AML chemotherapy regimen bone marrow morphology prediction method and device based on bone marrow next-generation sequencing

Abstract

The invention provides an AML chemotherapy regimen bone marrow morphology prediction method and device based on bone marrow next-generation sequencing. The method comprises the following steps: acquiring bone marrow next-generation sequencing data of an AML patient before treatment, extracting an associated mutant gene list from the bone marrow next-generation sequencing data, then mapping the associated mutant gene list to an AML field knowledge graph, acquiring each target node successfully mapped by the associated mutant gene list in the AML field knowledge graph, and determining the target node according to the acquired target node. Feature settlement corresponding to each target node is extracted from the AML domain knowledge graph, the feature settlement is a connected sub-graph which takes the target node as the center and is determined according to a predefined topology rule, the feature settlement corresponding to each target node is input into a bone marrow morphology prediction model, and a bone marrow morphology prediction model is obtained. The method solves the problem that in the prior art, the bone marrow morphology cannot be predicted before chemotherapy, so that an original refractory patient cannot be recognized in a prospective manner.

Description

Technical Field

[0001] This invention relates to the field of primary refractory risk prediction technology, and in particular to a method and device for predicting bone marrow morphology in AML chemotherapy regimens based on next-generation bone marrow sequencing. Background Technology

[0002] Acute myeloid leukemia (AML) is a highly heterogeneous malignant hematologic malignancy, and its treatment and prognostic assessment remain significant clinical challenges. Currently, standard chemotherapy, represented by the "7+3" regimen, remains the preferred choice for inducing remission in AML patients suitable for intensive chemotherapy. However, in practice, some AML patients may develop primary refractory disease with this standard chemotherapy regimen. This means that after standard chemotherapy, bone marrow morphological assessment reveals that the leukemia cells have not been effectively eliminated, and the bone marrow cannot restore normal hematopoietic function, directly leading to treatment failure.

[0003] Therefore, bone marrow morphology results after chemotherapy are considered the "gold standard" for determining primary refractory AML. Thus, if the expected bone marrow morphology after standard chemotherapy can be assessed before a patient undergoes AML chemotherapy, the risk of primary refractory AML can be prospectively evaluated, providing crucial evidence for developing more effective individualized treatment plans. However, current clinical practice lacks effective technical means to achieve this goal before treatment. Summary of the Invention

[0004] This invention provides a method and device for predicting bone marrow morphology in AML chemotherapy regimens based on next-generation bone marrow sequencing, in order to solve the problems in the prior art.

[0005] On one hand, this invention provides a method for predicting bone marrow morphology in AML chemotherapy regimens based on next-generation bone marrow sequencing, comprising: Obtain bone marrow next-generation sequencing data of AML patients before treatment, and extract a list of associated mutated genes from the bone marrow next-generation sequencing data; The list of associated mutant genes is mapped to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the list of associated mutant genes. The nodes of the AML domain knowledge graph are used to represent biological entities, and the edges are used to represent the biological relationships between the corresponding biological entities. Extract the feature clusters corresponding to each target node from the AML domain knowledge graph, wherein the feature clusters are connected subgraphs determined by predefined topology rules with the target node as the center; The feature clusters corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy.

[0006] Preferably, the associated mutant gene list is mapped to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the associated mutant gene list, specifically including: Obtain the standard identifier of each associated variant gene in the list of associated mutant genes; For each associated variant gene, the standard identifier of the associated variant gene is matched with the standard identifier of the biological entity represented by each node in the AML domain knowledge graph to locate the target node corresponding to the associated variant gene; wherein, the standard identifier is specifically HGNC ID or NCBI Gene ID.

[0007] Preferably, extracting the feature clusters corresponding to each target node from the AML domain knowledge graph specifically includes extracting the feature clusters corresponding to each target node in the following manner: Determine the location of the target node in the AML domain knowledge graph, and obtain the connected subgraph of each node and edge within the K-hop neighbor rule with the location as the center, as the feature cluster corresponding to the target node, where K is an integer from 1 to 5.

[0008] Preferably, the method further includes: Determine the network centrality of the target node in the AML domain knowledge graph; Based on the network centrality, the value of K corresponding to the target node is determined, wherein the lower the network centrality of the target node, the larger the value of K, or the higher the network centrality of the target node, the smaller the value of K.

[0009] Preferably, the method further includes: Knowledge triples are extracted from multiple biomedical databases, wherein each knowledge triple includes a head entity, a relation, and a tail entity; wherein the biomedical databases specifically include any one or more of the following: NCBI Gene database, KEGG pathway database, Reactome pathway database, GO annotation database, DrugBank database, STRING protein interaction database, and ClinVar clinical variant database. The head and tail entities in each knowledge triple are standardized to determine their corresponding standard identifiers. The standardized knowledge triples are integrated to construct an initial knowledge graph; The initial knowledge graph is subjected to quality verification and conflict resolution to obtain the AML domain knowledge graph.

[0010] Preferably, the training of the bone marrow morphology prediction model includes at least the following model training steps: Training samples of AML patients with known standard chemotherapy bone marrow morphology were obtained from the sample set. The training samples included bone marrow next-generation sequencing data of the AML patients before treatment and corresponding real bone marrow morphology labels. A second list of associated mutant genes was extracted from the bone marrow next-generation sequencing data in the training samples; The second list of associated mutant genes is mapped to the AML domain knowledge graph to obtain each second target node in the AML domain knowledge graph that is successfully mapped by the second list of associated mutant genes. Extract the second feature clusters corresponding to each second target node from the AML domain knowledge graph; The second feature clusters corresponding to each second target node are input into the model to be trained to obtain the second expected bone marrow morphology of the AML patient receiving standard chemotherapy. The second expected bone marrow morphology is compared with the real bone marrow morphology labels in the training samples, and the parameters in the model to be trained are corrected based on the comparison results. Determine if the loop termination condition is met; If the conditions are met, the training loop is terminated, and the model to be trained with the corrected parameters is used as the bone marrow morphology prediction model; or, If the condition is not met, new training samples are obtained from the sample set, and the model training steps are executed again.

