A method and system for optimal cross pairing of poultry skeletal muscle traits

Abstract

The application provides a poultry skeletal muscle trait optimal cross-pairing design method and system, relates to the poultry breeding technical field, and includes obtaining leg skeletal muscle transcriptome, whole genome, leg muscle weight and tibia length data of a candidate parent population at a specific development time node. By encoding the transcriptome and genome data, a skeletal muscle allelic expression trajectory tensor is constructed. Further, nonlinear time series embedding and pattern recognition are performed on the tensor to obtain gene expression patterns representing non-additive effects. The trait-gene dynamic association network is constructed by fusing the pattern representation and phenotype record, and the key regulatory path is screened out. Based on the key regulatory path, a reinforcement learning model is used to explore and optimize in the potential space, and the optimal parent pairing result that can maximize the skeletal muscle trait performance is output. The application realizes the precise design from multi-omics data to cross decision, and significantly improves the breeding efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of poultry breeding technology, and more specifically, to a method and system for designing optimal hybridization pairings for poultry skeletal muscle traits. Background Technology

[0002] In the current field of poultry genetics and breeding, in the pursuit of skeletal muscle trait improvement, traditional hybridization pairing methods mainly rely on screening of parental phenotypic characteristics and pedigree information analysis. Although this has promoted breeding progress to some extent, it is difficult to accurately analyze the interaction mechanism at the allele level.

[0003] Existing technologies typically employ static molecular marker-assisted selection methods, inferring hybrid vigor from genotype data at limited loci. However, this approach fails to capture the synergistic regulatory patterns of paternal and maternal allele expression trajectories on skeletal muscle formation during critical post-hatching stages of embryonic development. Furthermore, it neglects the correlation between gene expression patterns and non-additive genetic effects at different developmental stages, resulting in limited accuracy and stability in hybridization prediction and hindering breakthroughs in targeted optimization of skeletal muscle traits.

[0004] Therefore, there is an urgent need for a method and system for designing optimal hybridization pairings for poultry skeletal muscle traits to solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for designing optimal crossbreeding pairs for poultry skeletal muscle traits, thereby improving the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0006] In a first aspect, this application provides a method for designing optimal crossbreeding pairs for poultry skeletal muscle traits, including:

[0007] Obtain leg skeletal muscle tissue transcriptional sequencing reads, whole genome sequencing reads, leg muscle weighing records, and tibia length measurement records from candidate paternal and maternal populations at preset time points;

[0008] The leg skeletal muscle tissue transcription sequencing reads and whole genome sequencing reads were processed for allelic expression trajectory encoding. By constructing a trajectory representation that reflects the dominant changes in allelic origin, the skeletal muscle allelic expression trajectory tensor corresponding to each candidate individual was obtained.

[0009] The skeletal muscle allelic expression trajectory tensor is subjected to nonlinear temporal embedding and pattern recognition processing. The long-term dependence of allelic expression at different developmental nodes is captured by a preset neural network. The transition probability of expression state is identified by a hidden Markov model and the expression pattern is classified to obtain the non-additive effect gene expression pattern characterization corresponding to each parent-parent combination.

[0010] Based on the characterization of non-additive effect gene expression patterns, leg muscle weighing records, and tibia length measurement records, a trait-gene dynamic association network was constructed, and key regulatory pathways associated with skeletal muscle trait development were screened.

[0011] Based on the key regulatory pathway representation, the pairing scheme is optimized and the output is processed. The reinforcement learning model is used to explore the potential space composed of allele expression patterns and trait associations, simulate the dynamic process of allele interaction, and obtain the optimal parent pairing result that maximizes skeletal muscle trait performance.

[0012] Secondly, this application also provides an optimal crossbreeding pairing design system for poultry skeletal muscle traits, comprising:

[0013] The acquisition unit is used to acquire leg skeletal muscle tissue transcriptional sequencing reads, whole genome sequencing reads, leg muscle weighing records, and tibia length measurement records from candidate paternal and maternal populations at preset time points.

[0014] The encoding unit is used to encode the allelic expression trajectory of the leg skeletal muscle tissue transcription sequencing reads and the whole genome sequencing reads. By constructing a trajectory representation that reflects the dominant change in allelic source, the skeletal muscle allelic expression trajectory tensor corresponding to each candidate individual is obtained.

[0015] The processing unit is used to perform nonlinear temporal embedding and pattern recognition processing on the skeletal muscle allelic expression trajectory tensor, capture the long-term dependence of allelic expression between different developmental nodes through a preset neural network, identify the transition probability of expression state by applying a hidden Markov model, and classify the expression patterns to obtain the non-additive effect gene expression pattern characterization corresponding to each parent-parent combination.

[0016] The construction unit is used to construct a trait-gene dynamic association network based on the non-additive effect gene expression pattern characterization, leg muscle weighing records, and tibia length measurement records, and to screen out key regulatory pathways associated with skeletal muscle trait development.

[0017] The output unit is used to optimize and output the pairing scheme based on the key regulatory pathway representation. It explores the potential space composed of allele expression patterns and trait associations through a reinforcement learning model, simulates the dynamic process of allele interaction, and obtains the optimal parent pairing result that maximizes skeletal muscle trait performance.

[0018] The beneficial effects of this invention are as follows:

[0019] This invention integrates multi-time-node transcriptomic, whole-genome, phenotypic, and reciprocal cross population data from leg skeletal muscle tissue of candidate parents to achieve a precise characterization of the dynamic interactions of alleles during skeletal muscle development. By constructing an allele expression trajectory tensor and utilizing neural networks and hidden Markov models for nonlinear temporal embedding and pattern recognition, it effectively captures the synergistic and competitive relationships of parental alleles at specific developmental stages, identifying gene expression patterns with significant non-additive effects. Based on this, a trait-gene dynamic association network is constructed by fusing multi-source spatiotemporal data to screen key regulatory pathways closely related to skeletal muscle trait development. Finally, based on a reinforcement learning model, exploration and optimization are performed in the latent space to simulate allele interaction dynamics, thereby outputting the optimal parent pairing result that maximizes skeletal muscle trait performance. This method overcomes the limitations of traditional hybridization breeding relying on experience and static markers, achieving precise and interpretable design from allele expression dynamics to hybrid vigor prediction, significantly improving the efficiency and accuracy of targeted improvement of poultry skeletal muscle traits.

[0020] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the optimal hybridization pairing design method for poultry skeletal muscle traits described in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the optimal hybridization pairing design system for poultry skeletal muscle traits described in an embodiment of the present invention.

