary traits: pulp cell wall structure, pulp cell turgor pressure, and intercellular binding force. Pulp cell wall structure is further determined by more fundamental traits such as cellulose content, hemicellulose content, and pectin content. Based on these relationships, the system adds child nodes layer by layer, starting from the root node. The determination of each layer of nodes needs to be based on genetic laws, such as epistatic effects (the expression of one gene is influenced by another gene) and additive effects (the cumulative effects of multiple genes). When adding child nodes, the system checks whether the current node directly corresponds to a molecular marker or germplasm resource entity. For example, if the pectin content trait already has a known molecular marker, such as an SSR marker linked to the pectin synthase gene, then that node can be designated as a leaf node and decomposition can stop. If the current node is still a complex trait, decomposition continues. The final tree structure includes root nodes, intermediate nodes, and leaf nodes, with all leaf nodes corresponding to measurable molecular markers or specific germplasm resources.

[0033] For each trait node in the decision logic tree, the system calculates its topological features. First, the node's depth in the tree is determined: the root node has a depth of 1, its child nodes have a depth of 2, and so on. Then, the number of directly connected lower-level branches is counted, i.e., the number of child nodes the node has. These two values ​​constitute the node's topological feature vector. For example, a node with a depth of 3 and 2 child nodes has a topological feature of (3,2). The system performs weighted aggregation of the topological features of all nodes. The weights are set based on the node's importance in breeding decisions, which can be determined by expert scoring or the frequency of attention given to the trait in historical breeding data. The aggregation process involves calculating the product of each node's depth and the number of branches, multiplying this by the node's weight, and finally summing the products of all nodes to obtain the breeding complexity. Breeding complexity is a dimensionless value reflecting the total amount of information dimensions that need to be processed during the breeding decision-making process. Higher complexity means more complex trait relationships and marker combinations need to be analyzed.

[0034] Simultaneously, a Hidden Markov Model (HMM) was used to perform state transition analysis on historical call records. First, the event sequence of each germplasm resource being called was extracted from the historical call records, such as calls in March 2018 and May 2019. These call events constitute an observable state sequence. The system defines the latent state as the genetic background state of the germplasm resource, such as northern cultivar groups, southern cultivar groups, and European / American cultivar groups. These latent states are obtained by clustering the genomic data of the germplasm resources. The number of clusters is determined based on genetic diversity analysis. The parameters of the HMM are initialized, including the initial state probability distribution, the state transition probability matrix, and the observation probability matrix. Then, the Baum-Welch algorithm is used to iteratively train the observed sequences of all germplasm resources, updating the model parameters in each iteration until the model converges. The converged model can describe the transition patterns between different genetic background states and the probability of observing a specific call event in each state. Next, for each germplasm resource, based on its most recent invocation event and the converged model, the forward algorithm is used to calculate the probability of the resource being in each latent state at the current moment. Then, combined with the observed probability matrix, the probability of the resource being invoked again in the current latent state is calculated. This probability is the label association strength of the germplasm resource, which quantifies the potential association between the resource and the current breeding objective.

[0035] Furthermore, S3 includes: The breeding complexity is mapped to the threshold of the number of blocks when performing haplotype block segmentation on germplasm resource data. Based on the block number threshold, the maximum number of molecular markers that each haplotype block can contain is determined. The maximum number of molecular markers is used as the grouping granularity for tag combination on germplasm resource data. The grouping granularity determines the upper limit of the number of markers processed in each combination operation when extracting molecular marker data from the peach germplasm resource bank. All germplasm resource data in the peach germplasm resource bank are sorted in descending order according to the marker association strength to obtain germplasm resource sequences sorted from high to low marker association strength. The intensity transition positions where the marker association strength changes significantly are identified from the germplasm resource sequences. The germplasm resource sequences are divided into at least two intensity intervals according to the intensity transition positions. The physical distance or genetic distance between adjacent intensity intervals is measured and used as the marker spacing between adjacent groups. The germplasm resource data within each intensity interval is divided into haplotype blocks according to the grouping granularity, so that the number of molecular markers in each haplotype block does not exceed the maximum number of molecular markers. The address offset between adjacent haplotype blocks is determined according to the marker spacing to ensure that the data in different haplotype blocks will not conflict during parallel access.

[0036] In this embodiment, breeding complexity is mapped to a threshold for the number of blocks when performing haplotype segmentation on germplasm resource data. The system maintains a mapping table that maps the numerical range of breeding complexity to different block number thresholds. For example, when the breeding complexity is between 0 and 100, the threshold is 10; when it is between 100 and 200, the threshold is 20; and so on. The block number threshold represents the maximum number of molecular markers that each haplotype block can contain when performing haplotype segmentation. This threshold determines the size of the subsequent grouping granularity. For example, if the threshold is 10, the grouping granularity is 10, meaning that the number of markers in each block cannot exceed 10.