[0011] Preferably, the feature clusters corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy, including: The graph embedding encoder set at the front end of the bone marrow morphology prediction model transforms the feature clusters corresponding to each target node into corresponding feature vectors. The feature vectors corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy.

[0012] Preferably, the graph embedding encoder is specifically a graph neural network; and, The graph embedding encoder transforms the feature clusters corresponding to each target node into corresponding feature vectors in the following manner: The graph embedding encoder aggregates the neighborhood information of each node in the feature cluster through a message passing mechanism to generate the node embedding vector of each node. Then, the graph pooling layer aggregates the node embedding vectors into a single fixed-dimensional graph-level embedding vector to obtain the feature vector corresponding to the target node.

[0013] Preferably, extracting gene mutation features from the bone marrow next-generation sequencing data specifically includes: Obtain each mutated gene from the bone marrow next-generation sequencing data, wherein the mutated gene specifically includes any one or more of the following: single nucleotide variant genes, insertion or deletion mutation genes, and fusion genes; Screening out associated variant genes that appear in the preset AML-related gene set from various variant genes; The selected associated variant genes are used to generate the gene mutation list.

[0014] Secondly, the present invention provides a bone marrow morphology prediction device for AML chemotherapy regimens based on next-generation bone marrow sequencing, comprising: The acquisition unit is used to acquire bone marrow next-generation sequencing data of AML patients before treatment and extract a list of associated mutated genes from the bone marrow next-generation sequencing data. A mapping unit is used to map the list of associated mutant genes to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that has been successfully mapped by the list of associated mutant genes. The nodes of the AML domain knowledge graph are used to represent biological entities, and the edges are used to represent the biological relationships between the corresponding biological entities. An extraction unit is used to extract feature clusters corresponding to each target node from the AML domain knowledge graph, wherein the feature clusters are connected subgraphs determined by predefined topological rules with the target node as the center; The prediction unit is used to input the feature clusters corresponding to each target node into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy.

[0015] The bone marrow morphology prediction method for AML chemotherapy regimens based on next-generation bone marrow sequencing provided in this application includes acquiring bone marrow next-generation sequencing data of AML patients before treatment, extracting a list of associated mutated genes from the bone marrow next-generation sequencing data, mapping the list of associated mutated genes to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the list of associated mutated genes. The nodes in the AML domain knowledge graph are used to represent biological entities, and the edges are used to represent the biological relationships between corresponding biological entities. Then, the feature clusters corresponding to each target node are extracted from the AML domain knowledge graph. The feature clusters are connected subgraphs determined according to predefined topological rules with the target node as the center. Then, the feature clusters corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy. This method introduces a knowledge graph in the AML domain to obtain feature clusters, thereby achieving semantic dimensionality enhancement from isolated gene mutations associated with mutated genes to multidimensional biological functional networks. It then combines a bone marrow morphology prediction model to obtain the expected bone marrow morphology of AML patients receiving standard chemotherapy, solving the technical problem that existing technologies cannot predict bone marrow morphology before chemotherapy, thus failing to prospectively identify patients with primary refractory disease. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a method for predicting bone marrow morphology in AML chemotherapy regimens based on next-generation bone marrow sequencing, provided by the present invention. Figure 2 A schematic diagram of the training process of the bone marrow morphology prediction model in the bone marrow morphology prediction method for AML chemotherapy regimens based on next-generation bone marrow sequencing provided by the present invention. Figure 3 A structural block diagram of a device for predicting bone marrow morphology in AML chemotherapy regimens based on next-generation bone marrow sequencing, provided by the present invention. Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this 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 this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] As mentioned earlier, bone marrow morphology results after chemotherapy are considered the "gold standard" for determining whether AML patients are in primary refractory disease after receiving standard chemotherapy. Therefore, if the expected bone marrow morphology after standard chemotherapy can be assessed before a patient undergoes AML chemotherapy, the risk of primary refractory disease can be prospectively evaluated, providing a crucial basis for developing more effective individualized treatment plans. However, current clinical practice lacks effective technical means to achieve this goal before treatment.

[0020] In view of this, embodiments of this application provide a method and apparatus for predicting bone marrow morphology in AML chemotherapy regimens based on next-generation bone marrow sequencing, which can be used to solve the problems in the prior art. For example... Figure 1 The diagram shown illustrates the specific flowchart of the bone marrow morphology prediction method for AML chemotherapy regimens based on next-generation bone marrow sequencing provided in this application. The method includes the following steps: Step S11: Obtain bone marrow next-generation sequencing data of AML patients before treatment, and extract a list of associated mutated genes from the bone marrow next-generation sequencing data.

[0021] The bone marrow next-generation sequencing data refers to the data obtained by performing gene sequencing on physical samples of bone marrow smears from AML patients using bone marrow next-generation sequencing technology. For example, bone marrow next-generation sequencing technology can identify mononuclear cells or CD34 in physical samples of bone marrow smears. + Cell samples are enriched and subjected to whole-exome sequencing (WES) or targeted capture sequencing to obtain next-generation sequencing data of the bone marrow. This next-generation sequencing technology is a high-throughput gene detection technology that can be used for molecular diagnosis of gene mutations, fusion genes, chromosomal abnormalities, etc., in patients.

[0022] For example, a bone marrow smear physical sample of the AML patient can be prepared first using Wright-Gymsa staining or other methods in the existing technology, and then the bone marrow next-generation sequencing technology can be used to perform gene sequencing on the bone marrow smear physical sample to obtain the bone marrow next-generation sequencing data.