[0024] In the diagram: 701, acquisition unit; 702, encoding unit; 703, processing unit; 704, construction unit; 705, output unit. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0026] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] Example 1

[0028] This embodiment provides a method for designing optimal hybridization pairs for poultry skeletal muscle traits.

[0029] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4 and S5.

[0030] Step S1: Obtain leg skeletal muscle tissue transcriptional sequencing reads, whole genome sequencing reads, leg muscle weighing records, and tibia length measurement records from the candidate paternal and maternal populations at preset time points.

[0031] Understandably, this invention first obtains leg skeletal muscle tissue samples from candidate paternal and maternal populations at multiple preset developmental time points (mid-to-late embryonic development and early post-hatching, with mid-to-late development preset to 10 days and early development preset to 5 days), and uses high-throughput sequencing technology to obtain their raw transcriptome sequencing reads to capture dynamic changes in gene expression. Simultaneously, whole-genome resequencing is performed on the same individuals to obtain raw sequencing reads covering the entire genome, providing allele resolution references for subsequent accurate determination of parental transcript origin. At the phenotypic level, leg muscle fresh weight and tibia length are recorded simultaneously at each transcriptome sampling time point, forming phenotypic measurement records strictly corresponding to molecular data. Furthermore, this invention obtains F1 generation population samples through reciprocal cross experiments and clearly records the pedigrees of all experimental individuals (including parents and offspring), thereby constructing a complete data foundation capable of supporting allele-specific expression analysis and hybridization effect assessment.

[0032] Step S2: The leg skeletal muscle tissue transcription sequencing read data and whole genome sequencing read data are processed for allelic expression trajectory encoding. By constructing a trajectory representation that reflects the dominant changes in allelic sources, the skeletal muscle allelic expression trajectory tensor corresponding to each candidate individual is obtained.

[0033] It is understandable that the essence of this step is to precisely map each transcript fragment to its paternal or maternal allele. Based on this, by integrating the allele-specific expression levels at multiple consecutive developmental time points, a three-dimensional allele expression trajectory tensor is constructed. The dimensions of this tensor correspond to different genes, developmental time points, and the expression intensities of the paternal and maternal parents, respectively. In this step, step S2 includes steps S21, S22, and S23.

[0034] Step S21: Based on the leg skeletal muscle tissue transcription sequencing read data and the whole genome sequencing read data, perform parental origin mapping processing. By specifically comparing the whole genome sequencing read data of the candidate paternal population and the candidate maternal population with each leg skeletal muscle tissue transcription sequencing read data, identify their unique parental allele source and obtain the allele-specific read set of each candidate individual at multiple time points.

[0035] Understandably, this step first constructs parent-specific reference sequences from the whole-genome sequencing reads of candidate paternal and maternal populations. Based on this, bidirectional alignment is performed on the transcribed sequencing reads of each leg skeletal muscle tissue. This specific alignment considers the global similarity score of the sequences, effectively distinguishing the fuzzy mapping problem caused by highly repetitive sequence regions in the poultry genome. When dealing with the specific scenario of skeletal muscle development, this method pays special attention to the alignment quality of key gene loci for myocyte differentiation. By dynamically adjusting the alignment parameters to adapt to the sequence characteristics of different functional regions, it ensures accurate identification of the parental origin of myofibril-forming gene transcripts. The final allelic-specific read set not only contains the parental attribution information of each transcribed read but also retains its dynamic expression level over developmental time, providing a precise data foundation for subsequent allelic expression trajectory analysis.

[0036] The formula for calculating the global similarity score is as follows:

[0037] ;

[0038] Where S(r,R) p R is the specific reference sequence for the transcript read r and the parent p. p The final comparison score, S global (r, R) p) represents the preset standard global sequence alignment score, λ represents the weighting coefficient of the specific variant site score, and v i For V p A specific "specific variant site" in the set, V p ∩r represents the set of specific variant sites V of parent p. p In the context, the subset of sites that are exactly covered by the sequence of the currently processed transcript read r, base(r,v) i The return value indicates the location v at the corresponding genomic coordinates of read r. i The "bases" measured at the location, base(R) p ,v i The return value represents the parent p at the set of specific variant sites V. p The "bases" measured at the location, δ, are based on the specific variation site v. i Above, the "bases" carried by the read segment r are similar to the parental reference sequence R. p The score is given based on whether the "bases" at this site are consistent. If they are consistent, a positive reward value (+2) is returned; if they are inconsistent, a negative reward value (-1) is returned. η is the penalty coefficient, such as 1.5.

[0039] Step S22: Perform fuzzy alignment disambiguation processing based on the allele-specific read set. By constructing an allele-specific k-mer fingerprint and calculating the local alignment confidence, reads with confidence less than or equal to a preset threshold are removed to obtain skeletal muscle development-related allele expression sequences with confidence greater than the preset threshold.

[0040] Understandably, this step first constructs allele-specific k-mer fingerprints, which involves extracting all short sequence fragments of length k from the specific reference sequences of the father and mother, along with their frequency of occurrence and genomic location, forming a parent-specific sequence tag library. When calculating local alignment confidence, this method does not rely on the overall score of the global alignment. Instead, it decomposes the read to be evaluated into consecutive k-mers, checks the presence and uniqueness of each k-mer in the father and mother fingerprint libraries one by one, and calculates the evidence weight for supporting preliminary parental assignment and opposing parental assignment within the read length range. The confidence calculation integrates the number of specific k-mer matches, their rarity in the parent genome (to reduce the weight of highly repetitive sequence k-mers), and the degree of sequence differences between parents (prioritizing k-mers with base differences between the two parents as decisive evidence). Given the potential for highly homologous family members among myogenic genes in poultry skeletal muscle development, this method effectively identifies and downweights reads originating from multiple gene families and lacking clear parental distinction. Finally, by filtering using a pre-set confidence threshold, low-confidence reads with weak supporting evidence or numerous contradictory pieces of evidence are removed, resulting in a high-confidence set of allelic expression sequences related to skeletal muscle development. This provides a reliable data quality foundation for subsequent precise quantification of allelic expression dynamics.

[0041] The formula for constructing allele-specific k-mer fingerprints is shown below:

[0042] ;

[0043] Among them, F p (k) represents the allele-specific k-mer fingerprint set of parent p, s represents a DNA sequence fragment of length k, called a k-mer, and c p (s) represents the DNA sequence fragment s in the reference genome R of parent p. p The number of occurrences in, among which, reference genome R p The results of the experiment can be obtained directly from the database. Φ p (s) is the genomic location context feature vector of DNA sequence fragment s in parent p, such as whether it is located in a gene coding region, whether it is located in a muscle-specific enhancer region, GC content, and Hamming distance to the corresponding k-mer of the other parent, etc. K(R) p ,k) represents the sequence from the reference sequence R p Let I be the set of all possible k-mers extracted from the set, where I represents the set of all objects that satisfy the "condition". Extract each DNA sequence fragment s and generate a triplet for it. All these triples together constitute a fingerprint set.

