A method and system for constructing large-scale system evolution trees

Abstract

This invention discloses a method and system for constructing large-scale phylogenetic trees. First, based on the tree diameter of empirical phylogenetic trees in a publicly available database and the statistical characteristics of site heterogeneity in real amino acid multiple sequence alignments, simulated data is generated to construct a supervised learning dataset. For amino acid multiple sequence alignment data of large-scale taxonomic units, a trunk tree is constructed and sequences are placed. Connectivity partitioning is performed based on the topology, partition size, and structural span constraints of the placed tree, and partition topology trees and representative information dictionaries are extracted. Using a two-stage deep learning model based on Transformer and residual convolutional neural networks, combined with bundle search and adjacency join, phylogenetic subtrees for each partition are independently constructed. Finally, based on the partition topology trees and representative information dictionaries, a global phylogenetic tree is constructed through root re-determination, anchor point localization, and recursive concatenation. This invention combines divide-and-conquer with deep learning techniques, effectively balancing the accuracy, efficiency, and scalability of large-scale tree construction.

Description

Technical Field

[0001] This invention belongs to the fields of bioinformatics and deep learning technology, specifically relating to a method and system for constructing large-scale system evolution trees. Background Technology

[0002] A phylogenetic tree is a tree-like structural model depicting the evolutionary relationships between species or biological sequences. It is an important tool for studying the lineage history of life and revealing the kinship between taxa. A phylogenetic tree consists of nodes and branches. Leaf nodes represent existing or known taxa, such as species or genome sequences; internal nodes represent common ancestors that have not been directly observed but are inferred; branches connect different nodes and represent the accumulated genetic changes in the lineage during evolution. Branch length is usually expressed as the expected number of substitutions per amino acid or nucleotide sequence site, used to quantitatively measure the cumulative degree of evolutionary differences. Currently, researchers have proposed various phylogenetic tree construction methods based on multiple sequence alignment (MSA), mainly including statistical inference-based methods and deep learning-based methods.

[0003] Generally, methods for constructing system evolution trees based on statistical inference can be divided into two categories: 1) Distance-based statistical inference methods first estimate the pairwise distances between taxa using MSA data. Then, based on the distance matrix, a phylogenetic tree is searched in the tree space such that the inter-leaf path lengths induced by this tree are as consistent as possible with these distances. These methods mainly include Unweighted Group Average (UPGMA), Fastest Evolution (FastME), and Neighbor-Joining (NJ).

[0004] 2) Feature-based statistical inference methods directly use the positional character states in the MSA (Sequence Alignment Analysis) as the analysis object to characterize the evolutionary process between sequences at the positional level. These methods explicitly model the sequence evolution process in tree space, relying on predefined models, parameter sets, and evaluation metrics to assess the fit of a candidate phylogenetic tree to a given MSA. They search the tree space and select the phylogenetic tree that best suits the MSA under the given evaluation metrics. These methods mainly include maximum parsimony, maximum likelihood estimation, and Bayesian inference.

[0005] Deep learning-based phylogenetic tree construction methods are a type of data-driven approach. During the training phase, the simulated sequence alignment (MSA) is typically used as input, and the simulated phylogenetic tree or its derived information serves as the supervisory signal. A neural network model automatically learns statistical patterns of site patterns, covariant structures, and evolutionary signals from multiple sequence alignments, establishing a mapping between multiple sequence alignments and the phylogenetic tree. In the inference phase, deep learning methods can quickly obtain results through a single forward propagation. Currently, deep learning-based phylogenetic tree construction methods are still under development, and researchers have proposed several representative algorithms, such as the Tree_learning algorithm, the Fusang algorithm, and the Phyloformer algorithm.

[0006] The above-mentioned system evolution tree construction method usually has the following problems in practical applications: 1) Distance-based statistical inference methods reduce computational complexity by compressing MSA into a distance matrix between classification units, which has good computational efficiency and scalability. However, its tree building process relies on local merging criteria. When there is a deviation in distance estimation, it is difficult to make global corrections in subsequent steps. When the number of classification units increases or the evolutionary relationship becomes more complex, the topological accuracy and branch length accuracy of the system evolution tree are limited.

[0007] 2) Feature-based statistical inference methods, while having high reliability in terms of construction accuracy and statistical interpretability, usually rely on predefined evolutionary models and their parameters, and need to search and optimize in a huge tree space. As the number of classification units increases, the computational complexity rises rapidly, and in large-scale scenarios, they face problems such as long running time, high consumption of computing resources and insufficient scalability.

[0008] 3) Deep learning-based system evolution tree construction methods have certain advantages in feature representation capabilities and inference speed, but existing methods can only handle a limited scale of classification units, lack scalability, and there is still room for further improvement in the accuracy of system evolution tree construction.

[0009] In summary, existing methods for constructing system evolution trees struggle to balance efficiency, accuracy, and scalability: distance-based statistical inference methods offer good scalability but have limited quality in system evolution trees; feature-based statistical inference methods offer high accuracy but incur significant computational costs; and deep learning-based methods offer fast inference speeds, but their performance in constructing system evolution trees on large scales still needs improvement.

[0010] Therefore, how to balance tree construction accuracy, computational efficiency, and scalability in large-scale system evolution tree construction tasks is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0011] The technical problem to be solved by the present invention is to provide a method and system for constructing large-scale system evolution trees, which addresses the shortcomings of the prior art and solves the technical problem that existing system evolution tree construction methods are unable to balance tree construction accuracy, computational efficiency and scalability in large-scale system evolution tree construction tasks.

[0012] The present invention adopts the following technical solution: A method for constructing large-scale system evolution trees includes the following steps: Based on the statistical characteristics of tree diameter of empirical system evolution trees in public databases and the statistical characteristics of site heterogeneity parameters of real amino acid multiple sequence alignment data, a system evolution tree and corresponding amino acid multiple sequence alignment data are simulated and generated. The amino acid multiple sequence alignment data is used as the input for supervised learning, and the corresponding system evolution tree is used as the label for supervised learning. A supervised learning dataset for training and testing the system evolution subtree inference model is constructed. For amino acid multiple sequence alignment data of large-scale taxonomic units, a trunk tree is given or constructed based on the amino acid multiple sequence alignment data. Non-trunk sequences are placed into the trunk tree to obtain a placement tree. Based on the topology, partition size constraints and structural span constraints of the placement tree, all sequences are connected to partition, so that each partition corresponds to a connected subtree on the placement tree. At the same time, the hierarchical topological relationship between partitions and the representative information of adjacent partitions are extracted to obtain the sequence partitioning result, partition topology tree and representative information dictionary. The system evolution subtree construction model is trained based on the supervised learning dataset, and the trained system evolution subtree construction model is used to independently construct system evolution subtrees for each partition. The system evolution subtree construction model includes an evolution distance prediction model, a distance matrix optimization model, and a bundle search and adjacency method. The system evolution subtree construction model first uses the evolution distance prediction model to predict the pairwise evolution distance matrix between classification units within the partition, then uses the distance matrix optimization model to optimize the pairwise evolution distance matrix, and uses bundle search and adjacency method to construct the system evolution subtree of the corresponding partition based on the optimized evolution distance matrix. Based on the partition topology tree and the representative information dictionary, the evolutionary subtrees of each partition system are spliced ​​together. By re-rooting, determining splicing anchor points, pruning duplicate nodes, and recursively splicing, a global system evolution tree containing all classification units and without duplicate classification units is constructed.

[0013] Preferably, the process of constructing the supervised learning dataset includes: By referencing empirical system evolution trees and real amino acid multiple sequence alignment data in publicly available databases, the parameter distribution used to simulate system evolution trees and amino acid multiple sequence alignment data is determined. System evolution trees and corresponding amino acid multiple sequence alignment data are generated through simulation, and the resulting dataset is divided into training set, validation set, and test set.

[0014] Preferably, the process of simulating and generating a phylogenetic tree and the corresponding amino acid multiple sequence alignment data includes: A basic system evolution tree is generated using the birth-death process, and the branch length of the basic system evolution tree is adjusted by introducing evolution rate heterogeneity to obtain a non-overestimation simulated system evolution tree. The branch length of the obtained simulated system evolution tree is scaled by referring to the diameter distribution of empirical system evolution trees, and excessively short external branches are adjusted. Based on the replacement model and the adjusted system evolution tree, the corresponding amino acid multiple sequence alignment data is simulated using bioinformatics software.