[0037] All germplasm resource data in the peach germplasm resource bank are sorted in descending order based on the marker association strength, resulting in a germplasm resource sequence ordered from high to low. The marker association strength is a value between 0 and 1; a higher strength indicates a greater likelihood that the resource will be selected under the current breeding objective. The system scans this sequence, calculates the difference in marker association strength between adjacent resources, forming a difference sequence. Then, based on a preset transition threshold, such as 0.05, the system identifies positions where the difference is greater than the threshold; these positions are the strength transition positions. Based on these transition positions, the entire germplasm resource sequence is divided into multiple strength intervals, with resources within each interval having similar marker association strengths. For example, resources with a strength above 0.8 constitute a high-strength interval, 0.6-0.8 constitute a medium-strength interval, and below 0.6 constitute a low-strength interval.

[0038] The physical or genetic distance between adjacent intensity intervals is measured. Physical distance refers to the average physical location difference of the germplasm resources corresponding to the two intervals on the genome. If the germplasm resources do not have a clear physical location, the genetic distance can be used, which is the reciprocal of the genetic similarity calculated based on molecular markers. The system first determines the central germplasm of each intensity interval, and then calculates the physical distance between the central germplasm of two adjacent intervals based on the reference genome coordinates or the genetic distance based on the marker genotype. This distance value is used as the marker spacing between adjacent groups. The marker spacing determines the address length that needs to be traversed when jumping from one group to the next group during data access, which helps to avoid address conflicts and optimize data prefetching.

[0039] The system performs haplotype segmentation of germplasm resource data within each intensity interval based on the grouping granularity. For each intensity interval, the system first acquires the molecular marker genotype data of all germplasm resources within that interval on the target genome segment. Then, it divides these markers into sliding windows according to the grouping granularity, i.e., the maximum number of markers in each segment. For example, if the grouping granularity is 10 and there are 25 markers in a certain interval, the system will divide it into three segments: the first segment contains markers 1-10, the second segment contains markers 11-20, and the third segment contains markers 21-25. Each segment is a haplotype segment, and the markers within a segment may be in a state of linkage disequilibrium genetically. Simultaneously, the system determines the address offset of adjacent haplotype segments in storage space based on the marker spacing. For example, if the marker spacing is 1000 base pairs, after the first segment is completed, the starting address of the second segment will be offset by 1000 base pairs relative to the starting address of the first segment. This allows multiple segments of data to be read simultaneously without interference during subsequent parallel access.

[0040] Furthermore, S4 includes: For each trait node in the decision logic tree, query the annotation file of the peach reference genome to determine the starting and ending physical coordinates of the functional genes associated with the trait corresponding to the trait node on the peach reference genome. The continuous interval between the starting and ending physical coordinates is taken as the candidate genome segment associated with the trait node, and the correspondence between the trait node and the candidate genome segment is taken as the mapping relationship. Collect candidate genome segments corresponding to all trait nodes in the decision logic tree, merge and remove duplicates of the physical locations of all candidate genome segments on the peach reference genome, and obtain a preliminary genome segment distribution composed of non-overlapping genome intervals; For each genomic region in the preliminary genomic region distribution, the average value of the marker association strength of all molecular markers falling within that genomic region is calculated. This average value is used as the region association score for that genomic region. The region association score is compared with a preset screening threshold. Genomic regions with region association scores greater than or equal to the preset screening threshold are retained, while genomic regions with region association scores less than the preset screening threshold are removed. All the retained genomic regions are used as the target genomic region distribution.

[0041] In this embodiment, for each trait node in the decision logic tree, the annotation file of the peach reference genome is queried to determine the start and end physical coordinates of the functional genes associated with the trait corresponding to that trait node on the peach reference genome. The peach reference genome annotation file is a standard GFF or GTF format file, which records the chromosomal location information of each gene, transcript, exon, and other element. The system first searches for a list of genes associated with the trait in a pre-constructed trait-gene association database based on the trait name represented by the trait node. The trait-gene association database is obtained through literature mining and experimental data integration; for example, genes related to fruit flesh firmness include PpEXP2 and PpPG. For each related gene, the system finds its chromosome number and start and end coordinates in the annotation file. These coordinate intervals are merged to obtain the candidate genomic regions associated with the trait node. For example, if the firmness of the fruit is associated with three genes located at positions 1000-2000, 3000-4000, and 5000-6000 on chromosome 1, the candidate segment may be the union of these three segments, or they may be merged into a continuous segment according to the linkage relationship. The system records the correspondence between the trait node and this segment as a mapping relationship.

[0042] The process involves collecting candidate genomic regions corresponding to all trait nodes in the decision logic tree, merging and deduplicating the physical locations of all candidate regions on the peach reference genome, and obtaining a preliminary genomic region distribution composed of non-overlapping genomic intervals. The merging process needs to handle overlaps between candidate regions from different nodes. For example, if node A's region is on chromosome 1 (1000-3000) and node B's region is on chromosome 1 (2000-4000), after merging, we get chromosome 1 (1000-4000). If node C's region is on chromosome 2 (5000-6000), it does not overlap with the region on chromosome 1. Therefore, the preliminary genomic region distribution contains two intervals: chromosome 1 (1000-4000) and chromosome 2 (5000-6000). Each interval after merging and deduplication is a continuous, non-overlapping genomic region.