[0044] The formula for calculating the confidence level of local alignment is as follows:

[0045] ;

[0046] Where Confidence(r) is the local alignment confidence score of read segment r, S(r,k) is the set of consecutive k-mers that read segment r is decomposed into, and w(s) i ) represents the weight of the i-th k-mer, I(s) i ∈F assigned (k) represents the set of k-mer fingerprints of the i-th k-mer belonging to the parent to which the initial reading segment r is assigned. If true, return 1; otherwise, return 0. α is the penalty coefficient, and I(s) i ∈F other (k) returns 1 if the set of k-mer fingerprints of the i-th k-mer belonging to another parent (the parent to which the reading segment r is initially assigned) is true, otherwise returns 0.

[0047] Step S23: Perform temporal trajectory encoding processing based on the skeletal muscle development-related allelic expression sequences. By statistically analyzing the sequence abundance ratio from the parent and parent at each time node, and based on their order and increase / decrease relationship at different developmental nodes, construct a three-dimensional dynamic trajectory of allelic expression to obtain the skeletal muscle allelic expression trajectory tensor.

[0048] Understandably, this step first systematically organizes the quality-controlled skeletal muscle development-related allelic expression sequences along the time dimension. For each preset developmental time node (e.g., embryonic day 16, hatching day 1, day 7, etc.), the read counts from the paternal and maternal parents are counted separately. The precise allelic expression ratio at that time point is obtained by calculating the ratio of paternal to maternal sequences. Based on this, the continuous changing trend of the ratio between different developmental nodes is analyzed, with particular attention to the expression dominance shift that occurs in specific stages such as the critical period of myofibril formation (embryonic day 16-18) and the myotube fusion period (hatching day 1-hatching day 3). Next, the three dimensions—time dimension (continuous developmental nodes), gene dimension (set of genes related to skeletal muscle development), and allelic expression ratio dimension (paternal / maternal expression ratio)—are systematically integrated to construct a three-dimensional dynamic trajectory of allelic expression using tensor modeling. This trajectory not only records the expression state of each gene at a single time point, but more importantly, captures the dynamic alternation of paternal and maternal allelic expression dominance throughout the entire developmental timeline. The resulting skeletal muscle allele expression trajectory tensor, as a structured data container, provides a complete mathematical representation basis for subsequent analysis of the spatiotemporal dynamic characteristics of allele interactions.

[0049] The formula for calculating the allele expression ratio is shown below:

[0050] ;

[0051] Where AER(g,t) is the allelic expression ratio of gene g at a specific developmental time point t, and Count pat (g,t) represents the number of high-quality sequencing reads at time point t that uniquely match gene g and are confirmed to originate from the father. mat (g,t) represents the number of high-quality sequencing reads that uniquely match gene g at time point t and are confirmed to originate from the maternal parent. is a smoothing factor, taken as 0.5, and log2 is a logarithmic transformation to base 2.

[0052] Step S3: Perform nonlinear temporal embedding and pattern recognition processing on the skeletal muscle allelic expression trajectory tensor, capture the long-term dependence of allelic expression between different developmental nodes through a preset neural network, and apply a hidden Markov model to identify the transition probability of expression state and classify the expression patterns to obtain the non-additive effect gene expression pattern characterization corresponding to each parent-parent combination.

[0053] Understandably, this step first utilizes a pre-defined neural network structure to perform nonlinear temporal embedding processing on the skeletal muscle allele expression trajectory tensor. This process effectively captures the complex dependencies between paternal and maternal allele expression at continuous time points from embryonic development to post-hatching growth, especially those long-term interaction patterns spanning multiple developmental stages. Based on this, a Hidden Markov Model is applied to model gene expression states, identifying the transition probabilities between different expression states, such as dominant expression, co-expression, and repression, thereby quantifying the dynamic characteristics of allele interactions. Finally, by classifying these temporal and state characteristics, gene expression patterns exhibiting significant non-additive effects in hybrid offspring are screened and characterized. In this step, step S3 includes steps S31, S32, and S33.

[0054] Step S31: Perform developmental time-series dynamic embedding processing based on the skeletal muscle allele expression trajectory tensor. Using a pre-defined bidirectional deep neural network with gated recurrent units, capture the periodic cooperative and competitive relationship between the paternal and maternal allele expression intensities at pre-defined specific nodes to obtain a time-series dynamic embedding representation.

[0055] Understandably, this step uses a pre-defined bidirectional deep neural network with gated recurrent units to extract deep features from the time-series data in the tensor. The gated recurrent unit, as a recurrent neural network structure, dynamically controls the flow of information through its internal update and reset gate mechanisms, effectively capturing long-term dependencies in the time series and avoiding the gradient vanishing problem in long-sequence training of traditional recurrent networks. The bidirectional deep neural network analyzes the time series simultaneously from both forward (from early to late) and backward (from late to early) directions, comprehensively analyzing allele expression. The dynamic change patterns during development; for the skeletal muscle development scenario in poultry, this treatment pays special attention to the periodic synergistic (e.g., synchronous enhancement of expression) and competitive (e.g., alternating expression dominance) relationships between paternal and maternal allele expression intensities during specific stages such as the critical period of myofibril formation (16-18 days of embryonic development) and the period of myotube fusion (1-3 days after hatching); through hierarchical learning of the network, the high-dimensional allele expression trajectory is transformed into a low-dimensional temporal dynamic embedding representation, which condenses the spatiotemporal features of allele interactions and provides structured input for subsequent state transition modeling.

[0056] The formula for the pre-defined bidirectional deep neural network with gated recurrent units is shown below:

[0057] ;

[0058] Where, r t To reset the gate vector, σ is the sigmoid activation function, and W... xr W xz and W xh Let x be the weight matrix input to the reset gate, update gate, and candidate hidden space, respectively. t For the input vector, W hr W hz and W hh h represents the weight matrix from the hidden state to the reset gate, update gate, and candidate hidden space, respectively. t-1 b is the hidden state from the previous time step. r b z and b h For the bias term, z t To update the gate vector, Let be the candidate hidden state, and tanh be the hyperbolic tangent activation function. For element-wise multiplication (Hadamard product), h t The state is hidden at time step t.

[0059] The transformation formula for the temporal dynamic embedding representation is shown below:

[0060] ;

[0061] Among them, e (j) For the temporal dynamic embedding representation of gene j, N t The total number of time steps. Let be the hidden state of gene j at time step t.