[0015] Preferably, the process of constructing a backbone tree and phylogenetic placement of amino acid multiple sequence alignment data for large-scale taxonomic units includes: Given a trunk tree, or constructing a trunk tree based on some representative sequences in the amino acid multiple sequence alignment data; then using a system evolution placement algorithm based on the weighted least squares criterion, non-trunk sequences are placed into the trunk tree according to the distance fitting results between the non-trunk sequences and the trunk tree, thus obtaining the placement tree.

[0016] Preferably, the process of connecting partitions includes: A post-order traversal of the placement tree is performed. The total number of sequences contained in the current local region is calculated from bottom to top as the partition size, and the sum of the distances from the current node to the farthest leaf node in different sub-directions is calculated as the structural span. When the partition size of the current region does not exceed a preset threshold, or the structural span does not reach a preset threshold, the current region is kept in the same partition. Otherwise, a new partition boundary is formed at the corresponding position, and the corresponding subtree is divided. Among them, the set of edges with the same partition number and connected on the placement tree constitutes a partition, and the sequence corresponding to the edge set constitutes the sequence set of the partition.

[0017] Preferably, the process of extracting the hierarchical topological relationships between partitions and representative information of adjacent partitions includes: When the parent node in the placement tree belongs to a different partition than its child node, the parent-child hierarchical relationship between the corresponding partitions is recorded as a hierarchical topological relationship between the partitions, and a partition topology tree with the partition number as the node is constructed. At the same time, representative subtrees in the upward and downward directions are constructed at the partition boundaries. The representative subtrees are composed of representative classification units near the boundary and their connection relationships, and the corresponding representative subtrees are stored in the representative information dictionary.

[0018] Preferably, the process of constructing a system evolution subtree construction model for each partition independently includes: Amino acid multiple sequence alignment data of the corresponding partitions are extracted and one-hot encoded. An evolutionary distance prediction model based on Transformer is used to extract and characterize the encoded amino acid multiple sequence alignment data to obtain the pairwise evolutionary distance matrix between taxonomic units within the partition. A distance matrix optimization model based on residual convolutional neural network is used to correct the pairwise evolutionary distance matrix to obtain the optimized evolutionary distance matrix.

[0019] Preferably, the distance matrix optimization model uses a joint loss function to optimize the neural network parameters. The joint loss function includes a distance regression loss to constrain the error between the optimized evolutionary distance matrix and the true evolutionary distance matrix, a local Q-ranking loss to constrain the ranking relationship of candidate merging clusters, a residual regularization term to limit the magnitude of distance matrix correction, and a relative change regularization term to limit the relative change ratio of the distance matrix before and after optimization.

[0020] Preferably, the process of constructing the system evolution subtree of the corresponding partition using bundle search and adjacency method based on the optimized evolution distance matrix includes: A merging Q matrix for candidate cluster pairs is constructed according to the merging criteria of the adjacency method, and the candidate cluster pairs are sorted in ascending order of Q value. The top K candidate cluster pairs are selected from the top-ranked candidate cluster pairs, and E candidate cluster pairs are selected as the expansion objects of the current state according to the preset cluster pair selection strategy. Merging operations are performed on all expansion objects to generate multiple candidate intermediate states. The candidate intermediate states are sorted in ascending order of path score, and the top B candidate intermediate states are retained for the next round of search. The above process is repeated until the classification units in the current partition are merged, and the system evolution subtree of the current partition is obtained.

[0021] Secondly, embodiments of the present invention provide a large-scale system evolution tree construction system, the system comprising: The supervised learning dataset construction module is used to simulate and generate system evolution trees and corresponding amino acid multiple sequence alignment data based on the statistical characteristics of empirical system evolution trees and real amino acid multiple sequence alignment data in public databases, and to construct supervised learning datasets. The partitioning module, connected to the supervised learning dataset construction module, is used to construct the backbone tree, place the system evolution, and partition the amino acid multiple sequence alignment data of large-scale classification units, and extract the partition topology tree and representative information dictionary. The system evolution subtree construction module is connected to the partitioning module and is used to perform deep learning-based system evolution subtree construction on each partition to obtain the system evolution subtree of each partition; The splicing module, connected to the partitioning module and the system evolution subtree construction module, is used to recursively splice the system evolution subtrees of each partition according to the partition topology tree and the representative information dictionary to obtain a global system evolution tree that contains all classification units and does not contain duplicate classification units.

[0022] Thirdly, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described large-scale system evolution tree construction method.

[0023] Fourthly, embodiments of the present invention provide a computer-readable storage medium including a computer program, which, when executed by a processor, implements the steps of the above-described large-scale system evolution tree construction method.

[0024] Fifthly, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described large-scale system evolution tree construction method.

[0025] In a sixth aspect, embodiments of the present invention provide an electronic device, including a computer program, which, when executed by the electronic device, implements the steps of the above-described large-scale system evolution tree construction method.

[0026] Compared with the prior art, the present invention has at least the following beneficial effects: A method for constructing large-scale phylogenetic trees addresses the scarcity of real labels and enhances the model's generalization ability by simulating the statistical characteristics of real biological data to build a supervised dataset. It utilizes connected component partitioning to decompose the large-scale global problem into controllable local subproblems, significantly reducing computational complexity. Simultaneously, it preserves global structural information using partitioned topology trees and representative information dictionaries, ensuring the accuracy of subsequent concatenation. A two-stage model combining Transformer prediction, residual convolution optimization, and bundle search and adjacency methods leverages the powerful feature extraction capabilities of deep learning while retaining the fast tree-building advantages of distance-based methods, and mitigates the error accumulation of greedy algorithms through bundle search. A recursive concatenation strategy ultimately achieves the construction of a global tree with no repetition and full coverage. In tests with 1000 classification units, this invention outperforms IQ-TREE and NJ in terms of running speed, tree topology accuracy, and branch length accuracy.

[0027] Furthermore, simulation parameters were determined by referencing empirical statistical characteristics from publicly available databases such as Hug's Tree of Life and HOGENOM, rather than by random setting. This makes the simulated evolutionary tree and sequence data more closely resemble real biological evolution in terms of branch length distribution and amino acid composition. By introducing rate change events and reasonable birth-death processes, complex evolutionary history and population structure were simulated. Using this high-quality dataset as training input allows the trained model to maintain high prediction accuracy when faced with unseen real biological data, avoiding the problem of overfitting on simulated data and failing on real data. Clearly defining training, validation, and test sets provides a standard procedure for the scientific evaluation and hyperparameter tuning of the model, ensuring the reliability of the system evolutionary subtree inference model.

[0028] Furthermore, by introducing evolutionary rate heterogeneity to adjust branch lengths, a non-overmetric simulated tree was generated. This aligns with the objective fact that different lineages exhibit inconsistent evolutionary rates in real biological evolution, making it more representative than a simple overmetric tree. Branch lengths were scaled based on the diameter distribution of empirical data, and excessively short outer branches were adjusted to ensure the simulated tree's topological scale remained consistent with trees in the real database. Combining the replacement model and the AliSim tool, amino acid multiple sequence alignment data were generated based on the adjusted tree. This ensured that the tree-to-sequence generation process strictly adhered to bioinformatics principles, providing high-quality input-label pairs for the deep learning model, which is the cornerstone of the model's ability to accurately predict evolutionary distances.

[0029] Furthermore, it supports building from scratch, specifying the trunk, or directly using an existing tree, adapting to different application scenarios. It employs the Apples-2 algorithm based on the weighted least squares criterion. This algorithm is computationally efficient and accurate when placing non-trunk sequences into the trunk tree, quickly determining the optimal insertion position of a new sequence in the evolutionary tree. The placement tree obtained through sequence placement not only includes all sequences but also preserves their relative evolutionary relationships. This provides accurate topological basis for subsequent connected component partitioning, ensuring that partitioning is performed on the correct evolutionary path, avoiding partition chaos caused by incorrect sequence positions, and guaranteeing the local accuracy of subsequent subtree construction.

[0030] Furthermore, partition size constraints and structural span constraints are introduced. Size constraints prevent excessively large single partitions from causing computational explosion, while structural span constraints prevent partitions from having excessive spans in the evolutionary tree, leading to topological breaks. A bottom-up post-order traversal approach is adopted for computation and partitioning, which is logically rigorous and ensures that each partition is a connected subtree in the placement tree. The partitioning conditions are relaxed when specific conditions are met, increasing the algorithm's flexibility. This partitioning method transforms large-scale problems into multiple equally sized, structurally independent subproblems, greatly enhancing the potential for parallel processing and overall tree-building efficiency.

