A system for map-based reasoning analysis for cellular antigen processing defects

Abstract

This invention discloses a graph reasoning and analysis system for addressing defects in cellular antigen processing, belonging to the field of knowledge graph reasoning technology, to solve the problem of slow interpretable localization of defect presentation. This invention constructs a detection likelihood set by integrating HLA typing, somatic cell variation, transcriptomic expression, immunopeptidomics, and surface HLA detection. It assembles direct and cross-presentation subgraphs using pattern gating and loading complex integrity rules, generating reachable paths and observation consistency matrices. It distinguishes between structural contradictions and detection limitations, performs node pruning and link contraction based on constraint propagation to form an individualized graph, and outputs defective nodes and paths based on minimum interpretable paths and causal graph cost search. It generates validation and intervention lists and clinical decision support outputs, thereby reducing unnecessary detection and intervention attempts and improving the timeliness and consistency of case analysis and strategy formulation.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph reasoning technology, and more specifically, to a knowledge graph reasoning and analysis system for addressing deficiencies in cell antigen processing. Background Technology

[0002] In clinical immunoinformatics, researchers construct knowledge graphs of antigen processing and presentation processes, mapping genomes, transcription, immune peptide profiles, and human leukocyte antigen (HLA) detection onto the graphs according to a unified time and terminology. They then perform path reasoning on the lysis, transport, and modification links to locate impaired presentation links and form interpretable decision support.

[0003] However, existing solutions have bottlenecks. First, the edges of the map are usually static and difficult to update with the interferon state and cell type. This makes it easy to confuse reversible transcriptional repression with irreversible structural defects. Moreover, the evidence collection and processing cycles are inconsistent, and the causal relationship on the map is difficult to verify. This leads to the easy omission of immune peptide profiles in scenarios with limited sample size and modified peptides, which are often written as neutral blanks. This causes the posterior defect localization and verification order to be deviated, affecting clinical implementation and reuse. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the following solution is proposed to solve the problem of slow defect presentation interpretation and localization in the above-mentioned background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A graph reasoning and analysis system for addressing deficiencies in cell antigen processing includes a case data mapping module, an antigen assembly module, a presentation requirement determination module, an individual strategy analysis module, and a decision adjustment module, with each module connected by a signal.

[0007] The case data mapping module is used to obtain the HLA locus typing, somatic cell variation and copy number, transcriptional expression, immunopeptidomics and surface HLA detection results of cases, and to complete the standardized mapping.

[0008] The antigen assembly module is used to assemble a knowledge graph of antigen processing and presentation, including lysis, transport, modification, loading, and display, and distinguishes between direct presentation and cross-presentation.

[0009] The presentation requirement determination module is used to calculate link connectivity and observation consistency on the assembled antigen processing and presentation knowledge graph, and determine whether the presentation requirement is met based on the preset presentation requirement conditions.

[0010] The individual strategy analysis module is used to determine the association between unavailable channels and presentation patterns based on HLA site copy deletion, inactivation mutation and domain destruction when the conditions are not met. It then uses constraint propagation to generate a map editing sequence, prunes nodes and shrinks links to obtain an individualized map.

[0011] The decision adjustment module is used to perform evidence propagation and minimum interpretation path reasoning on the individualized graph by combining the detection likelihood set, outputting defect nodes, defect paths and evidence chains, and constructing a verification and intervention causal graph. It searches for the minimum cost path from the current presentation state to the target state that meets the presentation requirements, and generates clinical decisions for output.

[0012] Furthermore, the case data mapping module includes:

[0013] Based on a unified time reference and glossary, the field standardization and aliasing of HLA locus typing, somatic cell variation and copy number, transcript expression, immunopeptidomics and surface HLA detection results were standardized and compared with aliases.

[0014] Allele nomenclature standardization and heterozygosity consistency verification were performed on HLA locus typing.

[0015] Genome coordinate alignment and functional annotation were performed on somatic cell variations and copy numbers, and markers of copy deletion, inactivation mutations and domain disruption were extracted.

[0016] Gene expression status tags are generated from transcriptional expression.

[0017] Immunopeptidomics was used to perform spectroscopic identification and confidence screening, and to map peptides to presentation sites. Surface HLA detection was numerically normalized and surface expression tags were generated.