[0062] Step S32: Perform equipotential expression state transition modeling processing based on the temporal dynamic embedding representation, and establish the dominant-cooperative-inhibitory state transition probability matrix by applying a hidden Markov model;

[0063] Understandably, this step establishes a three-state system of "dominance-cooperation-suppression" using a Hidden Markov Model. The dominant state indicates that a particular parental allele consistently holds an expression advantage; the cooperative state reflects a balanced interaction between parental alleles; and the suppression state represents a significant suppression of the expression of a specific allele. The model calculates state transition probabilities, paying particular attention to state transition patterns during key developmental stages, from embryonic days 16-18 (myofibrillar formation) to chicks 1-3 days after hatching (myofibrillar fusion). Considering the specific characteristics of reciprocal crosses in poultry, this modeling also takes into account asymmetric transition patterns that may occur when highly selective breeds are crossed with local breeds. Specifically, the probability of transition from paternal dominance to cooperative states may differ from that from maternal dominance to cooperative states. The resulting transition probability matrix not only quantifies the likelihood of transitions between states but, more importantly, reveals the evolutionary patterns of allele interaction patterns along the developmental timeline, providing a theoretical basis for identifying key time windows for heterosis formation.

[0064] The formula for calculating the state transition probability is as follows:

[0065] ;

[0066] in, ξ represents the re-estimated state transition probability. t (a,b) represents the posterior probability of transitioning from state a to state b at time t, γ t (a) represents the posterior probability of being in state a at time t;

[0067] Step S33: Perform non-additive effect pattern recognition processing based on the state transition probability matrix, classify the expression patterns using a spectral clustering algorithm, and obtain the non-additive effect gene expression pattern representation corresponding to each parent-parent combination.

[0068] Understandably, this step uses spectral clustering algorithms to perform unsupervised classification of state transition patterns. First, a similarity matrix is ​​constructed to measure the similarity of different parent-child combinations in state transition sequences. The similarity calculation is based on the distribution characteristics of transition probabilities rather than Euclidean distance, thus effectively capturing nonlinear relationships. Next, the high-dimensional state space is mapped to a low-dimensional feature space through Laplace matrix eigenvalue decomposition, and the distribution characteristics of feature vectors are used to identify natural groupings. For the poultry skeletal muscle development scenario, this processing pays special attention to non-additive patterns that appear in state transitions, such as certain hybrid combinations exhibiting a continuous synergistic state (indicating heterosis) or frequent state switching (indicating genetic conflict) from late embryonic to early post-hatching. Finally, through semantic annotation of the clustering results, a characterization of the non-additive effect gene expression pattern corresponding to each parent-child combination is obtained. This characterization not only distinguishes the categories of expression patterns but also reveals the dynamic characteristics of allele interactions on the developmental timeline, providing structural insights into understanding the molecular basis of heterosis.

[0069] The formula for calculating the similarity of the parent-child combination in the state transition sequence is as follows:

[0070] ;

[0071] Among them, S ab Let A be the similarity value between parent combination a and parent combination b. (a) and A (b) These represent the state transition probability matrices for parent combination a and parent combination b, respectively. Let be the Jensen-Shannon divergence between the state transition probability matrices of parent combination a and parent combination b, β be the scaling parameter, and exp denote the exponential function.

[0072] Step S4: Based on the non-additive effect gene expression pattern characterization, leg muscle weighing records, and tibia length measurement records, construct a trait-gene dynamic association network and screen out key regulatory pathways associated with skeletal muscle trait development.

[0073] Understandably, this step integrates the characterization of non-additive gene expression patterns with leg muscle weighing and tibia length measurement records in a multi-dimensional manner. By constructing a trait-gene dynamic association network, it reveals the intrinsic link between molecular-level expression changes and macroscopic trait formation. In this process, the potential time lag effect between gene expression and phenotypic expression during skeletal muscle development is specifically considered; that is, changes in the expression of specific genes may require a certain developmental stage before being reflected in changes in leg muscle weight or tibia length. By analyzing the topological features and node association strength in the network, key regulatory pathways that maintain stable associations with skeletal muscle trait changes at multiple developmental time points are identified. These pathways reflect the most fundamental causal relationship between gene expression patterns and phenotypic development. In this step, step S4 includes steps S41, S42, and S43.

[0074] Step S41: Based on the non-additive effect gene expression pattern characterization, leg muscle weighing record and tibia length measurement record, perform multi-source spatiotemporal data fusion processing. By dynamically aligning the gene expression pattern with the phenotypic measurement data, and calculating the covariance structure of expression level with leg muscle weight and tibia length in the time dimension, a spatiotemporal correlation feature matrix is ​​obtained.

[0075] Understandably, this step uses a dynamic alignment method to match gene expression patterns with phenotypic measurement data at the same time points (e.g., embryonic days 16 and 18 to hatch days 1 and 7, etc.), ensuring that the gene expression level at each developmental time point precisely corresponds to the corresponding leg muscle weight and tibia length values. Next, the covariance structure of expression levels with leg muscle weight and tibia length over time is calculated. This calculation not only assesses the linear correlation strength between gene expression and phenotypic changes but also captures the co-fluctuation patterns of both in the developmental sequence, such as the temporal correlation between the increase in expression levels and leg muscle weight gain during myofibril formation. For the poultry skeletal muscle development scenario, this processing specifically considers potential measurement errors in phenotypic data and noise in gene expression, smoothing the time series through sliding window covariance analysis to enhance robustness. The final spatiotemporal correlation feature matrix is ​​a multidimensional data structure where rows represent genes or gene clusters, columns represent time points, and element values ​​quantify the correlation strength between gene expression and phenotypic traits at specific developmental stages, providing feature input for subsequent network construction.

[0076] The formula for calculating the covariance structure of expression levels with leg muscle weight and tibia length over the time dimension is shown below:

[0077] ;

[0078] in, The time delay covariance between gene j and phenotypic trait q at time point t with a time delay of d, τ∈W tLet the sliding window τ belong to the set of sliding windows centered at time point t, and g j '(τ) represents the standardized expression level of gene j within the sliding window τ. p represents the average expression level of gene j after normalization within the sliding window τ. q '(τ+d) is the standardized measurement of phenotypic trait q within a sliding window τ+d. It is the average value of the standardized measurements of phenotypic trait q over a sliding window τ+d.