[0031] Furthermore, a partition topology tree is constructed with partition numbers as nodes, clarifying the parent-child hierarchical relationships between subtrees. This ensures that during the splicing phase, the algorithm knows which subtrees to splice first and which to splice later, guaranteeing the ordered nature of the global construction. Representative upward and downward subtrees are constructed and stored in a dictionary. These representative subtrees, composed of classification units near the boundaries, act as interfaces or anchors between partitions. During subsequent splicing, this representative information allows for precise location of the splicing position and pruning of duplicate nodes. This solves the boundary effects and information loss problems common in divide-and-conquer algorithms, ensuring that the final spliced ​​global tree is topologically continuous and conflict-free.

[0032] Furthermore, by leveraging the axial self-attention mechanism of the Transformer, long-range dependencies and global covariance information in multi-sequence alignment can be captured. This allows for a better understanding of the complex evolutionary relationships between sequences, thus predicting a more accurate initial pairwise evolutionary distance matrix. A residual convolutional neural network is introduced to correct the initial distance matrix. Convolutional layers excel at extracting local features, while the residual structure helps train deep networks and prevents gradient vanishing. This makes the final output evolutionary distance matrix numerically closer to the true value, providing high-quality input for subsequent adjacency-based tree construction and significantly improving tree construction accuracy.

[0033] Furthermore, the distance regression loss ensures the numerical consistency between the optimized distance matrix and the true matrix; the local Q-sorting loss innovatively incorporates the merging order of subsequent adjacency-based tree construction into the loss function, guiding the model to learn the correct node merging priority; and the two regularization terms effectively limit the correction magnitude of the distance matrix, avoiding over-correction and overfitting problems. This makes the optimized distance matrix more compatible with the adjacency-based tree construction logic, fundamentally improving the topological accuracy of the system's evolving subtrees.

[0034] Furthermore, by selecting multiple candidates from the top-ranked cluster pairs based on Q-values ​​for expansion and retaining the top B optimal intermediate states for the next round of search, the irreversible impact of early erroneous merging on the final tree structure is effectively avoided; both deterministic and random expansion modes take into account search efficiency and path diversity; the path score ranking mechanism ensures the retention of the optimal path; while maintaining the computational efficiency advantage of the traditional distance method, the accuracy of constructing the system's evolutionary subtrees is significantly improved.

[0035] It is understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0036] In summary, this invention decomposes the task of building large-scale phylogenetic trees into four stages: supervised learning dataset construction, sequence placement and connectivity partitioning, construction of partitioned phylogenetic subtrees, and recursive splicing. This approach effectively balances tree construction accuracy, computational efficiency, and scalability, demonstrating superior overall performance. Furthermore, it shows good applicability and application potential on real biological data, providing an effective solution for building large-scale phylogenetic trees.

[0037] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0038] Figure 1 This is a flowchart of the large-scale system evolution tree construction method in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the supervised learning dataset simulation strategy in this invention; Figure 3 This is a schematic diagram of the simulation system evolution tree in this invention; Figure 4 This is a schematic diagram of the connectivity partitioning process in this invention; Figure 5 This is a schematic diagram of the partition topology tree generated during the connectivity partitioning process in this invention; Figure 6 This is a schematic diagram illustrating the process of generating a representative information dictionary during the connectivity partitioning process in this invention; Figure 7 This is a schematic diagram of the subtree construction process of the system evolution tree in this invention; Figure 8 This is an architecture diagram of the first stage of the system evolution subtree model in this invention; Figure 9 This is an architecture diagram of the second stage of the system evolution subtree model in this invention; Figure 10 This is a schematic diagram of the recursive splicing process in this invention; Figure 11 This is a schematic diagram illustrating the performance of the system evolution subtree model in this invention compared with the distance method (NJ), the maximum likelihood method (IQ-TREE), and the deep learning method (Phyloformer) under 50 classification units and a comparison length of 500aa. From top to bottom, the diagram shows the topological performance and the branching performance. Figure 12 This is a schematic diagram of the generalization performance of the system evolution subtree model under different classification units in this invention. From top to bottom, it shows the schematic diagram of topological performance and the schematic diagram of branch performance. Figure 13This is a schematic diagram of the generalization performance of the system evolution subtree model in this invention under different alignment lengths, wherein, from top to bottom, it is a schematic diagram of topological performance and a schematic diagram of branch performance; Figure 14 This is a schematic diagram illustrating the topological performance of the system evolution subtree model in this invention with the distance method (NJ), the maximum likelihood method (IQ-TREE), and the deep learning method (Phyloformer) at different classification units with a comparison length of 500aa. The classification units are 20, 30, 40, 60, 70, and 80 from left to right and from top to bottom. Figure 15 This is a schematic diagram illustrating the performance of the system evolution subtree model in this invention with the distance method (NJ), the maximum likelihood method (IQ-TREE), and the deep learning method (Phyloformer) at different classification units with a comparison length of 500aa. The classification units are 20, 30, 40, 60, 70, and 80 from left to right and from top to bottom. Figure 16 This is a schematic diagram illustrating the topological performance of the system evolution subtree model in this invention with the distance method (NJ), the maximum likelihood method (IQ-TREE), and the deep learning method (Phyloformer) at different alignment lengths in 50 classification units. The lengths from left to right and from top to bottom are 250aa, 750aa, 1000aa, and 1500aa, respectively. Figure 17 This is a schematic diagram illustrating the performance of the system evolution subtree model in this invention with the distance method (NJ), the maximum likelihood method (IQ-TREE), and the deep learning method (Phyloformer) at different comparison lengths in 50 classification units. The lengths from left to right and from top to bottom are 250aa, 750aa, 1000aa, and 1500aa, respectively. Figure 18 This is a schematic diagram illustrating the performance of the large-scale phylogenetic tree construction method, the distance method (NJ), the distance method combined with divide-and-conquer (NJ), and the maximum likelihood method combined with divide-and-conquer (IQ-TREE) in constructing phylogenetic trees in this invention with 100 taxonomic units and a comparison length of 500aa. Figure 19 shows an example of the predicted and actual system evolution trees constructed by the large-scale system evolution tree construction method of the present invention with 100 classification units and a comparison length of 500aa. (a) is the predicted system evolution tree and (b) is the actual system evolution tree. Figure 20 This is a schematic diagram illustrating the performance of the large-scale phylogenetic tree construction method, the distance method (NJ), the distance method combined with divide-and-conquer (NJ), and the maximum likelihood method combined with divide-and-conquer (IQ-TREE) in constructing phylogenetic trees in this invention with 200 taxonomic units and a comparison length of 500aa. Figure 21 shows an example of the predicted and actual system evolution trees constructed by the large-scale system evolution tree construction method of the present invention with 200 classification units and a comparison length of 500aa. In this figure, (a) is the predicted system evolution tree and (b) is the actual system evolution tree. Figure 22 This is a schematic diagram illustrating the performance of the large-scale phylogenetic tree construction method, the distance method (NJ), the distance method combined with divide-and-conquer (NJ), and the maximum likelihood method combined with divide-and-conquer (IQ-TREE) in constructing phylogenetic trees with 500 classification units and a comparison length of 500aa in this invention. Figure 23 shows an example of the predicted system evolution tree and the actual system evolution tree in the large-scale system evolution tree construction method of the present invention with 500 classification units and 500aa alignment length, where (a) is the predicted system evolution tree and (b) is the actual system evolution tree. Figure 24 This is a schematic diagram illustrating the performance of the large-scale phylogenetic tree construction method, the distance method (NJ) combined with the divide-and-conquer approach, and the maximum likelihood method (IQ-TREE) combined with the divide-and-conquer approach in this invention, when constructing a phylogenetic tree with 1000 taxonomic units and a comparison length of 500aa. Figure 25 shows an example of the predicted and actual system evolution trees constructed by the large-scale system evolution tree construction method of the present invention with 1000 classification units and a comparison length of 500aa. (a) is the predicted system evolution tree and (b) is the actual system evolution tree. Figure 26 is a performance comparison diagram of the phylogenetic tree constructed by the phylogenetic subtree model, the distance method (NJ), and the maximum likelihood method (IQ-TREE) in the cyanobacteria dataset in this invention, where (a) is a schematic diagram of topological performance and (b) is a schematic diagram of branch performance. Figure 27 shows an example of the phylogenetic tree predicted by the phylogenetic subtree model in the cyanobacteria dataset and the actual phylogenetic tree, where (a) is the predicted phylogenetic tree and (b) is the actual phylogenetic tree. Figure 28 This is a comparison chart of the topological performance of the system evolution subtree model, the distance method (NJ), and the maximum likelihood method (IQ-TREE) constructed in the Hug Tree of Life subset dataset in this invention; Figure 29 This is a schematic diagram illustrating the performance of the large-scale phylogenetic tree construction method of the present invention, which combines the divide-and-conquer distance method (NJ) and the divide-and-conquer maximum likelihood method (IQ-TREE), on the Hug Tree of Life dataset and the mammal dataset. Figure 30This is a schematic diagram of the large-scale phylogenetic tree predicted by the large-scale phylogenetic tree construction method of the present invention in the Hug Tree of Life dataset; Figure 31 This is a schematic diagram of a large-scale phylogenetic tree predicted in a mammalian dataset using the large-scale phylogenetic tree construction method of this invention. Figure 32 A schematic diagram of a computer device provided in an embodiment of the present invention; Figure 33 This is a block diagram of a chip provided according to an embodiment of the present invention.