[0023] The bone marrow next-generation sequencing data may include multiple mutated genes, specifically including one or more of the following: single nucleotide variants, insertion or deletion variants, and fusion genes. From the perspective of the intrinsic driving force of the disease, some of these mutated genes may be related to the AML patient's disease (i.e., having AML). In other words, these mutated genes play a driving role in the pathogenesis of AML, and these related mutated genes can be called associated mutated genes. Of course, some of these mutated genes may not be related to the AML patient's disease; these unrelated mutated genes can be called unassociated mutated genes.

[0024] In this application, the method for extracting the list of associated mutated genes from the bone marrow next-generation sequencing data can be as follows: first, obtain each mutated gene in the bone marrow next-generation sequencing data (because some genes have not mutated), and then screen out the associated mutated genes that appear in a pre-defined AML-related gene set. This pre-defined AML-related gene set records each gene related to AML. Then, generate the gene mutation list from the screened associated mutated genes. In practical applications, this pre-defined AML-related gene set includes at least the following genes: TP53, FLT3, NPM1, DNMT3A, IDH1, IDH2, RUNX1, and ASXL1.

[0025] Therefore, one way to extract the list of associated mutated genes from the bone marrow next-generation sequencing data is to first extract the mutated genes from the bone marrow next-generation sequencing data, for example, by using bioinformatics tools such as GATK, VarScan, and Mutect2. Then, by combining the pre-defined AML-related gene set, further screening of associated variant genes can be performed on these mutated genes, and then generating the gene mutation list from these associated variant genes.

[0026] Step S12: Map the list of associated mutant genes to the AML domain knowledge graph to obtain the target nodes in the AML domain knowledge graph that have been successfully mapped by the list of associated mutant genes.

[0027] This AML domain knowledge graph is a structured, semantic, and computable biomedical knowledge base built specifically for acute myeloid leukemia (AML). The bottom layer of this AML domain knowledge graph can be triples consisting of head entities, relationships, and tail entities. Structurally, the nodes of this AML domain knowledge graph represent biological entities (including the aforementioned head and tail entities), which can include: genes, proteins, signaling pathways, biological processes, molecular functions (such as protein tyrosine kinase activity), diseases (AML), drugs, and phenotypes. The edges of this AML domain knowledge graph represent the biological relationships between corresponding biological entities, which can include involved, encoded, regulated, inhibited, activated, located, treated, and associated with.

[0028] In step S12, the specific implementation method can be to first obtain the standard identifier of each associated variant gene in the associated mutant gene list. The standard identifier is an internationally recognized number used to uniquely and unambiguously represent the corresponding associated variant gene. Its core function is to eliminate semantic inconsistencies caused by historical naming, species aliases, spelling variants, or database heterogeneity. Therefore, in order to ensure the accuracy of subsequent association with nodes in the AML domain knowledge graph, the standard identifier of each associated variant gene in the associated mutant gene list can be obtained first. In practical applications, the standard identifier can be specifically HGNC ID or NCBI Gene ID.

[0029] For example, you can first obtain the NCBI Gene ID of each associated variant gene in the list of associated mutant genes, and use it as the standard identifier for each associated variant gene.

[0030] Next, for each associated mutant gene, the standard identifier of the associated mutant gene can be matched with the standard identifier of the biological entity represented by each node in the AML domain knowledge graph to locate the target node corresponding to the associated mutant gene. For example, the standard identifier of the biological entity represented by each node in the AML domain knowledge graph can be obtained. This standard identifier can also be HGNC ID or NCBI Gene ID, and then standardized with the standard identifier of the associated mutant gene. Then, the standard identifier of the associated mutant gene is matched with the standard identifier of the biological entity represented by each node in the AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the list of associated mutant genes, that is, to obtain each target node in the AML domain knowledge graph that is successfully matched.

[0031] In this step, each associated mutant gene in the associated mutant gene list can be designated as the current associated mutant gene. Then, the standard identifier of this current associated mutant gene is matched against the standard identifiers of the biological entities represented by each node in the AML domain knowledge graph, thereby obtaining the successfully matched target nodes. In practical applications, the specific matching method can be exact matching (meaning a perfect match), fuzzy matching, or semantic matching. Since this application uses matching between standard identifiers, exact matching can be used to improve the accuracy of the matching.

[0032] Step S13: Extract the feature clusters corresponding to each target node from the AML domain knowledge graph.

[0033] The feature cluster is a connected subgraph centered on the target node and determined according to predefined topological rules. Therefore, it is a miniature knowledge network in the AML domain knowledge graph. In this way, isolated nodes in the AML domain knowledge graph can be upgraded into network units with functional context through the feature cluster, and thus used more comprehensively for subsequent prediction.

[0034] Specifically, the predefined topology rule can be a K-hop neighborhood rule. This means starting from the target node, traversing all connected edges in any direction (without limiting incoming or outgoing edges), and including all nodes reachable within K steps and their connecting edges to form a connected subgraph. In this application, K can be an integer from 1 to 5, such as 1, 2, 3, 4, or 5. A larger value of K results in more nodes in the connected subgraph, leading to more comprehensive coverage but also increasing the likelihood of introducing too many noisy nodes. Conversely, a smaller value of K results in fewer nodes in the connected subgraph, making it less likely to introduce noisy nodes, but also covering fewer nodes and potentially missing some useful ones.

[0035] Therefore, the specific implementation of step S13 can be as follows: for each target node, the feature clusters corresponding to the target node can be extracted in the following way. Specifically, the position of the target node in the AML domain knowledge graph can be determined, and the connected subgraph of each node and edge in the K-hop neighbor rule can be obtained with the position as the center, which is used as the feature clusters corresponding to the target node.