[0079] Step S42: Based on the spatiotemporal correlation feature matrix, a dynamic correlation network is constructed. A weighted network model is constructed based on graph theory principles, where nodes represent non-additive effect gene clusters and phenotypic traits, edge weights represent spatiotemporal correlation strength, and a time delay factor is introduced to capture the lag effect of gene expression on phenotype, thus obtaining a preliminary trait-gene dynamic correlation network.

[0080] Understandably, this step constructs a weighted network model using graph theory principles, defining each non-additive effect gene cluster and phenotypic trait (leg muscle weight, tibia length) as a network node. Node attributes include their dynamic characteristics during development. The calculation of edge weights not only considers the direct correlation values ​​in the spatiotemporal correlation feature matrix but also introduces a time delay factor to capture the lag effect of gene expression on phenotype. For example, changes in the expression of certain genes during the embryonic period may not be reflected in leg muscle weight growth until early post-hatching.

[0081] This step specifically designed a dynamic weighting function, using a sliding time window to analyze the maximum mutual information value between gene expression and phenotypic changes, thus determining the optimal time delay parameter. A threshold filtering strategy was employed during network construction to retain biologically significant strong connections while eliminating weak associations caused by random noise. The resulting preliminary trait-gene dynamic association network not only presents the static association topology between genes and traits, but more importantly, it characterizes the spatiotemporal dynamic causal relationship of "gene expression → phenotypic formation" during development through weighted directed edges, providing a network foundation for subsequent critical path screening.

[0082] The dynamic weight function formula is shown below:

[0083] ;

[0084] Among them, w yq (t) represents the dynamic edge weights between gene cluster y and phenotypic trait q at time point t. To obtain the maximum value among all possible delays d, λ d This is the correlation strength adjustment factor under delay d, taken as the reciprocal of delay d. The absolute value of the association strength between gene cluster y and phenotypic trait q at time point t delayed by d.

[0085] The mutual information calculation is based on existing formulas and will not be elaborated here.

[0086] Step S43: Based on the preliminary trait-gene dynamic association network, key regulatory pathways are screened. By analyzing the modular structure and node centrality index in the network, key regulatory pathways associated with skeletal muscle trait development are identified.

[0087] Understandably, this step first uses a modular detection formula to divide the network into communities, identifying functional modules that are tightly connected in the topology (by comparing with a preset threshold, those greater than the preset threshold are considered tightly connected). These modules typically correspond to combinations of gene clusters and related traits involved in specific skeletal muscle development processes (such as myofibril formation and myotube fusion). Based on this, multiple node centrality indicators (including degree centrality, betweenness centrality, and eigenvector centrality) are comprehensively used to evaluate the importance of network nodes. Degree centrality identifies hub genes with the most connections, betweenness centrality identifies bridge nodes located between different modules, and eigenvector centrality considers the number of nodes' neighbors. Importance: This treatment, targeting the spatiotemporal characteristics of avian skeletal muscle development, focuses on persistently critical nodes that maintain high centrality across multiple developmental time points, as well as critical nodes that appear only at specific developmental stages (such as the critical period of myofibril formation at embryonic days 16-18). Through a comprehensive evaluation of the internal connectivity density and node centrality of modules, gene pathways that are both centrally located in the topological structure and closely related to the development of multiple skeletal muscle traits are screened. The resulting key regulatory pathway representation not only includes gene sequence information within the pathway but also records the activity intensity of the pathway at different developmental stages and its contribution to phenotypic traits, providing precise molecular targets for subsequent hybridization optimization.

[0088] The modular detection formula is shown below:

[0089] ;

[0090] Where Q is the modularity value, which measures the quality of network community partitioning; a larger value indicates a more obvious community structure. M is the sum of the weights of all edges in the network, and m and n both represent nodes in a dynamically connected network. mn The elements of the adjacency matrix represent the edge weights between nodes m and n, and k is the weight of the edge between node m and node n. m and k n Let δ(c) represent the weighted degrees of node m and node n, respectively. m ,c n The Kroneckerdelta function has a value of 1 when node m and node n belong to the same community, and 0 otherwise.

[0091] The formula for calculating weighted degree centrality is as follows:

[0092] ;

[0093] Among them, C D (m) is the degree centrality value of node m, N(m) is the set of neighboring nodes of node m, and E is the total number of nodes in the network;

[0094] The betweenness centrality formula is shown below:

[0095] ;

[0096] Among them, C B (v) represents the betweenness centrality of node v, σ cm (v) represents the number of shortest paths from node v to m, σ cm w is the number of shortest paths from node v to m. cv and w cm These are the weights of the edges from nodes c to v and m, respectively.

[0097] The formula for calculating the eigenvector centrality is as follows:

[0098] ;

[0099] Where, x m Let w be the eigenvector centrality value of node m, λ be the eigenvalue, and w be the eigenvector. mn Let be the weight of the edge from node m to n.

[0100] The formula for determining the key control path is as follows:

[0101] ;

[0102] Where S(P) is the comprehensive score of path P, and a comprehensive score greater than a preset threshold indicates a critical path; A, F, and C are weight coefficients that balance path activity, topological centrality, and phenotypic contribution; |P| is the number of edges in the path; e is the number of edges in the path; and w... e (t) represents the weight of edge e at time point t, where T is the set of development time points, and C is the weight of edge e at time point t. B (v t The betweenness centrality of nodes in the path at time t, I q (P) indicates whether path P is significantly associated with phenotypic trait q (such as leg muscle weight), obtained based on statistical tests.

[0103] Step S5: Optimize and output the pairing scheme based on the key regulatory pathway representation. Explore the potential space composed of allele expression patterns and trait associations through a reinforcement learning model to simulate the dynamic process of allele interaction and obtain the optimal parent pairing result that maximizes skeletal muscle trait performance.

[0104] Understandably, this step constructs a reinforcement learning environment based on key regulatory pathway representations, modeling the genetic interaction effects of different parental combinations as an exploration process in a state space. In this environment, the agent learns an optimal strategy that maximizes skeletal muscle trait performance by simulating the dynamic interaction patterns of alleles during development. The reinforcement learning model effectively explores the potential space composed of allele expression patterns and trait associations through continuous trial and error and strategy updates, ultimately outputting parent pairing results that achieve the optimal balance between genetic stability and trait performance, providing a scientific basis for practical breeding work. In this step, step S5 includes steps S51, S52, and S53.

[0105] Step S51: Based on the key regulatory path representation, construct the hybridization decision environment by measuring the strength, stability and contribution to the phenotype of each key regulatory path as state features and defining the parent candidate combination as an optional action to obtain the initialized Markov decision process model.