[0043] For each genomic region in the preliminary genomic segment distribution, the average marker association strength of all molecular markers falling within that region is calculated, and this average is used as the region association score for that genomic region. The marker association strength of molecular markers is a value calculated in the previous step, representing the degree of association between each marker and the current breeding target. The system first filters all markers whose physical location falls within the current genomic region from the marker database, obtains the marker association strength of each marker, and then calculates the arithmetic mean of these strengths. If there is no marker in the region, the region association score is 0. The region association score of each region is compared with a preset screening threshold, for example, a threshold of 0.5. Genomic regions with a region association score greater than or equal to 0.5 are retained, as these regions are considered to be highly related to the breeding target. Regions with scores less than 0.5 are removed, as these regions have low correlation and are not considered in subsequent analyses. Finally, all the genomic regions retained constitute the target genomic segment distribution, which focuses on the genomic regions most relevant to the breeding target.

[0044] Furthermore, referring to Figure 3 S5 includes: S51, determine the maximum number of markers to group all molecular markers within the target genome segment distribution according to the grouping granularity, divide the target genome segment distribution into multiple consecutive haplotype blocks according to the maximum number of markers, and assign a unique block index to each haplotype block; S52, extract the logical arrangement order of all leaf nodes in the decision logic tree. The logical arrangement order is determined by the breadth-first traversal order from the root node to the leaf node. Reorder multiple haplotype blocks according to the logical arrangement order so that the reordered haplotype block sequence corresponds one-to-one with the leaf node sequence of the decision logic tree. S53. For each haplotype block after reordering, extract the allelic variation information of all molecular markers in the haplotype block, and arrange the allelic variation information according to the position of the haplotype block in the haplotype block sequence to form a two-dimensional matrix structure. The rows of the matrix correspond to different germplasm resources, and the columns of the matrix correspond to molecular markers in the haplotype block. The two-dimensional matrix structure is the genome feature map.

[0045] In this embodiment, the maximum number of markers to group all molecular markers within the target genome segment distribution is determined based on the grouping granularity. The target genome segment distribution is then divided into multiple consecutive haplotype blocks based on the maximum number of markers, and each haplotype block is assigned a unique block index. The grouping granularity is a numerical value, such as 10, indicating that each haplotype block contains a maximum of 10 molecular markers. The system traverses each genomic interval in the target genome segment distribution, arranging all molecular markers within the interval according to their physical order on the chromosome. Then, a sliding window is used to divide the genome according to the grouping granularity; for example, starting from the first marker, every 10 markers form a block. If fewer than 10 markers remain at the end, they are also grouped into a separate block. Each block is assigned an incremental index number, such as 1, 2, 3,... In this way, the entire target genome segment distribution is divided into a series of consecutive haplotype blocks, each block consisting of a group of physically adjacent markers.

[0046] The logical order of all leaf nodes in the decision logic tree is extracted. Leaf nodes are nodes without child nodes and directly correspond to molecular markers or germplasm resource entities. The system uses a breadth-first traversal algorithm to traverse the decision logic tree, starting from the root node and visiting each node from left to right. When a leaf node is encountered, its identifier is recorded. After traversal, a list of leaf node sequences is obtained. This sequence reflects the logical order in which traits are considered during breeding decisions. The system then reorders the previously divided haplotype blocks according to this leaf node sequence. Specifically, a one-to-one correspondence is established between each leaf node in the leaf node list and a haplotype block. Typically, a leaf node may correspond to one or more blocks, which needs to be determined based on the mapping relationship in the knowledge base. If a one-to-many relationship exists, these blocks are arranged according to their genomic order, and the positions of the leaf nodes are replaced as a whole, ultimately resulting in a new haplotype block sequence. This sequence corresponds one-to-one with the leaf node sequence of the decision logic tree, thus ensuring that the data organization matches the decision logic.

[0047] For each reordered haplotype block, allelic variation information of all molecular markers within that haplotype block is extracted. This allelic variation information is then arranged according to the position of the haplotype block within its sequence, forming a two-dimensional matrix structure. The rows of this matrix correspond to different germplasm resources, and the columns correspond to the molecular markers within the haplotype block. This two-dimensional matrix structure constitutes the genome feature map. Specifically, the system first reads the genotype data of all germplasm resources from the germplasm resource bank. For each molecular marker locus within the current haplotype block, the allelic variation type at that locus for each material is obtained, such as A, T, C, G, or - deletion. This allelic variation information is then filled into a temporary matrix with germplasm resources as rows and marker loci as columns. After processing one block, the corresponding column groups are concatenated into the final matrix according to their positions in the sequence. After processing all blocks, a complete two-dimensional matrix is ​​formed, with the number of rows representing the number of germplasm resources and the number of columns representing the total number of markers within all blocks. This matrix not only contains genotype information of molecular markers, but also implicitly contains structural information of decision logic trees, hence it is called a genome feature map.