[0036] Of course, for each target node, before using the K-hop neighbor rule to determine its characteristic clusters, the corresponding K value can be determined in the following ways. For example, one way is to directly fix the K value to a certain fixed value, such as 2 or 3; another way is to combine the network centrality of the target node to determine its K value. In this way, the network centrality of the target node in the AML domain knowledge graph can be determined first, and then the K value corresponding to the target node can be determined based on the network centrality. In this case, the lower the network centrality of the target node, the larger the value of K, or the higher the network centrality of the target node, the smaller the value of K.

[0037] Network centrality is a quantitative indicator that characterizes the structural importance and information hub status of a target node in the AML knowledge graph. It is used to describe the density, reachability, and path control capabilities of the connections between the biological entity corresponding to the target node and the biological entities corresponding to other nodes. Therefore, in this application, the value of K can be determined by combining the network centrality of the target node. In practical applications, the network centrality can specifically be degree centrality, closeness centrality, or betweenness centrality.

[0038] The degree centrality is the number of edges directly connected to the target node (i.e., the number of adjacent nodes), reflecting the activity level of the target node as a local interaction center; the proximity centrality is the reciprocal of the sum of the shortest path lengths from the target node to all other reachable nodes in the AML domain knowledge graph, thus reflecting the global reachability efficiency of the target node as an information propagation relay station; the betweenness centrality is the frequency percentage of the target node located on the shortest path between any two nodes in the graph, thus reflecting the control weight of the target node on cross-module information flow.

[0039] After determining the network centrality of the target node in the AML domain knowledge graph, the value of K for each target node can be determined based on the network centrality. Since the value of K in the K-hop neighbor rule is the topological expansion radius in the feature cluster extraction process, which is used to represent the maximum number of hops extending outward from the target node, it will directly affect the coverage breadth and semantic density of the extracted connected subgraph. Therefore, in this application, the value of K and the network centrality have a strictly inverse monotonic mapping relationship to improve the accuracy of subsequent predictions. That is, the lower the network centrality of the target node, the larger the value of K, or the higher the network centrality of the target node, the smaller the value of K.

[0040] This application employs a segmented mapping approach to determine the K value for each target node based on network centrality. For example, the network centrality can be divided into multiple intervals, each corresponding to a different K value. Specifically, the network centrality can be divided into three intervals: [0, 0.3), [0.3, 0.7), and [0.7, 1]. The K value for the interval [0, 0.3) is 3, for [0.3, 0.7) it is 2, and for [0.7, 1] it is 1. Then, for each target node, the interval to which its network centrality belongs is determined, thereby determining the K value for that target node. After obtaining the K values ​​for each target node, the characteristic clusters corresponding to each target node can be obtained using the K-hop neighbor rule.

[0041] Step S14: Input the feature clusters corresponding to each target node into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy.

[0042] After obtaining the feature clusters corresponding to each target node through step S13, in step S14, the feature clusters corresponding to each target node can be input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy. The bone marrow morphology prediction model can be used to predict the expected bone marrow morphology of the AML patient receiving standard chemotherapy. The expected bone marrow morphology may include the percentage of bone marrow blast cells, the degree of erythroid / granulocytic / megakaryocytic developmental abnormalities, whether complete remission, partial remission, or no remission has been achieved, etc.

[0043] In this application, since the feature clusters are connected subgraphs in the AML domain knowledge graph, they need to be converted into feature vectors before being input into the bone marrow morphology prediction model. Therefore, a graph embedding encoder can be set at the front end of the bone marrow morphology prediction model. The specific implementation of step S14 is as follows: first, the feature clusters corresponding to each target node are converted into corresponding feature vectors by the graph embedding encoder set at the front end of the bone marrow morphology prediction model. Then, the feature vectors corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy.

[0044] In practical applications, this graph embedding encoder can be specifically a graph neural network (GNN). In this way, the graph embedding encoder can transform the feature clusters corresponding to each target node into corresponding feature vectors in the following way: the number of layers of the GNN can be set to 2 to 4, the hidden dimension of each layer can be 64 to 256, and the activation function can be LeakyReLU or GeLU. When the feature cluster size is large (number of nodes > 100), the GNN can also adopt a sparse adjacency sampling strategy to reduce computational complexity.

[0045] Specifically, for the graph embedding encoder to transform the feature clusters corresponding to each target node into corresponding feature vectors, the graph embedding encoder can aggregate the neighborhood information of each node in the feature cluster through a message passing mechanism to generate the node embedding vector of each node. For example, GNN performs message passing in an iterative manner, so each layer corresponds to one information aggregation process. Specifically, in the i-th layer, for any node v in the feature cluster, its node embedding vector h_v(i) is jointly updated and generated by its own previous layer embedding h_v(i-1) and the previous layer embedding h_u(i-1) of all its first-order neighbor nodes u∈N(v). The update formula can be expressed as: h_v(i)=σ(W(i)·AGGREGATE({h_u(i-1):u∈N(v)}∪{h_v(i-1)})).

[0046] In this update formula, AGGREGATE is the aggregation function, which in practical applications can be a summation function (Sum), a mean function (Mean), a maximum value function (Max), or an attention-weighted mean function; σ is a non-linear activation function; and W(i) is the learnable weight matrix of the i-th layer. This mechanism enables node v not only to perceive directly connected biological entities, but also to indirectly perceive more distant functionally related entities through multi-hop propagation, thereby capturing long-range biological dependencies such as signal cascades, feedback regulation, and pathway crossings commonly found in AML, significantly improving the overall ability to characterize pathway functional states.