[0106] Immediately understandable is that this step transforms these path features into a state space, where the strength of each key regulatory pathway is quantified by the magnitude of its expression level variation over developmental time, stability is assessed by the standard deviation of pathway activity fluctuations at different developmental stages, and contribution to phenotype is measured by the grey relational coefficient between pathway activity and traits such as leg muscle weight and tibia length. Parental candidate combinations are defined as optional actions, each corresponding to a specific allele combination. Based on this, a Markov decision process model is constructed. The state transition probability of this model is determined by the variation in expression patterns of key regulatory pathways under different parental combinations, and the reward function is directly related to the optimization objective of skeletal muscle trait performance. For poultry crossbreeding scenarios, this model specifically considers the asymmetric effects that may occur in reciprocal crosses, i.e., the same pair of parents may produce different regulatory pathway activation patterns under different cross directions. The final initialized Markov decision process model provides a complete mathematical framework for subsequent reinforcement learning optimization, making the crossbreeding decision process quantifiable and optimizable.

[0107] The state transition probability function based on the path activity pattern is shown below:

[0108] ;

[0109] Wherein, P(S) c' |S c ,a c ) indicates that in state S c Next, execute action a c Then, it transitions to the new state S. c' The probability, θ is the temperature parameter, S'' is all possible states in the state space, S is the set of all possible states, D KL (S c ‖S c' The Kullback-Leibler divergence is used to measure the divergence from state S. c To S c' "Information distance", D KL (S c ||S'') is the Kullback-Leibler divergence, used to measure the divergence from state S c The "information distance" to state S''.

[0110] The reward function is as follows:

[0111] ;

[0112] Among them, R(S) c ) indicates that the state is S c The immediate reward obtained, ω1 represents the path activity intensity I(P) c The weighting coefficients of ) The intensity of pathway activity I(P) c The average of the weight coefficients of θ(I(P)), where ω2 represents the path stability θ(I(P)). c The weighting coefficients of )) are ω3, which represents the phenotypic contribution. For pathway activity intensity The maximum value, max(θ), represents the path stability θ(I(P)). c The maximum value of (ρ(I(P)) c ),W)+ρ(I(P c ),L)) / 2 represents the pathway activity I(P) c The mean of the Pearson correlation coefficients between the two phenotypic traits (leg muscle weight W and tibia length L) and the two phenotypic traits (leg muscle weight W and tibia length L).

[0113] Step S52: Based on the Markov decision process model, perform hybridization strategy optimization. Using the deep deterministic strategy gradient algorithm, under the condition of the difference in positive and negative crossover effects between the candidate parent population and the candidate parent population, explore and update the strategy network to obtain the optimized hybridization strategy network.

[0114] Understandably, this step optimizes the policy using a deep deterministic policy gradient algorithm. This algorithm employs an actor-critic framework, where the actor network generates continuous crossbreeding policy actions (i.e., parent pairing selection), and the critic network evaluates the action-value function to guide policy updates. During training, the algorithm implements an exploration mechanism by adding random noise to the action space, allowing the model to try non-intuitive crossbreeding combinations. It also utilizes an experience replay buffer to store state-action-reward sequences to improve learning efficiency. For the specific scenario of the differences in reciprocal cross effects in poultry breeding, this processing is specifically designed to address the issue of non-reciprocal crossbreeding effects. A symmetric reward function is used to evaluate the phenotypic differences of the same parent line in orthogonal crosses (with highly selective varieties as the male parent) and reciprocal crosses (with local varieties as the male parent), enabling the strategy network to learn hybridization asymmetry caused by biological mechanisms such as genomic imprinting or sex-specific expression. Through multiple iterations, the strategy network gradually converges to the optimal strategy that maximizes the long-term cumulative reward (i.e., the skeletal muscle phenotypic optimization goal). The resulting optimized hybridization strategy network not only captures the complex mapping relationship between parent line combinations and phenotypic performance, but also internalizes the regulatory law of the difference in orthogonal and reciprocal cross effects on the hybridization results.

[0115] The loss function of the critic network is shown below:

[0116] ;

[0117] Where, L(θ) Q () represents the loss function of the critic network. For the expected computation of transferred samples sampled from the experience replay buffer D, A t1 Let A be the state at time step t, representing the path feature vector under a specific hybridization combination. t2 In state A t1 The next action to choose, A t3 Indicates an immediate reward, a reward value calculated based on skeletal muscle phenotypic performance, A. t4 Represents the new state after the transition, Q(A) t1 A t2 |θ Q y represents the critic network's prediction of the state-action pair. t This represents the target value of the critic network.

[0118] The policy gradient update formula is shown below:

[0119] ;

[0120] in, The strategy performance index J is related to the actor network parameter θ. μ gradient, This represents the state A obtained by randomly sampling from the experience replay buffer D. t1 , Indicates the target value with respect to action A t2 The gradient indicates the direction for improvement of the action. The policy function represents the network parameter θ. μ The gradient of μ is the gradient of the actor network (policy function), and its parameters are θ. μ .

[0121] Step S53: Perform optimal pairing output processing based on the optimized hybridization strategy network. By cross-validating the parent combination with the highest reward value obtained by the strategy network with the preset historical stable pairing scheme, the parent pairing result that maximizes the performance of leg skeletal muscle traits and is closest to the historical stable pairing scheme is obtained.

[0122] Understandably, this step first utilizes the reasoning ability of the optimized hybridization strategy network to evaluate candidate parental combinations in a simulated hybridization decision-making environment, selecting the pairing schemes with the highest reward values. These reward values ​​comprehensively reflect the predicted skeletal muscle trait performance potential and genetic stability. Subsequently, these high-potential pairing schemes, discovered through reinforcement learning, are cross-validated with pre-set historically stable pairing schemes (typically hybrid combinations validated through long-term field trials and possessing reliable production performance). This validation process is not a simple comparison but rather compares the correlation between the two schemes, using cosine similarity to assess the closeness of their genetic mechanisms. The final output of the optimal parental pairing is the intersection of the above dual selection: it possesses both the theoretical potential to maximize trait performance discovered by the reinforcement learning model and maintains a high degree of consistency with historical successes in terms of genetic action patterns, thus ensuring that breeding decisions are both innovative and groundbreaking while maintaining the predictability and stability required for practical production applications.

[0123] The formula for calculating the optimal parent pairing scheme is as follows:

[0124] ;

[0125] Among them, a * For the optimal parent pairing scheme, The maximization operator represents finding a candidate solution a in set A' that maximizes the objective function value. i a i Candidate pairing scheme, R(a) i ) represents the reward value for the candidate pairing. S(a) represents the maximum reward value in set A'. i ) represents the pairing scheme a iThe maximum similarity with the historical stable scheme is calculated using cosine similarity, where γ is the balance coefficient.