[0039] Among them, 60. Computer equipment; 61. Processor; 62. Memory; 63. Computer program; 600. Electronic device; 610. Processing unit; 620. Storage unit; 6201. Random access memory unit; 6202. Cache memory unit; 6203. Read-only memory unit; 6204. Program / utility; 6205. Program module; 630. Bus; 640. Display unit; 650. Input / output interface; 660. Network adapter; 700. External device. Detailed Implementation

[0040] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0042] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0043] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0044] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0045] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0046] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0047] This invention provides a method for constructing large-scale phylogenetic trees. First, referencing the statistical characteristics of empirical phylogenetic trees and real amino acid multiple sequence alignment (MMI) data from publicly available databases, a phylogenetic tree and corresponding MMI MMI data are simulated and generated, constructing a supervised learning dataset for training and testing the phylogenetic subtree inference model. Then, a trunk tree is constructed and sequences are placed on the MMI MMI data of large-scale classification units, and connected partitions are created based on the topology of the placement tree, partition size constraints, and structural span constraints. Further, each partition is independently inferred using a two-stage deep learning-based phylogenetic subtree construction model. First, the pairwise evolutionary distance matrix between classification units is predicted and optimized, then bundle search and adjacency join are combined to construct the partition's phylogenetic subtree. Finally, based on the partition topology tree and a representative information dictionary, a global phylogenetic tree is constructed through operations such as root re-determination, anchor point determination, repeated node pruning, and recursive splicing. This method targets large-scale classification units and decomposes the task of building large-scale phylogenetic trees into four stages: supervised learning dataset construction, sequence placement and connectivity partitioning, construction of partitioned phylogenetic subtrees, and recursive splicing. It provides a novel and effective solution for building large-scale phylogenetic trees, which can balance the accuracy, efficiency, and scalability of phylogenetic tree construction.

[0048] Please see Figure 1 This invention provides a method for constructing a large-scale system evolution tree, comprising the following steps: S1. Construct a supervised learning dataset for training and testing the phylogenetic subtree inference model. Use simulated amino acid multiple sequence alignment data as supervised learning input and the corresponding phylogenetic tree as supervised learning label to create a supervised learning dataset. Divide the supervised learning dataset into a training set, a validation set, and a test set. Please see Figure 2 The data simulation strategy of this invention. To make the simulation parameters closely resemble the properties of real-world biological sequences, the parameter distributions of empirical data in the Hug Tree of Life, HOGENOM database, and RAxMLGROVE database are referenced; rate change events are introduced, the occurrence of which follows a Poisson distribution; a system evolution tree with reasonable branch lengths is simulated; and the data is combined with... The model was replaced, and the corresponding amino acid multiple sequence alignment data was simulated using the AliSim tool to create a supervised learning dataset, specifically: S101. Referencing the statistical characteristics of tree diameter and site heterogeneity parameters from empirical data in publicly available databases, determine the parameter distribution used to simulate phylogenetic trees and amino acid multiple sequence alignment data; Please see Figure 3 This invention simulates the process of system evolution tree.

[0049] S102. Generate a basic system evolution tree using the birth-death process, and adjust the branch length of the basic system evolution tree by introducing evolution rate heterogeneity to obtain a non-hypermetric simulated system evolution tree. S103. Based on the diameter distribution of the empirical system evolution tree, scale the branch length of the simulated system evolution tree obtained in step S102, and adjust the excessively short external branches to obtain the final simulated system evolution tree. S104, Based on the replacement model The phylogenetic tree obtained in step S103 was used to simulate the corresponding amino acid multiple sequence alignment data using the bioinformatics software AliSim; S105. Using the amino acid multiple sequence alignments obtained in step S104 as data and the corresponding phylogenetic tree as labels, a supervised learning dataset is created. S106. Divide the dataset obtained in step S105 into a training set, a validation set, and a test set.

[0050] The beneficial effects of the above technical solution are: The simulation parameters are derived from empirical system evolution trees and real biological sequences. By simulating complex evolutionary history and population structure, the generalization ability of the model is enhanced. The simulation data maintains a high degree of consistency with real biological data in terms of tree diameter distribution, branch length distribution, amino acid composition, and cross-species heterogeneity, indicating that the supervised learning dataset can well reflect the statistical characteristics of real system evolution scenarios.

[0051] Please see Figure 4 This invention describes a connectivity partitioning process for large-scale amino acid multiple sequence alignment data.

[0052] S2. For amino acid multiple sequence alignment data of large-scale taxonomic units, a trunk tree is given or constructed based on the amino acid multiple sequence alignment data. Non-trunk sequences are placed into the trunk tree to obtain a placement tree. Based on the topology, partition size constraints, and structural span constraints of the placement tree, all sequences are partitioned into connected subtrees, so that each partition corresponds to a connected subtree on the placement tree. During the partitioning process, the topological relationships between partitions and the representative information of adjacent partitions are extracted, and finally the sequence partitioning results, partition topology tree, and representative information dictionary are obtained. S201. For amino acid multiple sequence alignment data of large-scale taxonomic units, provide a trunk tree, or construct a trunk tree based on some representative sequences in the amino acid multiple sequence alignment data; The construction of a trunk tree includes at least one of the following methods: 1) De novo construction mode: First, sample the amino acid multiple sequence alignment data to generate a reference tree. Perform tree clustering on the reference tree to divide all sequences into several clusters. Then, select the sequence with the highest data quality from each cluster as the backbone sequence. Based on the backbone sequence, use the maximum likelihood method or the system evolution subtree construction model of this invention to construct the backbone tree. 2) Trunk sequence mode: Constructs a trunk tree based on a pre-specified trunk sequence; 3) Trunk tree mode, directly using the existing trunk tree.

[0053] S202. Using the Apples-2 system evolution placement algorithm based on the weighted least squares criterion, the non-stem sequence is placed into the stem tree as a query sequence according to the distance fitting result between the non-stem sequence and the stem tree, thus obtaining the placement tree. S203. Based on the topology, partition size constraints, and structural span constraints of the placement tree, perform connected partitioning on all sequences, so that each partition corresponds to a connected subtree on the placement tree; The rules for determining connected partitions include: performing a post-order traversal of the placement tree, calculating the total number of sequences contained in the current local region from bottom to top as the partition size, and calculating the sum of the distances from the current node to the farthest leaf nodes in different sub-directions as the structural span; when the partition size of the current region does not exceed a preset threshold, or the structural span does not reach a preset threshold, the current region is retained in the same partition; otherwise, a new partition boundary is formed at the corresponding position, and the corresponding subtree is divided; under preset special circumstances, the partitioning conditions are relaxed; wherein, the set of edges with the same partition number and maintaining connectivity on the placement tree constitutes a partition, and the sequence corresponding to the edge set constitutes the sequence set of the partition.

[0054] S204. Extract the hierarchical topological relationships between partitions and representative information of adjacent partitions during the partitioning process; Please see Figure 5 The hierarchical topology extracted during the partitioning process in this invention is represented as a partition topology tree; please refer to [link to relevant documentation]. Figure 6 This invention generates a representative information dictionary from the representative information of adjacent partitions extracted during the partitioning process. When the parent node in the placement tree belongs to a different partition than its child node, the parent-child hierarchical relationship of the corresponding partition is recorded as a hierarchical topological relationship between partitions, and a partition topology tree with partition number as the node is constructed. At the same time, representative subtrees in the upward and downward directions are constructed at the partition boundaries. The representative subtrees are composed of representative classification units near the boundary and their connection relationships, which are used to represent the boundary connection information between the current partition and adjacent partitions. Typically, the topology of the leaf node with the closest distance in different directions is retained. Finally, the representative subtrees of the child partition corresponding to the parent partition and the upward representative subtrees of the child partition are stored in the representative information dictionary.

[0055] S205. Based on the processing results of steps S203 and S204, the sequence partitioning results, partitioning topology tree, and representative information dictionary are obtained.