[0047] After obtaining the node embedding vectors of each node, a graph pooling layer can be used to aggregate these vectors into a single fixed-dimensional graph-level embedding vector, thus obtaining the feature vector corresponding to the target node. The graph pooling method used in this layer can be global average pooling, global max pooling, hierarchical pooling (such as DiffPool), or attention-based set pooling. These methods aggregate the node embedding vectors into a single fixed-dimensional graph-level embedding vector, which can be, for example, 128 or 256 dimensions.

[0048] After obtaining the feature vectors corresponding to each target node, these vectors can be input into the bone marrow morphology prediction model. For example, the feature vectors corresponding to each target node can be fused using average pooling to generate a comprehensive feature vector representing the AML patient. This comprehensive feature vector is then input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient after receiving standard chemotherapy. Furthermore, this method can be further enhanced by assessing the risk level of primary refractory disease after standard chemotherapy based on the expected bone marrow morphology. For instance, the expected bone marrow morphology can be provided as a report to relevant physicians, who can then assess the risk level of primary refractory disease after standard chemotherapy and determine whether to proceed with the standard chemotherapy treatment regimen for the AML patient.

[0049] It should be further noted that the bone marrow morphology prediction model can be pre-trained in this application, wherein, for example... Figure 2 As shown, the training of this bone marrow morphology prediction model includes at least the following model training steps: Step S141: Obtain training samples of AML patients with known standard post-chemotherapy bone marrow morphology from the sample set. The training samples include pre-treatment bone marrow next-generation sequencing data of AML patients and corresponding real bone marrow morphology labels.

[0050] In practical applications, raw data can be collected first, which includes bone marrow next-generation sequencing data of multiple AML patients before treatment, as well as the actual bone marrow morphology of each AML patient after standard chemotherapy. In this way, the bone marrow next-generation sequencing data of each AML patient before treatment and the corresponding actual bone marrow morphology labels can be extracted from the raw data as training samples and the sample set can be generated.

[0051] In this step S141, training samples of AML patients with known bone marrow morphology after standard chemotherapy can be obtained from the sample set, for example, a training sample can be randomly obtained from the sample set.

[0052] Step S142: Extract the second associated mutation gene list from the bone marrow next-generation sequencing data in the training sample.

[0053] The specific implementation of step S142 can be done in the same way as step S11 above, by extracting the list of associated mutation genes (that is, the second list of associated mutation genes) from the bone marrow next-generation sequencing data of the AML patients before treatment in the training sample. This will not be elaborated here.

[0054] Step S143: Map the second associated mutant gene list to the AML domain knowledge graph to obtain each second target node in the AML domain knowledge graph that has been successfully mapped by the second associated mutant gene list.

[0055] The specific implementation of step S143 can be done in the same way as step S12 above, to map the second associated mutation gene list to the AML domain knowledge graph, thereby obtaining each target node in the AML domain knowledge graph that has been successfully mapped by the second associated mutation gene list (as the second target node). This will not be elaborated further here.

[0056] Step S144: Extract the second feature clusters corresponding to each second target node from the AML domain knowledge graph.

[0057] The specific implementation of step S144 can be done in the same way as step S13 above, to extract the feature clusters corresponding to each second target node from the AML domain knowledge graph (as the second feature clusters), which will not be elaborated here.

[0058] Step S145: Input the second feature clusters corresponding to each second target node into the model to be trained to obtain the second expected bone marrow morphology of the AML patient receiving standard chemotherapy.

[0059] The specific implementation of step S145 can be done in the same way as step S14 above, by inputting the second feature clusters corresponding to each second target node into the model to be trained (because it is still in the model training stage, it is called the model to be trained) to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy, which is used as the second expected bone marrow morphology. This will not be elaborated here.

[0060] Step S146: Compare the second expected bone marrow morphology with the real bone marrow morphology label in the training sample, and correct the parameters in the model to be trained based on the comparison results.

[0061] For example, the deviation between the second expected bone marrow morphology and the real bone marrow morphology label can be calculated using a loss function, and then the parameters in the model to be trained can be corrected based on the backpropagation of this deviation.

[0062] Step S147: Determine whether the loop termination condition is met. If yes, proceed to step S148; otherwise, proceed to step S149.

[0063] The loop termination condition can be, for example, reaching the upper limit of the number of loop rounds, or the model to be trained reaching convergence. In these cases, the loop termination condition is met; otherwise, it is not met.

[0064] Step S148: Terminate the loop training and use the parameter-corrected model to be trained as the bone marrow morphology prediction model.

[0065] Step S149: Obtain new training samples from the sample set and execute the model training steps again.

[0066] If the loop termination condition is met, the loop training can be terminated in step S148, and the model to be trained with the corrected parameters can be used as the bone marrow morphology prediction model. Otherwise, if the loop termination condition is not met, new training samples can be obtained from the sample set in step S149, and the model training steps can be executed again, that is, the operations of steps S141 to S147 can be executed again, and then it can be determined again whether the loop termination condition is met, until the loop terminates.

[0067] The bone marrow morphology prediction method for AML chemotherapy regimens based on next-generation bone marrow sequencing provided in this application includes acquiring bone marrow next-generation sequencing data of AML patients before treatment, extracting a list of associated mutated genes from the bone marrow next-generation sequencing data, mapping the list of associated mutated genes to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the list of associated mutated genes. The nodes in the AML domain knowledge graph are used to represent biological entities, and the edges are used to represent the biological relationships between corresponding biological entities. Then, the feature clusters corresponding to each target node are extracted from the AML domain knowledge graph. The feature clusters are connected subgraphs determined according to predefined topological rules with the target node as the center. Then, the feature clusters corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy. This method introduces a knowledge graph in the AML domain to obtain feature clusters, thereby achieving semantic dimensionality enhancement from isolated gene mutations associated with mutated genes to multidimensional biological functional networks. It then combines a bone marrow morphology prediction model to obtain the expected bone marrow morphology of AML patients receiving standard chemotherapy, solving the technical problem that existing technologies cannot predict bone marrow morphology before chemotherapy, thus failing to prospectively identify patients with primary refractory disease.