[0048] Furthermore, S53 includes: For each haplotype block after reordering, the corresponding physical interval is located in the distribution of the target genome segment based on the block index of the haplotype block. Genotype data of all germplasm resources corresponding to the physical interval are read from the peach germplasm resource bank. Allelic variation type of each molecular marker locus is extracted from the genotype data and converted into binary code or character code. Based on the positional order of haplotype blocks in the haplotype block sequence, the column start position and column end position of the haplotype block in the two-dimensional matrix structure are determined, and the allelic variant codes of all extracted molecular marker sites are sequentially filled into the corresponding columns of the two-dimensional matrix structure according to the column start position and column end position. Traverse all the reordered haplotype blocks and fill the corresponding positions of the two-dimensional matrix structure with the allelic variant codes of each haplotype block until all haplotype blocks have been processed, forming a two-dimensional matrix with the number of rows equal to the number of germplasm resources and the number of columns equal to the total number of molecular markers in all haplotype blocks. Use this two-dimensional matrix as a genome feature map.

[0049] In this embodiment, for each reordered haplotype block, the corresponding physical region is located in the target genome segment distribution based on the block index of that haplotype block. Genotype data of all germplasm resources corresponding to the physical region are read from the peach germplasm resource bank. Allelic variation types for each marker locus are extracted from the genotype data and converted into binary or character codes. The system first finds the corresponding physical genomic region based on the block index, for example, chromosome 1000-2000. Then, a query is sent to the peach germplasm resource bank requesting the return of genotype data for all germplasm resources within that physical region. The returned data is typically a three-dimensional structure: material × marker × allele. For each marker locus, the system obtains the allele genotype of each material. If the marker is a biallelic marker, alleles can be encoded as 0 and 1, for example, A is encoded as 0 and T as 1. If it is a polyallele marker, character encoding can be used, for example, directly using characters such as A and T. If there are missing values, special symbols such as - are used to indicate them. In this way, the allelic variation information of each marker site is converted into a unified encoding form.

[0050] Based on the positional order of haplotype blocks within the haplotype block sequence, the start and end positions of each column in the two-dimensional matrix structure are determined. Since the entire genome feature map is arranged sequentially by block, with each block occupying several consecutive columns, the system maintains a global column counter during map construction. For the first block, the column start position is 1, and the column end position is the number of markers within that block. The column start position for the second block is the column end position of the first block plus 1, and so on. Therefore, for the currently processed block, the system can calculate its corresponding column range in the map. Then, the allelic variant codes for each previously converted marker locus are sequentially filled into the corresponding columns of the two-dimensional matrix according to the order of the markers within the block. During filling, the data is filled row by row according to the order of the germplasm resources, ensuring that each row corresponds to the complete genotype of one material.

[0051] The process iterates through all the reordered haplotype blocks, repeating the above steps for each block. Allelic variant codes for each block are then inserted into the corresponding positions in the two-dimensional matrix. This process continues until all blocks have been processed. The resulting two-dimensional matrix has rows equal to the number of germplasm resources and columns equal to the total number of molecular markers within all haplotype blocks. This two-dimensional matrix is ​​the genome feature map, which fully represents the genotype combinations of all germplasm resources on the target genome segment. Furthermore, the column order corresponds to the leaf node order of the decision logic tree, thus achieving the integration of data organization and decision logic.

[0052] Furthermore, S6 includes: Based on the molecular marker sites corresponding to each column in the genome feature map, the physical address of the molecular marker site on the peach reference genome is determined. The address difference between two adjacent molecular marker sites is calculated based on the marker spacing. The address difference is used as the address jump step size when continuously reading molecular marker data from the peach germplasm resource bank. Based on the germplasm resource material corresponding to each row in the genome feature map, a data reading instruction corresponding to that germplasm resource material is generated. The data reading instruction contains the physical addresses of all molecular marker sites corresponding to the germplasm resource material and the address jump step size between adjacent physical addresses. The data reading instructions for different germplasm resource materials are sent in parallel to the storage controller of the peach germplasm resource bank. The storage controller reads the molecular marker data of multiple germplasm resource materials in parallel from the storage medium according to the address jump step size and merges them to form a molecular marker data stream. The decision-making timeline of the target breeding task in the intelligent breeding process is analyzed. The decision-making timeline includes the decision-making stages and the start and end times of each decision-making stage. The available decision-making time is calculated based on the time difference between the end time and the start time of the current decision-making stage. The maximum amount of data that can be transmitted from the peach germplasm resource bank within the current decision-making stage is calculated based on the data output bandwidth of the storage controller and the available decision-making time. The molecular marker data stream is divided into multiple data blocks according to the row structure of the genome feature map. Each data block corresponds to the complete molecular marker information of a germplasm resource material. The number of data blocks that can be included in a single data access is determined based on the maximum data volume. The number of data blocks is used as the batch processing data volume. The storage controller transmits the molecular marker data stream to the intelligent breeding decision module in batches according to the batch processing data volume.

[0053] In this embodiment, the physical address of each molecular marker site on the peach reference genome is determined based on the molecular marker site corresponding to each column in the genome feature map. The address difference between two adjacent molecular marker sites is calculated based on the marker spacing, and this address difference is used as the address jump step size when continuously reading molecular marker data from the peach germplasm resource bank. Each column of the genome feature map corresponds to a specific molecular marker. The system obtains the chromosome and physical coordinates of each marker by querying the marker database. The columns of the map are arranged sequentially. For two adjacent columns, the difference in their physical coordinates is calculated. If the marker spacing has been predetermined, it is directly used as the address jump step size; otherwise, it can be calculated based on the physical coordinate difference. For example, if marker A is at 1000 bp on chromosome 1 and marker B is at 1500 bp on chromosome 1, the step size is 500 bp. The step size guides the storage controller to skip unnecessary intermediate data and directly locate the position of the next target marker when performing continuous address access.