[0126] Example 2:

[0127] like Figure 2 As shown, this embodiment provides an optimal hybridization pairing design system for poultry skeletal muscle traits. (See also...) Figure 2 The system includes an acquisition unit 701, an encoding unit 702, a processing unit 703, a construction unit 704, and an output unit 705.

[0128] The acquisition unit 701 is used to acquire leg skeletal muscle tissue transcriptional sequencing reads, whole genome sequencing reads, leg muscle weighing records, and tibia length measurement records of the candidate paternal and maternal populations at preset time points.

[0129] The coding unit 702 is used to encode the allelic expression trajectory of the leg skeletal muscle tissue transcription sequencing read data and the whole genome sequencing read data. By constructing a trajectory representation that reflects the dominant change in allelic source, the skeletal muscle allelic expression trajectory tensor corresponding to each candidate individual is obtained.

[0130] The processing unit 703 is used to perform nonlinear temporal embedding and pattern recognition processing on the skeletal muscle allelic expression trajectory tensor, capture the long-term dependence of allelic expression between different developmental nodes through a preset neural network, identify the transition probability of expression state by applying a hidden Markov model, and classify the expression patterns to obtain the non-additive effect gene expression pattern characterization corresponding to each parent-parent combination.

[0131] Construction unit 704 is used to construct a trait-gene dynamic association network based on the non-additive effect gene expression pattern characterization, leg muscle weighing records and tibial length measurement records, and to screen out key regulatory pathways associated with skeletal muscle trait development.

[0132] The output unit 705 is used to optimize and output the pairing scheme based on the key regulatory pathway representation. It explores the potential space composed of allele expression patterns and trait associations through a reinforcement learning model, simulates the dynamic process of allele interaction, and obtains the optimal parent pairing result that maximizes skeletal muscle trait performance.

[0133] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0134] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0135] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method of designing optimal cross-pairing of poultry skeletal muscle traits, characterized in that, The method comprises the following steps: Obtaining leg skeletal muscle tissue transcript sequencing read data, whole genome sequencing read data, leg muscle weighing records and tibia length measurement records of candidate paternal populations and candidate maternal populations at a preset time node; Encoding the leg skeletal muscle tissue transcript sequencing read data and the whole genome sequencing read data by allele expression trajectory, obtaining a skeletal muscle allele expression trajectory tensor corresponding to each candidate individual by constructing a trajectory representing dominant changes in allele sources; Performing nonlinear time series embedding and pattern recognition processing on the skeletal muscle allele expression trajectory tensor, capturing long-term dependencies of allele expression between different development nodes by a preset neural network, identifying transition probabilities of expression states by applying a hidden Markov model, and classifying expression patterns to obtain a non-additive effect gene expression pattern representation corresponding to each paternal-maternal combination; Building a trait-gene dynamic correlation network according to the non-additive effect gene expression pattern representation, the leg muscle weighing records and the tibia length measurement records, and screening to obtain a key regulatory path representation associated with skeletal muscle trait development; Optimizing and outputting a mating scheme according to the key regulatory path representation, exploring in a potential space constituted by allele expression patterns and trait correlations by a reinforcement learning model, simulating a dynamic process of allele interaction, and obtaining an optimal parent pairing result that maximizes skeletal muscle trait performance.

2. The poultry skeletal muscle trait optimal cross-pair design method according to claim 1, characterized in that The method comprises the following steps: Encoding the leg skeletal muscle tissue transcript sequencing read data and the whole genome sequencing read data by allele expression trajectory, comprising: Performing parent source mapping processing on the leg skeletal muscle tissue transcript sequencing read data and the whole genome sequencing read data, identifying unique parent allele sources by specifically aligning whole genome sequencing read data of the candidate paternal populations and the candidate maternal populations with each leg skeletal muscle tissue transcript sequencing read data, and obtaining an allele-specific read set of each candidate individual at multiple time nodes; Performing fuzzy alignment disambiguation processing on the allele-specific read set, eliminating read sequences with a confidence value less than or equal to a preset threshold by constructing allele-specific k-mer fingerprints and calculating local alignment confidence values, and obtaining skeletal muscle development-related allele expression sequences with a confidence value greater than the preset threshold; 3. The poultry skeletal muscle trait optimal cross-pair design method according to claim 1, characterized in that Performing time series trajectory encoding processing on the skeletal muscle development-related allele expression sequences, constructing a three-dimensional allele expression dynamic trajectory by statistically analyzing sequence abundance ratios of paternal and maternal sources at each time node and according to their order and increase / decrease relationship between different development nodes, and obtaining the skeletal muscle allele expression trajectory tensor. The method comprises the following steps: Performing nonlinear time series embedding and pattern recognition processing on the skeletal muscle allele expression trajectory tensor, comprising: Performing development time series dynamic embedding processing on the skeletal muscle allele expression trajectory tensor, capturing periodic synergy and competition relationships of paternal and maternal allele expression intensities between preset specific nodes by a preset bidirectional deep neural network with a gating recurrent unit, and obtaining a time series dynamic embedding representation: According to the time sequence dynamic embedding representation, an isogenic expression state transition modeling process is performed, and a dominant-collaborative-inhibitory state transition probability matrix is established by applying a hidden Markov model; According to the state transition probability matrix, a non-additive effect mode recognition process is performed, and the expression mode is classified by a spectral clustering algorithm to obtain a non-additive effect gene expression mode representation corresponding to each parent-mother combination.

4. The poultry skeletal muscle trait optimal cross-pair design method according to claim 1, characterized in that According to the non-additive effect gene expression mode representation, leg muscle weighing record and tibia length measurement record, a trait-gene dynamic correlation network is constructed, including: According to the non-additive effect gene expression mode representation, leg muscle weighing record and tibia length measurement record, a multi-source space-time data fusion process is performed, the gene expression mode is dynamically aligned with the phenotype measurement data, and the covariance structure of the expression amount, leg muscle weight and tibia length in the time dimension is calculated to obtain a space-time correlation feature matrix; According to the space-time correlation feature matrix, a dynamic correlation network construction process is performed, a weighted network model is constructed based on the graph theory principle, the nodes represent non-additive effect gene clusters and phenotype traits, the edge weight represents the space-time correlation strength, and a time delay factor is introduced to capture the lag effect of gene expression on the phenotype to obtain a preliminary trait-gene dynamic correlation network; According to the preliminary trait-gene dynamic correlation network, a key regulation path screening process is performed, the modular structure and node centrality index in the network are analyzed, and a key regulation path representation associated with the skeletal muscle trait development is identified.