[0068] In steps S12 and S13 above, an AML domain knowledge graph was mentioned. This AML domain knowledge graph has a significant impact on the accuracy of the final identification result of this application. Therefore, a breast cancer approach can be further provided to pre-generate this domain knowledge graph. This involves first extracting knowledge triples from multiple biomedical databases. These knowledge triples include head entities, relations, and tail entities. Specifically, these biomedical databases include any one or more of the following: NCBI Gene Database, KEGG Pathway Database, Reactome Pathway Database, GO Annotation Database, DrugBank Database, STRING Protein Interaction Database, and ClinVar Clinical Variation Database. In other words, knowledge triples can be extracted from these authoritative biomedical databases. Then, the head and tail entities in each knowledge triple can be standardized to determine their corresponding standard identifiers. The purpose of this standardization is to convert the head and tail entities in the knowledge triples into corresponding standard identifiers, thereby unifying the description method. In particular, since the above involves extracting knowledge triples from multiple biomedical databases, the names and numbers of the same entities may differ in these databases. Therefore, this standardization process can be used to convert them into standard identifiers.

[0069] The standardized knowledge triples can then be further integrated to construct an initial knowledge graph. Each node in this initial knowledge graph is the head entity and tail entity in the triples, and the edges are the relations in the triples. The method of constructing the knowledge graph by integrating the knowledge triples can be based on existing technologies, and there are no specific limitations on this method.

[0070] After obtaining the initial knowledge graph, further quality checks and conflict resolution can be performed to ultimately obtain the AML domain knowledge graph. Quality checks can include three types of checks: entity integrity checks, relation consistency checks, and structural connectivity checks. Entity integrity checks can traverse all nodes to verify whether the biological entities of each node conform to the latest published standard identifiers, and remove obsolete or redirected nodes. Relation consistency checks are used to detect logical contradictions. Structural connectivity checks can use the BFS algorithm to verify whether there are isolated subgraphs (node ​​degree = 0) in the initial knowledge graph.

[0071] Conflict resolution can employ a hierarchical strategy. The first layer is automatic resolution, where for multiple relationships between the same head-tail biological entity pair, the edge with the highest confidence value is retained as the unique edge. The second layer is expert arbitration, where semantic conflicts that cannot be automatically determined are submitted to a hematology-oncology expert committee for adjudication. This hierarchical conflict resolution approach optimizes the initial knowledge graph, ultimately yielding the AML domain knowledge graph.

[0072] Based on the same inventive concept as the bone marrow morphology prediction method for AML chemotherapy regimens based on next-generation bone marrow sequencing provided in the embodiments of this application, the embodiments of this application can also provide a bone marrow morphology prediction device for AML chemotherapy regimens based on next-generation bone marrow sequencing. For any unclear points regarding the content of this device embodiment, please refer to the relevant content in the above method embodiments. Figure 3 The diagram shows a schematic representation of the bone marrow morphology prediction device 30 (hereinafter referred to as device 30) for AML chemotherapy regimens based on next-generation bone marrow sequencing. Device 30 includes: an acquisition unit 301, a mapping unit 302, an extraction unit 303, and a prediction unit 304, wherein: The acquisition unit 301 is used to acquire bone marrow next-generation sequencing data of AML patients before treatment and extract a list of associated mutated genes from the bone marrow next-generation sequencing data. Mapping unit 302 is used to map the list of associated mutant genes to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that has been successfully mapped by the list of associated mutant genes. The nodes of the AML domain knowledge graph are used to represent biological entities, and the edges are used to represent the biological relationships between the corresponding biological entities. Extraction unit 303 is used to extract feature clusters corresponding to each target node from the AML domain knowledge graph, wherein the feature clusters are connected subgraphs determined by predefined topology rules with the target node as the center; The prediction unit 304 is used to input the feature clusters corresponding to each target node into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy. The device 30 provided in the embodiments of this application adopts the same inventive concept as the method provided in the embodiments of this application. Since the method can solve the problems in the prior art, the device 30 can also solve the problems in the prior art. This will not be elaborated here.

[0073] Specifically, mapping the list of associated mutant genes to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the list of associated mutant genes may include: Obtain the standard identifier of each associated variant gene in the list of associated mutant genes; For each associated variant gene, the standard identifier of the associated variant gene is matched with the standard identifier of the biological entity represented by each node in the AML domain knowledge graph to locate the target node corresponding to the associated variant gene; wherein, the standard identifier is specifically HGNC ID or NCBI Gene ID.

[0074] Specifically, extracting feature clusters corresponding to each target node from the AML domain knowledge graph includes extracting the feature clusters corresponding to each target node in the following manner: Determine the location of the target node in the AML domain knowledge graph, and obtain the connected subgraph of each node and edge within the K-hop neighbor rule with the location as the center, as the feature cluster corresponding to the target node, where K is an integer from 1 to 5.

[0075] The device 30 may further include a K-value determination unit, used to determine the network centrality of the target node in the AML domain knowledge graph; and to determine the K value corresponding to the target node based on the network centrality, wherein the lower the network centrality of the target node, the larger the value of K, or the higher the network centrality of the target node, the smaller the value of K.

[0076] The device 30 may further include a knowledge graph generation unit for extracting knowledge triples from multiple biomedical databases. Each knowledge triple includes a head entity, a relation, and a tail entity. Specifically, the biomedical databases include any one or more of the following: NCBI Gene database, KEGG pathway database, Reactome pathway database, GO annotation database, DrugBank database, STRING protein interaction database, and ClinVar clinical variant database. The head and tail entities in each knowledge triple are standardized to determine their corresponding standard identifiers. The standardized knowledge triples are then integrated to construct an initial knowledge graph. The initial knowledge graph undergoes quality verification and conflict resolution to obtain the AML domain knowledge graph.