[0054] Based on the germplasm resources corresponding to each row in the genome feature map, the system generates data reading instructions for each germplasm resource. Each row represents one germplasm resource, and the marker sites in all columns of that row constitute the molecular marker data set for that material. The system generates one reading instruction for each material, containing a list of the physical addresses of all marker sites for that material, as well as the jump step size between adjacent addresses. Since the physical addresses of marker sites may differ between different materials (because different materials may have the same address at the same marker site, but their storage locations may differ), the address list for each instruction is specific to the storage location of that material. After generating the instructions, the system sends these instructions in parallel to the storage controller of the peach germplasm resource bank. The storage controller has multi-channel parallel processing capabilities, enabling it to receive and execute multiple instructions simultaneously. The controller reads data from storage media such as hard drives or SSDs according to the address list and step size in each instruction. For example, for instruction 1, the controller reads data from address A1, then skips the step size to reach A2, and so on; simultaneously, instruction 2 performs a similar operation on another channel. All read data streams converge and merge inside the controller to form a molecular marker data stream, which is a mixed stream of multiple material data.

[0055] The system analyzes the decision-making timeline of the target breeding task within the intelligent breeding process. The decision-making timeline is a predefined timetable describing the start and end times of each stage in the breeding decision-making process. For example, the first stage is parent selection, with a start time t1 and an end time t2; the second stage is hybrid combination design, with a start time t2 and an end time t3. Based on the start and end times of the current decision-making stage, such as the ongoing parent selection, the system calculates the available decision-making time, which is the end time minus the start time. Simultaneously, the system obtains the data output bandwidth of the storage controller, i.e., the amount of data that can be transferred from the storage medium to the decision module per unit time, for example, 100MB per second. Multiplying the bandwidth by the available decision-making time yields the maximum amount of data that can be transferred from the peach germplasm resource bank within the current decision-making stage.

[0056] The molecular marker data stream is divided into multiple data blocks according to the row structure of the genome feature map. Each data block corresponds to the complete molecular marker information of one germplasm resource material. For example, if the map has 1000 rows, the data stream can be divided into 1000 data blocks. The size of each block may vary. The number of data blocks that can be included in a single data access, i.e., the batch processing data volume, is determined based on the maximum data volume. For example, if the maximum data volume is 500MB and the average size of each data block is 5MB, the batch processing data volume is 100, meaning that complete data of 100 materials can be transmitted at a time. The storage controller transmits the molecular marker data stream to the intelligent breeding decision module in batches according to this batch processing data volume. For example, the first 100 data blocks are transmitted in the first batch, the next 100 in the second batch, and so on, until all data blocks are transmitted. In this way, data transmission is synchronized with the decision-making sequence, avoiding data congestion and waiting.

[0057] Furthermore, following S6, it also includes: The molecular marker data stream is input into a pre-configured peach variety breeding decision engine. The peach variety breeding decision engine decodes the molecular marker data stream in parallel according to the marker combination relationship defined by the genome feature map, and restores the haplotype combination of each germplasm resource material in the target genome segment. The peach variety breeding decision engine compares the restored haplotype combinations with the preset ideal haplotype template, calculates the genetic similarity between each germplasm resource material and the ideal haplotype template, and sorts all germplasm resource materials according to the genetic similarity to generate a list of recommended parent materials. The recommended parent material list is linked with the field phenotypic data corresponding to each germplasm resource material, so that breeding experts can determine the final hybridization combination scheme based on the recommended parent material list and field phenotypic data.

[0058] In this embodiment, based on the genome feature map and marker spacing, the corresponding molecular marker data stream is extracted in parallel from the peach germplasm resource bank. After adjusting the batch processing data volume of a single data access according to the decision sequence of the target breeding task in the intelligent breeding process, the molecular marker data stream is input into a pre-configured peach variety breeding decision engine. The peach variety breeding decision engine is a specially designed computing module used to process molecular marker data and provide breeding suggestions. The data stream received by the engine is transmitted in batches, with each batch containing molecular marker data from multiple germplasm resources. The engine first performs parallel decoding of the molecular marker data stream according to the marker combination relationship defined by the genome feature map. The construction method of the genome feature map determines which markers belong to the same haplotype block. The engine uses this block information to organize the marker data belonging to the same material in the data stream according to the blocks. Specifically, for each material, the engine restores the data to the allelic variation combination of the material in each haplotype block according to the column order of the markers in the map. For example, if block 1 in the map contains markers 1-10 and block 2 contains markers 11-20, the engine will combine the data of markers 1-10 into the haplotype of block 1, and the data of markers 11-20 into the haplotype of block 2. In this way, each germplasm resource material is restored to its haplotype combination in the target genome segment, that is, a set of ordered haplotype sequences.