[0056] The beneficial effects of the above technical solution are: By placing the trunk tree and using connected partitions, the task of constructing a large-scale system evolution tree is divided into several local subproblems of manageable size. Each partition corresponds to a connected subtree on the placement tree, thus ensuring the structural continuity and rationality of the partitioning results and reducing the computational complexity of system evolution tree construction. Furthermore, by simultaneously extracting the partition topology tree and representative information dictionary during the partitioning process, a splicing order and basis are provided for subsequent cross-partition splicing, thereby improving the rationality and stability of the global system evolution tree.

[0057] Please see Figure 7 The process of constructing the evolutionary subtree of the partitioning system in this invention.

[0058] S3. For each partition obtained in step S2, use the supervised learning dataset constructed in step S1 to train the system evolution subtree construction model—Representation-Beam-Searching Phylogenetic Inference (RBS-Phylo). The RBS-Phylo model consists of an evolutionary distance prediction model, a distance matrix optimization model, and beam search and adjacency methods. The trained RBS-Phylo model is used to independently construct the system evolution subtree of each partition. The dataset used in the training process is 50 classification units and 500aa alignment length amino acid multiple sequence alignment data. S301. For each partition obtained in step S2, extract the corresponding amino acid multiple sequence alignment data for that partition, and perform one-hot encoding on the amino acid multiple sequence alignment data. The model input is the shape. A tensor, where 22 represents 20 amino acids, as well as missing (-) and unknown characters; S302. In the first stage, an evolutionary distance prediction model based on Transformer is used to extract features and characterize distances in the encoded amino acid multiple sequence alignment data.

[0059] Please see Figure 8 The architecture diagram of the first stage of the RBS-Phylo model of this invention is as follows: S3021. After input, the data first passes through a site embedding layer. This mapping shares parameters across all sites, mapping the one-hot representation of each site to a 64-dimensional embedding space, resulting in... ; S3022. Construct initial sequence pair representations in the sequence distance averaging layer:

[0060] In the formula: Indicates the first The embedding representation of the sequence is used to obtain the shape as The sequence pair representation, 1225 is composed of Calculated; S3023. A six-layer axial self-attention module is used to update the sequence pair representation. Each layer contains two sub-self-attention blocks: first, attention calculation is performed on the site dimension (row) to integrate information between different sites within the same sequence pair; then, attention calculation is performed on the sequence pair dimension (column) to enable the representation of each sequence pair to incorporate information from other sequence pairs. Through this alternating information propagation mechanism, the sequence representation of each pair gradually integrates the global information of the entire MSA, achieving evolutionary distance modeling under global conditions. Furthermore, layer normalization is performed before each self-attention sub-module, and the sub-module is fused with the input via residual connections. S3024. After layer normalization, the data enters the feedforward neural network, which contains two fully connected layers. The hidden layer has a dimension of 256 and uses the GELU activation function to increase the non-linearity of the representation. S3025. Compress the representation of each sequence pair at each site into a scalar using a site-feedforward neural network. and use Activation function to ensure non-negative output:

[0061] The size is obtained Matrix; S3026. Average the values ​​at each point in the site averaging layer to obtain a length of... The pairwise evolution prediction vector.

[0062] S3027. In the evolution matrix construction module, adjust the pairwise evolution prediction vectors to a 50×50 symmetric pairwise evolution distance matrix, and set the diagonal elements to 0 to obtain the evolution distance matrix between classification units.

[0063] The evolution distance prediction model is trained using the mean squared error loss function; S303. Using a distance matrix optimization model based on residual convolutional neural networks, the pairwise evolutionary distance matrix obtained in step S302 is corrected to obtain a more accurate distance representation.

[0064] Please see Figure 9 The architecture diagram of the second stage of the RBS-Phylo model of this invention is as follows: S3031. Let the evolution distance matrix predicted in the first stage be... ,here Using this as a single-channel input model, feature mapping is performed through an input convolutional layer with a kernel size of 3×3, a stride of 1, and padding of 1. This convolutional layer contains 32 kernels, thus mapping the input single-channel distance matrix into 32 feature channels. Let this convolutional layer be denoted as... The output can then be represented as In the formula: Represents the ReLU activation function. The role of this layer is to initially extract local system evolution signals from the original distance matrix, laying the foundation for deep representation learning of subsequent residual blocks.

[0065] S3032. Deep feature extraction is performed on the evolutionary distance matrix using four cascaded residual blocks. Each residual block has a consistent internal structure, consisting of two convolutional layers and an identity skip connection. For the input features... The calculation process for a single residual block is as follows:

[0066]

[0067]

[0068] In this system, both convolutional layers maintain the same number of input and output channels, with a kernel size of 3×3, a stride of 1, and padding of 1. The output can then be represented as... ; S3033, High-dimensional features By mapping the output convolutional layer back to a single channel, the distance matrix residual correction term is obtained. The output convolutional layer also uses a 3×3 convolutional kernel with a stride of 1 and padding of 1 to ensure that the output matrix is ​​consistent with the input matrix in terms of spatial dimensions. S3034. Correct the original distance matrix using residual update: In the formula: This is the residual scaling factor, used to control the strength of the correction term; S3035, to Post-processing is performed to ensure that it satisfies symmetry, non-negativity, and that the diagonal elements are 0.

[0069] The distance matrix model is trained using a joint loss function:

[0070] in, It is the distance regression loss used to constrain the error between the optimized evolutionary distance matrix and the true evolutionary distance matrix. It is used to constrain the local Q-ordering loss of candidate merge clusters to the ordering relation. It is a residual regularization term that limits the magnitude of the distance matrix correction. It is a regularization term that restricts the relative change ratio of the distance matrix before and after optimization. , , and These are the weighting coefficients. The optimized evolutionary distance matrix is ​​obtained by optimizing and updating the neural network parameters using gradient descent and backpropagation algorithms.

[0071] The specific definitions of each part of the joint loss function are as follows: 1) The distance regression loss aims to make the optimized evolutionary distance matrix numerically close to the true evolutionary distance matrix, using masking smoothing. To mitigate the loss, the diagonal elements are masked to prevent trivial terms that are always zero from interfering with training. Defined as:

[0072]

[0073] in, It is the evolution distance matrix of the mask; It is the optimized evolution distance matrix; It is the actual evolutionary distance matrix; 2) The local Q-ranking loss is used to constrain the optimized evolutionary distance matrix, ensuring consistency with the true evolutionary distance matrix in the ranking of candidate merged cluster pairs during subsequent adjacency-based tree construction. First, based on the true distance matrix... The adjacency-based tree construction process is performed to obtain the teacher merging trajectory. Then, for each teacher state, a merging Q-matrix is ​​constructed based on the optimized evolutionary distance matrix and the true evolutionary distance matrix, respectively. The smallest merging Q-matrix in the true merging Q-matrix is ​​selected. Select one cluster pair as a positive sample, and then select from the remaining cluster pairs. Difficult negative samples are sampled first from cluster pairs with high Q-value ranking. The predicted evolutionary distance matrix is ​​required to maintain the order of positive samples over negative samples in terms of Q-value, and the predicted evolutionary distance matrix is ​​encouraged to maintain the correct merging priority in terms of ranking. Defined as:

[0074] in, This represents the set of positive samples selected by the true merged Q matrix in the current state; Represents the set of negative samples; The sorting interval; It is the prediction distance matrix; 3) The residual regularization term is used to limit the magnitude of the distance matrix correction. Defined as:

[0075] in, It is the total number of elements in the evolution distance matrix; 4) The relative change regularization term is used to limit the relative change ratio of the distance matrix before and after optimization. Defined as:

[0076] in, It is a minimal constant that prevents the denominator from being zero.

[0077] S304. Based on the optimized evolution distance matrix obtained in step S303, the system evolution tree inference of the current partition is performed using the cluster search and adjacency method, and the corresponding partition system evolution subtree is constructed. S3041. Based on the optimized evolutionary distance matrix obtained in step S303, construct the merging Q matrix of candidate cluster pairs according to the merging criterion of the adjacency method, and sort the candidate cluster pairs according to the Q value from smallest to largest. The Q value is calculated as follows:

[0078] in, , and It is a sequence number; It is the sequence number; It is the evolutionary distance matrix; S3042. Select the top K candidate cluster pairs from the top-ranked candidate cluster pairs, and select E candidate cluster pairs from them as the extension objects of the current state according to the deterministic mode or the random mode. Perform a merging operation on all extension objects to generate multiple candidate intermediate states. The deterministic model directly selects the E cluster pairs with the smallest Q values ​​for expansion.