[0077] The training of the bone marrow morphology prediction model includes at least the following model training steps: Training samples of AML patients with known standard chemotherapy bone marrow morphology were obtained from the sample set. The training samples included bone marrow next-generation sequencing data of the AML patients before treatment and corresponding real bone marrow morphology labels. A second list of associated mutant genes was extracted from the bone marrow next-generation sequencing data in the training samples; The second list of associated mutant genes is mapped to the AML domain knowledge graph to obtain each second target node in the AML domain knowledge graph that is successfully mapped by the second list of associated mutant genes. Extract the second feature clusters corresponding to each second target node from the AML domain knowledge graph; The second feature clusters corresponding to each second target node are input into the model to be trained to obtain the second expected bone marrow morphology of the AML patient receiving standard chemotherapy. The second expected bone marrow morphology is compared with the real bone marrow morphology labels in the training samples, and the parameters in the model to be trained are corrected based on the comparison results. Determine if the loop termination condition is met; If the conditions are met, the training loop is terminated, and the model to be trained with the corrected parameters is used as the bone marrow morphology prediction model; or, If the condition is not met, new training samples are obtained from the sample set, and the model training steps are executed again.

[0078] Specifically, the feature clusters corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy. This may include: The graph embedding encoder set at the front end of the bone marrow morphology prediction model transforms the feature clusters corresponding to each target node into corresponding feature vectors. The feature vectors corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy.

[0079] Specifically, the graph embedding encoder is a graph neural network; and, The graph embedding encoder transforms the feature clusters corresponding to each target node into corresponding feature vectors in the following manner: The graph embedding encoder aggregates the neighborhood information of each node in the feature cluster through a message passing mechanism to generate the node embedding vector of each node. Then, the graph pooling layer aggregates the node embedding vectors into a single fixed-dimensional graph-level embedding vector to obtain the feature vector corresponding to the target node.

[0080] Specifically, extracting gene mutation features from the bone marrow next-generation sequencing data may include: Obtain each mutated gene from the bone marrow next-generation sequencing data, wherein the mutated gene specifically includes any one or more of the following: single nucleotide variant genes, insertion or deletion mutation genes, and fusion genes; Screening out associated variant genes that appear in the preset AML-related gene set from various variant genes; The selected associated variant genes are used to generate the gene mutation list.

[0081] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute the bone marrow morphology prediction method for AML chemotherapy regimens based on next-generation bone marrow sequencing provided in this application embodiment. The method includes acquiring bone marrow next-generation sequencing data of AML patients before treatment, extracting a list of associated mutation genes from the bone marrow next-generation sequencing data, and then mapping the list of associated mutation genes to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the list of associated mutation genes. The nodes of the AML domain knowledge graph are used to represent biological entities, and the edges are used to represent the biological relationships between corresponding biological entities. Then, the feature clusters corresponding to each target node are extracted from the AML domain knowledge graph. The feature clusters are connected subgraphs determined according to predefined topological rules with the target node as the center. Then, the feature clusters corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy. This method introduces a knowledge graph in the AML domain to obtain feature clusters, thereby achieving semantic dimensionality enhancement from isolated gene mutations associated with mutated genes to multidimensional biological functional networks. It then combines a bone marrow morphology prediction model to obtain the expected bone marrow morphology of AML patients receiving standard chemotherapy, solving the technical problem that existing technologies cannot predict bone marrow morphology before chemotherapy, thus failing to prospectively identify patients with primary refractory disease.

[0082] In practical applications, this electronic device can be either a user-side electronic device or a server-side electronic device.

[0083] Obviously, since the processor 410 can call the logical instructions in the memory 430 to execute the method provided in the embodiments of this application, it can also solve the problems in the prior art.

[0084] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the bone marrow morphology prediction method for AML chemotherapy regimens based on bone marrow next-generation sequencing provided in the embodiments of this application. The method includes acquiring bone marrow next-generation sequencing data of AML patients before treatment, extracting a list of associated mutation genes from the bone marrow next-generation sequencing data, and then mapping the list of associated mutation genes to an AML domain knowledge graph to obtain each target node in the AML domain knowledge graph that is successfully mapped by the list of associated mutation genes. The nodes of the AML domain knowledge graph are used to represent biological entities, and the edges are used to represent the biological relationships between corresponding biological entities. Then, the feature clusters corresponding to each target node are extracted from the AML domain knowledge graph. The feature clusters are connected subgraphs determined according to predefined topological rules with the target node as the center. Then, the feature clusters corresponding to each target node are input into the bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy. This method introduces a knowledge graph in the AML domain to obtain feature clusters, thereby achieving semantic dimensionality enhancement from isolated gene mutations associated with mutated genes to multidimensional biological functional networks. It then combines a bone marrow morphology prediction model to obtain the expected bone marrow morphology of AML patients receiving standard chemotherapy, solving the technical problem that existing technologies cannot predict bone marrow morphology before chemotherapy, thus failing to prospectively identify patients with primary refractory disease.

[0086] Obviously, since the computer can execute the method provided in the embodiments of this application when the computer program is executed by the processor, it can also solve the problems in the prior art.

[0087] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, provides the method provided in the embodiments of this application.