[0059] The peach variety breeding decision engine compares the reconstructed haplotype combinations with a preset ideal haplotype template, calculating the genetic similarity between each germplasm resource material and the ideal haplotype template. The ideal haplotype template is an idealized haplotype combination pre-defined by breeding experts based on the target trait; it represents the allelic variation configuration most conducive to the expression of the target trait. The comparison process uses a block-by-block comparison method. For each haplotype block, it compares whether the allelic variation of the material in that block is consistent with the allelic variation of the ideal template in that block. If they are consistent, the block scores 1; otherwise, it scores 0. The scores of all blocks are added together and then divided by the total number of blocks to obtain the genetic similarity, which is between 0 and 1. For example, if there are a total of 10 blocks, and the material has 8 blocks consistent with the template, the similarity is 0.8. After calculating the similarity of all germplasm resources, the engine sorts all materials from high to low similarity and generates a recommended parent material list. The top of the list contains the materials with the highest similarity, which are most likely to be used as excellent parents for hybridization breeding.

[0060] The list of recommended parental materials is associated with the field phenotypic data corresponding to each germplasm resource material. The field phenotypic data are the actual trait performances observed over multiple years and multiple locations, such as single fruit weight, soluble solid content, fruit firmness, etc. These data are pre-stored in the germplasm resource library and associated with the unique identifier of each material. When generating the recommended list, the engine simultaneously extracts the phenotypic data of each material from the database and presents the phenotypic data together with the material information. The presentation can be in the form of a table, listing the material name, genetic similarity, and the values of key phenotypic traits. Breeders can, based on this display interface, comprehensively consider the genetic similarity and field phenotypic data to finally determine the parental combination plan for hybridization. For example, experts may select materials with high genetic similarity and excellent phenotypes as the male and female parents.

[0061] Further, after S3, it further includes: Encoding the grouping granularity and marker spacing as metadata tags, and associatively storing the metadata tags with the breeding target parameters to establish a correspondence library between the breeding target parameters and the data access mode; When receiving a new breeding target task, retrieve from the correspondence library whether there are historical grouping granularity and historical marker spacing that match the new breeding target parameters. If so, directly call the historical grouping granularity and historical marker spacing to construct the genomic feature map to accelerate the response speed to the new breeding target task.

[0062] Among them, in this embodiment, after determining the grouping granularity of the marker combinations for all germplasm resource data in the peach germplasm resource library and the marker spacing between adjacent groups according to the breeding complexity and marker association strength, the grouping granularity and marker spacing are encoded as metadata tags, and the metadata tags are associatively stored with the breeding target parameters to establish a correspondence library between the breeding target parameters and the data access mode; the metadata tags are a type of structured data description information, for example, recorded in XML or JSON format: <breeding target parameter>Fruit firmness is high< / breeding target parameter><grouping granularity>10< / grouping granularity><marker spacing>500bp< / marker spacing>. The system stores these metadata tags together with the corresponding breeding target parameters such as high fruit firmness in an independent correspondence library, which can be a database table or a key-value store. In this way, whenever the parsing of the breeding target parameters and the determination of the data access mode are completed, the system persists this correspondence.

[0063] When a new breeding target task is received, the system first parses the new breeding target parameters. Then, it searches the corresponding relational database for historical grouping granularity and historical marker spacing that match the new breeding target parameters. The matching process can use exact matching or similarity matching. Exact matching requires that the new breeding target parameters are completely consistent with the parameters recorded in the database. Similarity matching calculates the semantic similarity between parameters, and a match is considered successful when the similarity exceeds a preset threshold. For example, if the new breeding target is extremely high fruit firmness, and the database already contains records of high fruit firmness, the two have high semantic similarity and can be considered a match. If a matching historical record is found, the system directly uses the stored historical grouping granularity and historical marker spacing for subsequent construction of genome feature maps and extraction of molecular marker data streams. This skips the steps of recalculating breeding complexity and marker association strength, accelerating the response speed to new breeding target tasks. If no matching historical record is found, the system performs the full calculation process and stores the newly calculated grouping granularity and marker spacing in the corresponding relational database for future use.

[0064] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A smart breeding decision-making method for new peach varieties based on multi-omics data and molecular markers, characterized in that, include: S1, obtain the description text of the breeding target trait corresponding to the target breeding task, and parse the breeding target trait description text to obtain the breeding target parameters, and obtain the historical call records of germplasm resource data associated with the breeding target parameters in the peach germplasm resource bank within the historical breeding cycle; S2, construct a decision logic tree containing multiple trait nodes according to the breeding target parameters, extract the node topology features of each trait node from the tree structure of the decision logic tree, determine the breeding complexity corresponding to the breeding target parameters, and use a hidden Markov model to perform state transition analysis on the historical call records to obtain the label association strength corresponding to each germplasm resource data in the peach germplasm resource bank. S3, based on the breeding complexity and the marker association strength, determine the grouping granularity for tagging all germplasm resource data in the peach germplasm resource bank and the marker spacing between adjacent groups; S4, establish a mapping relationship between each trait node in the decision logic tree and the physical location of the peach reference genome, determine the preliminary genome segment distribution based on the mapping relationship, and filter the preliminary genome segment distribution based on the marker association strength to obtain the target genome segment distribution; S5, perform haplotype logical recombination on the germplasm resource data corresponding to the distribution of the target genome segment according to the grouping granularity, and construct a genome feature map that matches the topology of the decision logic tree; S6. Based on the genome feature map and the marker spacing, the corresponding molecular marker data stream is extracted in parallel from the peach germplasm resource bank, and the batch processing data volume of a single data access is adjusted according to the decision sequence of the target breeding task in the intelligent breeding process, so as to complete the accelerated decision-making of the peach new variety breeding process.