[0079] The randomized pattern performs no-replacement sampling on the top K cluster pairs based on their sorting position. If the rank of a cluster pair in the top K set is... The probability of it being sampled is defined as follows: In the formula: The parameter controls the flatness of the probability distribution. When the value is small, sampling tends to favor cluster pairs with higher rankings and smaller Q values; when... When the value is large, the probability difference between different cluster pairs decreases, thereby increasing the diversity of the search; S3043. For the multiple candidate intermediate states generated in step S3042, sort them according to their path scores in ascending order, and retain the top B candidate intermediate states for the next round of search. The path score calculation method includes at least one of the following: one is to accumulate the Q values ​​of the merged cluster pairs along the way. One approach is to set all path scores to zero, meaning that the quality of intermediate paths is not explicitly distinguished, and paths are preserved only based on the expansion order and bundle width. S3044. Repeat steps S3041 to S3043 until all classification units in the current partition are merged, and multiple candidate system evolution subtrees are obtained after deduplication.

[0080] S305. Perform steps S301 to S304 on each partition, and select the optimal system evolution tree as the system evolution subtree for each partition.

[0081] The beneficial effects of the above technical solution are: RBS-Phylo separates sequence feature extraction from multi-sequence alignment from system evolution tree construction, thus combining the representational power of deep learning methods with the rapid tree-building capability of distance-based methods. This significantly improves the efficiency of constructing evolutionary subtrees for partitioned systems while maintaining high tree-building accuracy. Furthermore, by optimizing the predicted evolutionary distance matrix and introducing a bundle search mechanism to retain multiple candidate intermediate states, the error accumulation of the greedy algorithm is mitigated, thereby improving the stability and reliability of partitioned system evolutionary subtree construction and providing higher-quality local subtree input for subsequent global system evolution tree splicing.

[0082] Please see Figure 10 A schematic diagram of the recursive splicing process of this invention.

[0083] S4. Based on the partition topology tree and representative information dictionary obtained in step S2, the evolutionary subtrees of each partition system obtained in step S3 are spliced ​​together. By re-rooting, determining splicing anchor points, pruning duplicate nodes, and recursively splicing, a global system evolution tree containing all classification units and without duplicate classification units is constructed. S401. Determine the splicing hierarchy between the evolutionary subtrees of each partition system based on the partition topology tree obtained in step S2, and determine the bottom-up splicing order of the evolutionary subtrees of each partition system obtained in step S3. S402. During the splicing process, the system evolution subtree corresponding to the current partition is taken as the parent tree, the local system evolution tree that has completed the splicing of the lower part is taken as the subtree, and the parent tree is rerooted to determine the direction. S403. Based on the representative information dictionary obtained in step S2, determine the parent tree splicing anchor point in the parent tree. Specifically: the algorithm reads the set of representative nodes of the sub-partition in the current parent partition. Since these representative nodes appear in both the parent and sub-partitions, they can be regarded as a topological reference connecting the parent tree and the subtree. Find the lowest common ancestor of these representative nodes in the parent tree and use it as the splicing anchor point on the parent tree side; S404. To avoid node duplication, prune the sub-partition representative nodes in the parent tree used to locate the splicing position. S405. Based on the representative information dictionary obtained in step S2, determine the subtree splicing anchor point in the subtree. Specifically, the algorithm uses the same group of representative nodes to locate the splicing anchor point in the subtree. Specifically, it finds the lowest common ancestor of this group of representative nodes in the subtree and uses it as the splicing anchor point on the subtree side. This anchor point reflects the most direct correspondence between the subtree and the parent tree, providing a positional basis for subsequent splicing. S406. To avoid node duplication, prune nodes in the subtree that are duplicates of the parent tree. S407. Perform splicing based on the splicing anchor points of the parent tree and child tree, and perform single-branch structure compression and other processing. S406. Recursively execute steps S402 to S407 until all partition system evolution subtrees are spliced ​​together. Perform manual root node cleaning and root removal on the entire tree to obtain the global system evolution tree.

[0084] The beneficial effects of the above technical solution are: By using a partitioned topology tree and a representative information dictionary to perform structure-guided recursive splicing of cross-partition system evolution subtrees, a global system evolution tree containing all classification units without duplicate classification units can be obtained, thereby improving the accuracy and stability of large-scale system evolution tree construction.

[0085] S5. Based on simulated sequence data and biological sequence data from public databases, the large-scale phylogenetic tree construction algorithm of this invention is tested to evaluate the accuracy of the topological structure, branch length estimation accuracy, and running time efficiency of the constructed phylogenetic tree. The results of comparative and generalization tests show that the algorithm of this invention has good overall performance on both simulated and real biological data.

[0086] In another embodiment of the present invention, a large-scale system evolution tree construction system is provided. This system can be used to implement the above-mentioned large-scale system evolution tree construction method. Specifically, the large-scale system evolution tree construction system includes a supervised learning dataset construction module, a partitioning module, a system evolution subtree construction module, and a splicing module.

[0087] The supervised learning dataset construction module is used to simulate and generate system evolution trees and corresponding amino acid multiple sequence alignment data based on the statistical characteristics of empirical system evolution trees and real amino acid multiple sequence alignment data in public databases, and to construct supervised learning datasets. The partitioning module, connected to the supervised learning dataset construction module, is used to construct the backbone tree, place the system evolution, and partition the amino acid multiple sequence alignment data of large-scale classification units, and extract the partition topology tree and representative information dictionary. The system evolution subtree construction module is connected to the partitioning module and is used to perform deep learning-based system evolution subtree construction on each partition to obtain the system evolution subtree of each partition; The splicing module, connected to the partitioning module and the system evolution subtree construction module, is used to recursively splice the system evolution subtrees of each partition according to the partition topology tree and the representative information dictionary to obtain a global system evolution tree that contains all classification units and does not contain duplicate classification units.

[0088] This invention provides a terminal device comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or function. The processor described in this embodiment can be used for the operation of a large-scale system evolution tree construction method, including: Based on the statistical characteristics of tree diameters of empirical system evolution trees in publicly available databases and the statistical characteristics of site heterogeneity parameters in real amino acid multiple sequence alignment (MMI) data, a system evolution tree and corresponding MMI MMI data are simulated and generated. Using the MMI MMI data as input for supervised learning and the corresponding system evolution tree as the supervised learning label, a supervised learning dataset is constructed for training and testing the system evolution subtree inference model. For MMI MMI data of large-scale classification units, a trunk tree is given or constructed based on the MMI MMI data. Non-trunk sequences are placed into the trunk tree to obtain a placement tree. Based on the topology, partition size constraints, and structural span constraints of the placement tree, all sequences are connected and partitioned, so that each partition corresponds to a connected subtree on the placement tree. Simultaneously, the hierarchical topological relationships between partitions and representative information of adjacent partitions are extracted to obtain the sequence partitioning results, partition topology trees, and... The system evolution subtree construction model is trained and tested based on the supervised learning dataset. The trained model is then used to independently construct system evolution subtrees for each partition. The system evolution subtree construction model includes an evolutionary distance prediction model, a distance matrix optimization model, and a bundle search and adjacency method. The model first uses the evolutionary distance prediction model to predict the pairwise evolutionary distance matrix between classification units within a partition, then uses the distance matrix optimization model to optimize the pairwise evolutionary distance matrix. Based on the optimized evolutionary distance matrix, bundle search and adjacency methods are used to construct the corresponding partition's system evolution subtree. According to the partition topology tree and the representative information dictionary, the system evolution subtrees of each partition are spliced ​​together. Through root re-determination, determining splicing anchor points, pruning duplicate nodes, and recursive splicing, a global system evolution tree containing all classification units and without duplicate classification units is constructed.

[0089] Please see Figure 32 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the large-scale system evolution tree construction method described in this embodiment. To avoid repetition, details are omitted here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the large-scale system evolution tree construction system described in this embodiment. To avoid repetition, details are omitted here.

[0090] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 32This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0091] The processor 61 may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0092] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0093] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0094] Please see Figure 33 The terminal device is an electronic device 600, which is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), a display unit 640, etc.

[0095] The storage unit stores program code, which can be executed by the processing unit 610 to perform the steps described in the method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 610 can perform actions such as... Figure 1 The steps are shown in the figure.

[0096] Storage unit 620 may include readable media in the form of volatile storage units, such as random access memory (RAM) 6201 and / or cache memory 6202, and may further include read-only memory (ROM) 6203.