[0088] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting bone marrow morphology of AML chemotherapy regimen based on bone marrow next-generation sequencing, characterized in that, The method comprises the following steps: obtaining bone marrow second-generation sequencing data of an AML patient before treatment, and extracting a list of associated mutation genes from the bone marrow second-generation sequencing data; mapping the list of associated mutation genes to an AML field knowledge graph to obtain each target node in the AML field knowledge graph that is successfully mapped by the list of associated mutation genes, wherein the nodes of the AML field knowledge graph represent biological entities, and the edges represent biological relationships between the corresponding biological entities; extracting a feature cluster corresponding to each target node from the AML field knowledge graph, wherein the feature cluster is a connected subgraph centered on the target node and determined according to a predefined topological rule; inputting the feature cluster corresponding to each target node into a bone marrow morphology prediction model to obtain the expected bone marrow morphology of the AML patient receiving standard chemotherapy.

2. The method of claim 1, wherein, Mapping the list of associated mutation genes to the AML field knowledge graph to obtain each target node in the AML field knowledge graph that is successfully mapped by the list of associated mutation genes, specifically comprising: obtaining the standard identifier of each associated variant gene in the list of associated mutation genes; for each associated variant gene, matching the standard identifier of the associated variant gene with the standard identifier of the biological entity represented by each node in the AML field knowledge graph to locate the target node corresponding to the associated variant gene; wherein the standard identifier is specifically HGNC ID or NCBI Gene ID.

3. The method of claim 1, wherein, Extracting a feature cluster corresponding to each target node from the AML field knowledge graph, specifically comprising extracting the feature cluster corresponding to each target node by the following way: determining the position of the target node in the AML field knowledge graph, and taking the position as the center to obtain the connected subgraph of each node and edge within K-hop neighbor rules as the feature cluster corresponding to the target node, wherein K is an integer from 1 to 5.

4. The method of claim 3, wherein, The method further comprises: determining the network centrality of the target node in the AML field knowledge graph; determining the value of K corresponding to the target node according to the network centrality, wherein the value of K is larger when the network centrality of the target node is lower, or the value of K is smaller when the network centrality of the target node is higher.

5. The method of claim 1, wherein, The method further comprises: extracting knowledge triples from a plurality of biomedical databases, wherein the knowledge triples comprise a head entity, a relationship, and a tail entity; wherein the biomedical databases specifically comprise any one or more of the following: NCBI Gene database, KEGG pathway database, Reactome pathway database, GO annotation database, DrugBank database, STRING protein interaction database, ClinVar clinical variation database; standardizing the head entity and the tail entity in each knowledge triple to determine the corresponding standard identifiers; integrating the standardized knowledge triples to construct an initial knowledge graph; Quality checking and conflict resolution are performed on the initial knowledge graph to obtain the AML field knowledge graph.

6. The method of claim 1, wherein, The training of the bone marrow morphology prediction model at least includes the following model training steps: A training sample of an AML patient with known bone marrow morphology after standard chemotherapy is obtained from a sample set, and the training sample includes bone marrow second-generation sequencing data of the AML patient before treatment and a corresponding true bone marrow morphology label; A second associated mutation gene list is extracted from the bone marrow second-generation sequencing data in the training sample; The second associated mutation gene list is mapped to the AML field knowledge graph to obtain each second target node in the AML field knowledge graph that is successfully mapped by the second associated mutation gene list; A second feature cluster corresponding to each second target node is extracted from the AML field knowledge graph; Each second target node is respectively inputted into a to-be-trained model to obtain a second expected bone marrow morphology of the AML patient receiving standard chemotherapy; The second expected bone marrow morphology is compared with the true bone marrow morphology label in the training sample, and the parameters in the to-be-trained model are corrected according to the comparison result; It is judged whether the loop termination condition is met; If yes, the loop training is terminated, and the to-be-trained model after parameter correction is taken as the bone marrow morphology prediction model; Or, If not, a new training sample is obtained from the sample set, and the model training steps are executed again.

7. The method of claim 1, wherein, Each target node is respectively inputted into a to-be-trained model to obtain a second expected bone marrow morphology of the AML patient receiving standard chemotherapy, including: Each target node is respectively inputted into a to-be-trained model to obtain a second expected bone marrow morphology of the AML patient receiving standard chemotherapy, including: The graph embedding encoder is specifically a graph neural network; and 8. The method of claim 7, wherein, The graph embedding encoder respectively converts the feature cluster corresponding to each target node into a corresponding feature vector by the following method: The graph embedding encoder aggregates the neighborhood information of each node in the feature cluster through a message passing mechanism to generate a node embedding vector of each node, and then aggregates the node embedding vectors into a single fixed-dimensional graph-level embedding vector through a graph pooling layer to obtain the feature vector corresponding to the target node. The gene mutation features are extracted from the bone marrow second-generation sequencing data, specifically including:

9. The method of claim 1, wherein, Each variant gene in the bone marrow second-generation sequencing data is obtained, wherein the variant gene specifically includes any one or more of the following: single nucleotide variant gene, insertion or deletion mutation gene, fusion gene; The associated variant genes appearing in a preset AML-related gene set are screened from each variant gene; The screened associated variant genes generate the gene mutation list. The method comprises the following steps:

10. A bone marrow morphology prediction device for AML chemotherapy regimen based on bone marrow next-generation sequencing, characterized by, A unit is configured to obtain bone marrow second-generation sequencing data of an AML patient before treatment and extract an associated mutation gene list from the bone marrow second-generation sequencing data; ​ mapping unit, configured to map the associated mutation gene list to an AML field knowledge graph to obtain each target node in the AML field knowledge graph that is successfully mapped by the associated mutation gene list, wherein a node of the AML field knowledge graph is used to represent a biological entity, and an edge is used to represent a biological relationship between corresponding biological entities; an extraction unit configured to extract a feature cluster corresponding to each target node from the AML field knowledge graph, wherein the feature cluster is a connected subgraph centered on the target node and determined according to a predefined topological rule; a prediction unit configured to input the feature cluster corresponding to each target node into a bone marrow morphology prediction model to obtain an expected bone marrow morphology of the AML patient receiving standardized chemotherapy.

Images