2. The method according to claim 1, characterized in that, S1 includes: The system receives target trait description text for the selection and breeding of new peach varieties from breeding experts, identifies trait keywords in the target trait description text through natural language processing technology, and performs semantic matching of the trait keywords with a preset trait ontology library to obtain breeding target parameters used to characterize specific breeding directions. Based on the breeding target parameters, a search is conducted in the metadata index of the peach germplasm resource bank to determine the unique identifier of at least one germplasm resource data that is associated with the breeding target parameters; Based on the unique identifier, a backtracking query is performed in the access log system of the peach germplasm resource bank to extract the timestamp, call frequency, and breeding project type associated with each germplasm resource data in the historical breeding cycle. The timestamp, call frequency, and breeding project type are aggregated and encapsulated to form a historical call record corresponding to each germplasm resource data.

3. The method according to claim 1, characterized in that, S2 includes: The breeding target parameters are used as the root node of the decision logic tree. The first-level trait sub-nodes under the root node are determined according to the trait determination relationship in the peach variety breeding process. For each first-level trait sub-node, the second-level trait sub-nodes are obtained by recursively expanding according to the epistatic effect and additive effect relationship in the peach genetic law, until all sub-nodes can correspond to specific molecular markers or germplasm resource entities, forming a multi-layer tree structure containing root nodes, intermediate nodes and leaf nodes, which serves as the decision logic tree. For each trait node in the decision logic tree, the level depth of the trait node in the tree structure and the number of lower branches directly connected to the trait node are identified. The level depth and the number of branches are combined to form the node topology feature of the trait node. The node topology features of all trait nodes are weighted and aggregated to obtain the breeding complexity, which is used to characterize the total amount of information dimensions that the breeding target parameters need to be processed in the breeding decision process. Extract the time series data of each germplasm resource data being called from the historical call records. Define each call event in the time series data as an observable state in the Hidden Markov Model (HMM), and define the genetic similarity relationship between germplasm resource data as a latent state in the HMM. By calculating the transition probability matrix between latent states and the output probability matrix of the observable state to the latent state, perform forward-backward iterative training on the time series data to obtain converged HMM parameters. Use the converged HMM parameters to calculate the posterior probability of each germplasm resource data being selected again under the current breeding objective, and use the posterior probability as the label association strength corresponding to that germplasm resource data.

4. The method according to claim 1, characterized in that, The S3 includes: The breeding complexity is mapped to a threshold for the number of blocks when performing haplotype block segmentation on germplasm resource data. The maximum number of molecular markers that each haplotype block can contain is determined based on the threshold for the number of blocks. The maximum number of molecular markers is used as the grouping granularity for combining markers in germplasm resource data. The grouping granularity determines the upper limit of the number of markers processed in each combination operation when extracting molecular marker data from the peach germplasm resource bank. All germplasm resource data in the peach germplasm resource bank are sorted in descending order according to the marker association strength to obtain a germplasm resource sequence sorted from high to low marker association strength. The intensity transition positions where the marker association strength changes significantly are identified from the germplasm resource sequence. The germplasm resource sequence is divided into at least two intensity intervals according to the intensity transition positions. The physical distance or genetic distance between adjacent intensity intervals is measured and the physical distance or genetic distance is used as the marker spacing between adjacent groups. The germplasm resource data within each intensity interval is divided into haplotype blocks according to the grouping granularity, such that the number of molecular markers in each haplotype block does not exceed the maximum number of molecular markers, and the address offset between adjacent haplotype blocks is determined according to the marker spacing to ensure that data in different haplotype blocks will not conflict during parallel access.

5. The method according to claim 1, characterized in that, The S4 includes: For each trait node in the decision logic tree, the annotation file of the peach reference genome is queried to determine the starting and ending physical coordinates of the functional gene associated with the trait corresponding to the trait node on the peach reference genome. The continuous interval between the starting and ending physical coordinates is taken as the candidate genome segment associated with the trait node, and the correspondence between the trait node and the candidate genome segment is taken as the mapping relationship. Collect all candidate genome regions corresponding to all trait nodes in the decision logic tree, merge and remove duplicates of the physical locations of all candidate genome regions on the peach reference genome, and obtain a preliminary genome region distribution composed of non-overlapping genome regions. For each genomic region in the preliminary genomic region distribution, the average value of the marker association strength of all molecular markers falling within the genomic region is calculated. This average value is used as the region association score of the genomic region. The region association score is compared with a preset screening threshold. Genomic regions with region association scores greater than or equal to the preset screening threshold are retained, while genomic regions with region association scores less than the preset screening threshold are removed. All the retained genomic regions are used as the target genomic region distribution.