[0097] Storage unit 620 may also include a program / utility 6204 having a set (at least one) program module 6205, such program module 6205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0098] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0099] Electronic device 600 can also communicate with one or more external devices 700 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem). This communication can be performed via input / output interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network, wide area network, and / or public network, such as the Internet) via network adapter 660. Network adapter 660 can communicate with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0100] This invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). More specific examples of the computer-readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical fiber, portable compact disk read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0101] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, radio frequency, etc., or any suitable combination thereof.

[0102] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0103] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the large-scale system evolution tree construction method in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps: Based on the statistical characteristics of tree diameters of empirical system evolution trees in publicly available databases and the statistical characteristics of site heterogeneity parameters in real amino acid multiple sequence alignment (MMI) data, a system evolution tree and corresponding MMI MMI data are simulated and generated. Using the MMI MMI data as input for supervised learning and the corresponding system evolution tree as the supervised learning label, a supervised learning dataset is constructed for training and testing the system evolution subtree inference model. For MMI MMI data of large-scale classification units, a trunk tree is given or constructed based on the MMI MMI data. Non-trunk sequences are placed into the trunk tree to obtain a placement tree. Based on the topology, partition size constraints, and structural span constraints of the placement tree, all sequences are connected and partitioned, so that each partition corresponds to a connected subtree on the placement tree. Simultaneously, the hierarchical topological relationships between partitions and representative information of adjacent partitions are extracted to obtain the sequence partitioning results, partition topology trees, and... The system evolution subtree construction model is trained and tested based on the supervised learning dataset. The trained model is then used to independently construct system evolution subtrees for each partition. The system evolution subtree construction model includes an evolutionary distance prediction model, a distance matrix optimization model, and a bundle search and adjacency method. The model first uses the evolutionary distance prediction model to predict the pairwise evolutionary distance matrix between classification units within a partition, then uses the distance matrix optimization model to optimize the pairwise evolutionary distance matrix. Based on the optimized evolutionary distance matrix, bundle search and adjacency methods are used to construct the corresponding partition's system evolution subtree. According to the partition topology tree and the representative information dictionary, the system evolution subtrees of each partition are spliced ​​together. Through root re-determination, determining splicing anchor points, pruning duplicate nodes, and recursive splicing, a global system evolution tree containing all classification units and without duplicate classification units is constructed.

[0104] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0105] 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.

[0106] Please see Figures 11 to 2 5. The results of comparative and generalization experiments in simulated datasets.

[0107] Among them, the nRF distance is used as the criterion for judging the accuracy of the system evolution tree topology construction; the nKF distance is used as the criterion for judging the accuracy of the system evolution tree branch length estimation; the smaller the nRF distance and nKF distance, the more accurate the system evolution tree construction.

[0108] In the comparative experiments, the RBS-Phylo model and large-scale phylogenetic tree construction method of the present invention are compared with at least one of the distance method (NJ), maximum likelihood method (IQ-TREE), and deep learning method (Phyloformer) to evaluate the phylogenetic tree construction effect of different methods. In the generalization experiments, the RBS-Phylo model of the present invention is tested under different classification unit sizes (20 to 80 classification units) and different alignment sequence lengths (250aa to 1500aa) to evaluate the generalization ability of the method.

[0109] Please see Figure 11 The RBS-Phylo of this invention outperforms NJ, IQ-TREE and Phyloformer in phylogenetic tree construction performance with 50 taxonomic units and a comparison length of 500aa.

[0110] Please see Figure 12 and Figure 13 The RBS-Phylo of this invention exhibits good phylogenetic tree construction performance and good generalization performance under different taxonomic unit sizes and different alignment sequence lengths.

[0111] Please see Figures 14 to 17 The RBS-Phylo method of this invention still shows better accuracy in constructing phylogenetic trees compared to the other three methods, under different taxonomic unit sizes and different alignment sequence lengths.

[0112] Please see Figure 18 In a large-scale system evolution tree construction task with 100 taxonomic units, the large-scale system evolution tree construction method of this invention was compared with comparative methods. The comparison results show that the method of this invention performs best in tree topology accuracy; the method of this invention performs well overall in branch length accuracy, although it is slightly inferior to the maximum likelihood method under the divide-and-conquer framework, the difference is small, and it has a significant advantage in running speed compared to the method.

[0113] Please refer to Figure 19, which shows an example of a system evolution tree of 100 classification units constructed by the method of the present invention.

[0114] Please see Figure 20 In a large-scale phylogenetic tree construction task with 200 taxonomic units, the large-scale phylogenetic tree construction method of this invention was compared with comparative methods. The comparison results show that the method of this invention performs best in tree topology accuracy and branch length accuracy while maintaining a fast running speed.

[0115] Please refer to Figure 21, which shows an example of a system evolution tree of 200 classification units constructed by the method of the present invention.

[0116] Please see Figure 22 In a large-scale phylogenetic tree construction task with 500 taxonomic units, the large-scale phylogenetic tree construction method of this invention was compared with comparative methods. The results show that the method of this invention performs best in topological accuracy while maintaining accurate branch length estimation and fast running speed.

[0117] Please refer to Figure 23, which shows an example of a system evolution tree of 500 classification units constructed by the method of the present invention.

[0118] Please see Figure 24 In a large-scale phylogenetic tree construction task with 1000 taxonomic units, the large-scale phylogenetic tree construction method of the present invention was compared with comparative methods. The results show that the method of the present invention performs best in tree topology accuracy and branch length accuracy while maintaining a fast running speed.

[0119] Please refer to Figure 25, which shows an example of an evolutionary tree of a system of 1000 taxonomic units constructed by the method of the present invention.

[0120] The experimental results on the simulated data above show that the method of the present invention can effectively balance topological accuracy, branch length estimation accuracy and computational efficiency in large-scale system evolution tree construction tasks.

[0121] Please refer to Figure 26 to Figure 31 The present invention compares the results of experiments and applications in real biological datasets.

[0122] Please refer to Figure 26, which compares the nRF and nKF distances of the cyanobacteria method, the distance method (NJ), and the maximum likelihood method (IQ-TREE) in the cyanobacteria dataset (https: / / datadryad.org / dataset / doi:10.5061 / dryad.pv6df). The results show that the method of the present invention is significantly better than NJ in both tree topology accuracy and branch length accuracy; compared with IQ-TREE, the method of the present invention is slightly better in topology accuracy, while still maintaining good performance in branch length estimation.

[0123] Please refer to Figure 27, an example of a cyanobacterial phylogenetic tree constructed by the method of the present invention.

[0124] Please see Figure 28 Hug's Tree of Life (https: / / www.nature.com / articles / nmicrobiol20164) In the MOESM205 dataset (8#), the system evolution tree construction results of the proposed method and the comparative methods are compared. The results show that the proposed method is significantly better than NJ and IQ-TREE in terms of tree topology accuracy.

[0125] Please see Figure 29 In the Tree of Life dataset and the Mammal dataset (https: / / orthomam.mbb.cnrs.fr / ), the nRF and nKF distances of the phylogenetic tree construction results of the proposed method and the comparative methods were compared. The results show that the proposed method performs best in tree topology accuracy and branch length accuracy while maintaining a fast running speed.

[0126] Please see Figure 30 An example of the Hug Tree of Life system evolution tree constructed by the method of the present invention.

[0127] Please see Figure 31 An example of a mammalian phylogenetic tree constructed by the method of the present invention.

[0128] The experimental results on real biological data show that the large-scale phylogenetic tree construction method of the present invention not only has good tree construction performance on simulated data, but also can still take into account topological accuracy, branch length estimation accuracy and computational efficiency in real application scenarios, and has good applicability and application potential.

[0129] The advantages of the system evolution tree construction results of the present invention can be seen from the following aspects.

[0130] First, by employing a divide-and-conquer strategy, this invention decomposes the large-scale system evolution tree construction task into four stages: supervised learning dataset construction, sequence placement and connectivity partitioning, partition system evolution subtree construction, and recursive splicing. This transforms the task into a local subproblem of controllable scale, reducing the computational complexity of global tree construction. As a result, this invention can quickly handle large-scale system evolution tree construction tasks. Second, by using deep learning methods, the sequence feature extraction process is separated from the system evolution tree construction process. This combines the representational ability of deep learning methods with the fast tree construction ability of distance methods, and introduces a cluster search mechanism to alleviate the error accumulation of the greedy algorithm, thereby improving the accuracy and speed of system evolution subtree construction. Third, compared with the cyanobacteria dataset, the Hug Tree of Life dataset has more missing sites in the amino acid multiple sequence alignment data. The method of this invention can adapt to this situation better, showing stronger robustness and stability, and can still maintain high construction accuracy even when there are many missing alignment data.