5. The poultry skeletal muscle trait optimal cross-pair design method according to claim 1, wherein According to the key regulation path representation, a mating scheme optimization and output process is performed, including: According to the key regulation path representation, a hybrid decision environment construction process is performed, the strength, stability and contribution of each key regulation path to the phenotype are quantified as state characteristics, and the parent candidate combination is defined as a selectable action to obtain an initialized Markov decision process model; According to the Markov decision process model, a hybrid strategy optimization process is performed, and the strategy network is explored and updated under the condition that the candidate father group and the candidate mother group have different positive and negative cross effect differences by using a deep deterministic policy gradient algorithm to obtain an optimized hybrid strategy network; According to the optimized hybrid strategy network, an optimal mating output process is performed, and the parent combination with the highest reward value explored by the strategy network is cross-validated with a preset historical stable mating scheme to obtain a parent mating result that can maximize the leg skeletal muscle trait performance and is closest to the historical stable mating scheme.

6. A poultry skeletal muscle trait optimal cross-pairing design system, characterized by, including: An acquisition unit is configured to acquire leg skeletal muscle tissue transcript sequencing read data, whole genome sequencing read data, leg muscle weighing record and tibia length measurement record of a candidate father group and a candidate mother group at a preset time node; An encoding unit is configured to perform isogenic expression trajectory encoding processing on the leg skeletal muscle tissue transcript sequencing read data and the whole genome sequencing read data, to obtain a skeletal muscle isogenic expression trajectory tensor corresponding to each candidate individual by constructing a trajectory representation reflecting the dominant change of isogenic sources. The processing unit is configured to perform nonlinear time series embedding and pattern recognition processing on the skeletal muscle allelic expression trajectory tensor, capture long-term dependencies of allelic expression between different development nodes by using a preset neural network, apply a hidden Markov model to identify transition probabilities of expression states, classify expression patterns, and obtain a non-additive effect gene expression pattern representation corresponding to each parent-offspring combination. The construction unit is configured to perform trait-gene dynamic association network construction based on the non-additive effect gene expression pattern representation, leg muscle weight record, and tibia length measurement record, and filter to obtain a key regulatory path representation associated with skeletal muscle trait development. The output unit is configured to perform optimization and output processing of a mating scheme based on the key regulatory path representation, explore a potential space constituted by allelic expression patterns and trait associations by using a reinforcement learning model, simulate a dynamic process of allelic interaction, and obtain an optimal parent-offspring pairing result that maximizes skeletal muscle trait performance.

7. The poultry skeletal muscle trait optimal cross-pair design system of claim 6, wherein, The encoding unit includes: The first encoding sub-unit is configured to perform parent source mapping processing on the leg skeletal muscle tissue transcription sequencing read data and whole genome sequencing read data, identify unique parent allelic sources by specifically aligning whole genome sequencing read data of candidate parent and candidate mother groups with each leg skeletal muscle tissue transcription sequencing read data, and obtain allelic-specific read set of each candidate individual at multiple time nodes. The second encoding sub-unit is configured to perform fuzzy alignment disambiguation processing on the allelic-specific read set, remove read data with a confidence value less than or equal to a preset threshold by constructing allelic-specific k-mer fingerprints and calculating local alignment confidence values, and obtain skeletal muscle development-related allelic expression sequences with a confidence value greater than the preset threshold. The third encoding sub-unit is configured to perform time series trajectory encoding processing on the skeletal muscle development-related allelic expression sequences, construct a three-dimensional allelic expression dynamic trajectory by statistically analyzing sequence abundance ratios of the parent and the mother at each time node and according to the order and change relationship between different development nodes, and obtain the skeletal muscle allelic expression trajectory tensor.

8. The poultry skeletal muscle trait optimal cross-pair design system of claim 6, wherein, The processing unit includes: The first processing sub-unit is configured to perform development time series dynamic embedding processing on the skeletal muscle allelic expression trajectory tensor, capture periodic synergy and competition relationships between expression intensities of the parent and the mother at preset specific nodes by using a preset bidirectional deep neural network with a gated recurrent unit, and obtain a time series dynamic embedding representation. The second processing sub-unit is configured to perform allelic expression state transition modeling processing on the time series dynamic embedding representation, and establish a dominant-synergistic-inhibitory state transition probability matrix by using a hidden Markov model. The third processing sub-unit is configured to perform non-additive effect pattern recognition processing on the state transition probability matrix, classify expression patterns by using a spectral clustering algorithm, and obtain a non-additive effect gene expression pattern representation corresponding to each parent-offspring combination.

9. The poultry skeletal muscle trait optimal cross-pair design system of claim 6, wherein, The construction unit includes: The first construction subunit is configured to perform multi-source spatiotemporal data fusion processing according to the non-additive effect gene expression pattern characterization, leg muscle weighing record and tibia length measurement record, to obtain a spatiotemporal correlation feature matrix by dynamically aligning the gene expression pattern with the phenotype measurement data and calculating the covariance structure of the expression amount, the leg muscle weight and the tibia length in the time dimension; The second construction subunit is configured to perform dynamic correlation network construction processing according to the spatiotemporal correlation feature matrix, to obtain a preliminary trait-gene dynamic correlation network by constructing a weighted network model based on the graph theory principle, wherein the nodes represent the non-additive effect gene clusters and the phenotypic traits, the edge weight represents the spatiotemporal correlation strength, and a time delay factor is introduced to capture the lag effect of the gene expression on the phenotype. The third construction subunit is configured to perform key regulatory path screening processing according to the preliminary trait-gene dynamic correlation network, to obtain a key regulatory path representation associated with the skeletal muscle trait development by analyzing the modular structure and the node centrality index in the network.

10. The poultry skeletal muscle trait optimal cross-pair design system of claim 6, wherein, The output unit includes: The first output subunit is configured to perform hybrid decision environment construction processing according to the key regulatory path representation, to obtain an initialized Markov decision process model by quantifying the strength, stability and contribution to the phenotype of each key regulatory path as state features and defining the parent candidate combinations as selectable actions; The second output subunit is configured to perform hybrid strategy optimization processing according to the Markov decision process model, to obtain an optimized hybrid strategy network by exploring and updating the strategy network through the deep deterministic policy gradient algorithm under the condition of the difference in the positive and negative cross effect of the candidate paternal population and the candidate maternal population; The third output subunit is configured to perform optimal pairing output processing according to the optimized hybrid strategy network, to obtain a parent pairing result that can maximize the leg skeletal muscle trait performance and is closest to the historical stable pairing scheme by cross-verification of the parent combination with the highest reward value obtained by the strategy network exploration and the preset historical stable pairing scheme.

Images