6. The method according to claim 1, characterized in that, The S5 includes: S51, determine the maximum number of markers to group all molecular markers within the target genome segment distribution according to the grouping granularity, divide the target genome segment distribution into multiple consecutive haplotype blocks according to the maximum number of markers, and assign a unique block index to each haplotype block; S52, extract the logical arrangement order of all leaf nodes in the decision logic tree. The logical arrangement order is determined by the breadth-first traversal order from the root node to the leaf node. Reorder the multiple haplotype blocks according to the logical arrangement order so that the reordered haplotype block sequence corresponds one-to-one with the leaf node sequence of the decision logic tree. S53, for each reordered haplotype block, extract allelic variation information of all molecular markers within the haplotype block, and arrange the allelic variation information according to the position of the haplotype block in the haplotype block sequence to form a two-dimensional matrix structure, wherein the rows of the matrix correspond to different germplasm resources and the columns of the matrix correspond to molecular markers within the haplotype block, and the two-dimensional matrix structure is the genome feature map.

7. The method according to claim 6, characterized in that, S53 includes: For each haplotype block after reordering, the corresponding physical interval is located in the distribution of the target genome segment according to the block index of the haplotype block. The genotype data of all germplasm resource materials corresponding to the physical interval are read from the peach germplasm resource bank. The allelic variation type of each molecular marker locus is extracted from the genotype data and the allelic variation type is converted into binary code or character code. Based on the positional order of the haplotype blocks in the haplotype block sequence, the column start position and column end position of the haplotype block in the two-dimensional matrix structure are determined, and the allelic variant codes of all extracted molecular marker sites are sequentially filled into the corresponding columns of the two-dimensional matrix structure according to the column start position and column end position. Traverse all the reordered haplotype blocks and fill the allelic variant codes of each haplotype block into the corresponding positions of the two-dimensional matrix structure until all haplotype blocks have been processed, forming a two-dimensional matrix with the number of rows equal to the number of germplasm resources and the number of columns equal to the total number of molecular markers in all haplotype blocks. Use this two-dimensional matrix as the genome feature map.

8. The method according to claim 1, characterized in that, The S6 includes: Based on the molecular marker sites corresponding to each column in the genome feature map, the physical address of the molecular marker site on the peach reference genome is determined. The address difference between two adjacent molecular marker sites is calculated based on the marker spacing. The address difference is used as the address jump step size when continuously reading molecular marker data from the peach germplasm resource bank. Based on the germplasm resource material corresponding to each row in the genome feature map, a data reading instruction corresponding to that germplasm resource material is generated. The data reading instruction contains the physical addresses of all molecular marker sites corresponding to the germplasm resource material and the address jump step size between adjacent physical addresses. The data reading instructions for different germplasm resource materials are sent in parallel to the storage controller of the peach germplasm resource bank. The storage controller reads the molecular marker data of multiple germplasm resource materials in parallel from the storage medium according to the address jump step size and merges them to form the molecular marker data stream. The decision sequence of the target breeding task in the intelligent breeding process is analyzed. The decision sequence includes decision stages and the start and end times of each decision stage. The available decision duration is calculated based on the time difference between the end time and the start time of the current decision stage. The maximum amount of data that can be transmitted from the peach germplasm resource bank within the current decision stage is calculated based on the data output bandwidth of the storage controller and the available decision duration. The molecular marker data stream is divided into multiple data blocks according to the row structure of the genome feature map. Each data block corresponds to a complete set of molecular marker information for a germplasm resource material. The number of data blocks that can be included in a single data access is determined based on the maximum data volume. The number of data blocks is used as the batch processing data volume. The storage controller transmits the molecular marker data stream to the intelligent breeding decision module in batches according to the batch processing data volume.

9. The method according to claim 1, characterized in that, Following S6, it also includes: The molecular marker data stream is input into a pre-configured peach variety breeding decision engine. The peach variety breeding decision engine performs parallel decoding of the molecular marker data stream according to the marker combination relationship defined by the genome feature map, and restores the haplotype combination of each germplasm resource material on the target genome segment. The peach variety breeding decision engine compares the restored haplotype combination with the preset ideal haplotype template, calculates the genetic similarity between each germplasm resource material and the ideal haplotype template, and sorts all germplasm resource materials according to the genetic similarity to generate a list of recommended parent materials. The recommended parent material list is linked with the field phenotypic data corresponding to each germplasm resource material, so that breeding experts can determine the final hybridization combination scheme based on the recommended parent material list and the field phenotypic data.

10. The method according to claim 1, characterized in that, Following S3, it also includes: The grouping granularity and the tag spacing are encoded into metadata tags, and the metadata tags are associated with the breeding target parameters and stored to establish a correspondence library between the breeding target parameters and the data access mode; When a new breeding target task is received, the corresponding relation database is searched to see if there is a historical grouping granularity and historical marker spacing that match the new breeding target parameters. If so, the historical grouping granularity and historical marker spacing are directly called to construct a genome feature map to accelerate the response speed to the new breeding target task.

Images