[0131] In summary, this invention provides a method and system for constructing large-scale phylogenetic trees. Combining a divide-and-conquer strategy with deep learning technology, it decomposes the task of constructing large-scale phylogenetic trees into four stages: supervised learning dataset construction, sequence placement and connectivity partitioning, construction of partitioned phylogenetic subtrees, and recursive splicing. This approach effectively balances tree construction accuracy, computational efficiency, and scalability, demonstrating superior overall performance. Furthermore, it shows good applicability and application potential on real biological data, providing an effective solution for constructing large-scale phylogenetic trees.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0133] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0134] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0135] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0136] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0137] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0138] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random-access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0139] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0142] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for constructing a large-scale system evolution tree, characterized in that, Includes the following steps: Based on the statistical characteristics of tree diameter of empirical system evolution trees in public databases and the statistical characteristics of site heterogeneity parameters of real amino acid multiple sequence alignment data, a system evolution tree and corresponding amino acid multiple sequence alignment data are simulated and generated. The amino acid multiple sequence alignment data is used as the input for supervised learning, and the corresponding system evolution tree is used as the label for supervised learning. A supervised learning dataset for training and testing the system evolution subtree inference model is constructed. For amino acid multiple sequence alignment data of large-scale taxonomic units, a trunk tree is given or constructed based on the amino acid multiple sequence alignment data. Non-trunk sequences are placed into the trunk tree to obtain a placement tree. Based on the topology, partition size constraints and structural span constraints of the placement tree, all sequences are connected to partition, so that each partition corresponds to a connected subtree on the placement tree. At the same time, the hierarchical topological relationship between partitions and the representative information of adjacent partitions are extracted to obtain the sequence partitioning result, partition topology tree and representative information dictionary. The system evolution subtree construction model is trained based on the supervised learning dataset, and the trained system evolution subtree construction model is used to independently construct system evolution subtrees for each partition. The system evolution subtree construction model includes an evolution distance prediction model, a distance matrix optimization model, and a bundle search and adjacency method. The system evolution subtree construction model first uses the evolution distance prediction model to predict the pairwise evolution distance matrix between classification units within the partition, then uses the distance matrix optimization model to optimize the pairwise evolution distance matrix, and uses bundle search and adjacency method to construct the system evolution subtree of the corresponding partition based on the optimized evolution distance matrix. Based on the partition topology tree and the representative information dictionary, the evolutionary subtrees of each partition system are spliced ​​together. By re-rooting, determining the splicing anchor point, pruning duplicate nodes, and recursively splicing, a global system evolution tree containing all classification units and without duplicate classification units is constructed.

2. The method for constructing a large-scale system evolution tree according to claim 1, characterized in that, The process of constructing the supervised learning dataset includes: By referencing empirical system evolution trees and real amino acid multiple sequence alignment data in publicly available databases, the parameter distribution used to simulate system evolution trees and amino acid multiple sequence alignment data is determined. System evolution trees and corresponding amino acid multiple sequence alignment data are generated through simulation, and the resulting dataset is divided into training set, validation set, and test set.

3. The method for constructing a large-scale system evolution tree according to claim 2, characterized in that, The process of simulating the generation of a phylogenetic tree and the corresponding amino acid multiple sequence alignment data includes: A basic system evolution tree is generated using the birth-death process, and the branch length of the basic system evolution tree is adjusted by introducing evolution rate heterogeneity to obtain a non-overestimation simulated system evolution tree. The branch length of the obtained simulated system evolution tree is scaled by referring to the diameter distribution of empirical system evolution trees, and excessively short external branches are adjusted. Based on the replacement model and the adjusted system evolution tree, the corresponding amino acid multiple sequence alignment data is simulated using bioinformatics software.

4. The method for constructing a large-scale system evolution tree according to claim 1, characterized in that, The process of constructing a backbone tree and phylogenetic placement for amino acid multiple sequence alignment data of large-scale taxonomic units includes: Given a trunk tree, or constructing a trunk tree based on some representative sequences in the amino acid multiple sequence alignment data; then using a system evolution placement algorithm based on the weighted least squares criterion, non-trunk sequences are placed into the trunk tree according to the distance fitting results between the non-trunk sequences and the trunk tree, thus obtaining the placement tree.

5. The method for constructing a large-scale system evolution tree according to claim 1, characterized in that, The process of connecting partitions includes: A post-order traversal of the placement tree is performed. The total number of sequences contained in the current local region is calculated from bottom to top as the partition size, and the sum of the distances from the current node to the farthest leaf node in different sub-directions is calculated as the structural span. When the partition size of the current region does not exceed a preset threshold, or the structural span does not reach a preset threshold, the current region is kept in the same partition. Otherwise, a new partition boundary is formed at the corresponding position, and the corresponding subtree is divided. Among them, the set of edges with the same partition number and connected on the placement tree constitutes a partition, and the sequence corresponding to the edge set constitutes the sequence set of the partition.

6. The method for constructing a large-scale system evolution tree according to claim 1, characterized in that, The process of extracting the hierarchical topological relationships between partitions and representative information of adjacent partitions includes: When the parent node in the placement tree belongs to a different partition than its child node, the parent-child hierarchical relationship between the corresponding partitions is recorded as a hierarchical topological relationship between the partitions, and a partition topology tree with the partition number as the node is constructed. At the same time, representative subtrees in the upward and downward directions are constructed at the partition boundaries. The representative subtrees are composed of representative classification units near the boundary and their connection relationships, and the corresponding representative subtrees are stored in the representative information dictionary.

7. The method for constructing a large-scale system evolution tree according to claim 1, characterized in that, The process of constructing a system evolution subtree independently for each partition in the system evolution subtree construction model includes: Amino acid multiple sequence alignment data for the corresponding partitions are extracted and one-hot encoded. An evolutionary distance prediction model based on Transformer is used to extract and characterize features from the encoded amino acid multiple sequence alignment data, obtaining a pairwise evolutionary distance matrix between taxonomic units within the partition. A distance matrix optimization model based on residual convolutional neural networks is used to correct the pairwise evolutionary distance matrix, obtaining an optimized evolutionary distance matrix. Cluster search and adjacency methods are used to independently construct the phylogenetic subtrees of the corresponding partitions.

8. The method for constructing a large-scale system evolution tree according to claim 7, characterized in that, The distance matrix optimization model uses a joint loss function to optimize the neural network parameters. The joint loss function includes a distance regression loss to constrain the error between the optimized evolutionary distance matrix and the true evolutionary distance matrix, a local Q-ranking loss to constrain the ranking relationship of candidate merging clusters, a residual regularization term to limit the magnitude of distance matrix correction, and a relative change regularization term to limit the relative change ratio of the distance matrix before and after optimization.

9. The method for constructing a large-scale system evolution tree according to claim 1, characterized in that, The process of constructing the system evolution subtree of the corresponding partition based on the optimized evolutionary distance matrix using bundle search and adjacency method includes: A merging Q matrix for candidate cluster pairs is constructed according to the merging criteria of the adjacency method, and the candidate cluster pairs are sorted in ascending order of Q value. The top K candidate cluster pairs are selected from the top-ranked candidate cluster pairs, and E candidate cluster pairs are selected as the expansion objects of the current state according to the preset cluster pair selection strategy. Merging operations are performed on all expansion objects to generate multiple candidate intermediate states. The candidate intermediate states are sorted in ascending order of path score, and the top B candidate intermediate states are retained for the next round of search. The above process is repeated until the classification units in the current partition are merged, and the system evolution subtree of the current partition is obtained.

10. A large-scale system evolution tree construction system, characterized in that, The system for performing the method according to any one of claims 1 to 9, the system comprising: The supervised learning dataset construction module is used to simulate and generate system evolution trees and corresponding amino acid multiple sequence alignment data based on the statistical characteristics of empirical system evolution trees and real amino acid multiple sequence alignment data in public databases, and to construct supervised learning datasets. The partitioning module, connected to the supervised learning dataset construction module, is used to construct the backbone tree, place the system evolution, and partition the amino acid multiple sequence alignment data of large-scale classification units, and extract the partition topology tree and representative information dictionary. The system evolution subtree construction module is connected to the partitioning module and is used to perform deep learning-based system evolution subtree construction on each partition to obtain the system evolution subtree of each partition; The splicing module, connected to the partitioning module and the system evolution subtree construction module, is used to recursively splice the system evolution subtrees of each partition according to the partition topology tree and the representative information dictionary to obtain a global system evolution tree that contains all classification units and does not contain duplicate classification units.

Images