[0018] A detection likelihood set is established based on the sample input amount, peptide type, and experimental procedure parameters, and then output along with the label.

[0019] Furthermore, the requirement determination module includes:

[0020] On the assembled antigen processing and presentation knowledge graph, the set of reachable paths from input to output is calculated in the order of cleavage, transport, modification, loading, and display.

[0021] The labels output by the case data mapping module are compared with the corresponding nodes to form an observation consistency matrix;

[0022] Based on the preset presentation requirements, the set of reachable paths and the observation consistency matrix are jointly determined, and the satisfaction result is output.

[0023] Furthermore, the individual strategy analysis module includes the following when it determines that the presentation requirements are not met:

[0024] HLA site copy deletion, inactivation mutation, and domain disruption markers were extracted from the case data mapping results.

[0025] The markers are projected onto the nodes and links associated with alleles to identify the set of unavailable channels and to build a channel association mapping table.

[0026] Based on the channel association mapping table, constraint propagation is used to generate a graph editing sequence, which performs pruning on associated nodes and shrinking on dependent links to obtain an individualized graph.

[0027] Furthermore, the constraint propagation method for generating graph editing sequences includes:

[0028] A constraint-satisfying network is constructed using node states and edge channel states as variables.

[0029] Inject node predecessor availability, loading complex integrity, and allele existence as constraints into the network.

[0030] Perform constraint propagation to reduce the feasible domain of variables and identify the minimum cut set that leads to inconsistency;

[0031] The editing actions are sorted according to topological order and inconsistency resolution priority rules, and the graph editing sequence is output.

[0032] Furthermore, the decision adjustment module includes:

[0033] Evidence propagation is performed on the individualized graph by combining the detection likelihood set and calculating the interpretation score of each node and each path;

[0034] A verification and intervention causal graph is constructed based on the interventionable points and the observed outputs. Nodes represent interventionable points and observed outputs, and edges represent regulation or functional dependencies.

[0035] A path cost function is constructed based on intervention costs, expected incremental increases, and penalties for inconsistencies in evidence.

[0036] Heuristic search is used to search for the minimum cost path from the current presentation state to the target state that satisfies the presentation requirements in the causal graph;

[0037] Output the minimum cost path and its corresponding evidence chain, and generate clinical decision support output.

[0038] Furthermore, the detection likelihood set established by the case data mapping module includes:

[0039] Generate sample-level detection parameters based on the sample input volume, preprocessing method, and batch size.

[0040] Peptide-level detection parameters are generated based on peptide charge-to-mass ratio range, retention time deviation, fragment ion coverage, identification confidence level, and modification type.

[0041] Surface expression detection parameters are generated based on antibody cloning, labeling channel overflow correction, and instrument gain for surface HLA detection.

[0042] Sample-level detection parameters, peptide-level detection parameters, and surface expression detection parameters are mapped to observation confidence labels and bound to corresponding nodes or observation records.

[0043] Furthermore, when distinguishing between direct presentation and cross-presentation, the antigen assembly module performs the following actions:

[0044] Based on cell type labels and presentation pattern markers, evidence is propagated only within subplots that match them;

[0045] Using the presentation layer nodes as the merge nodes, the consistency of the results from the direct presentation subgraph and the cross presentation subgraph is checked.

[0046] When the verification is inconsistent, the conflict path is marked and a presentation mode conflict prompt is output for graph editing in the individual strategy analysis module.

[0047] Furthermore, the edit log includes:

[0048] The unique identifier of the edited node and link, the corresponding allele, and the presentation pattern marker;

[0049] The source of evidence that triggered the edit, the timestamp, and the version identifier;

[0050] Edit action types and execution order; edit the state of nodes and links before and after editing.

[0051] Rollback markers and traceability check codes are used for recovery.

[0052] Furthermore, structured messages include:

[0053] List of defective nodes, set of defective paths, and corresponding chains of evidence;

[0054] Reversibility labeling, validation program list, and intervention program list;

[0055] Data version, algorithm version, and graph version identifiers; evidence source identifiers and timestamps;

[0056] This is used for mapping message headers and fields for integration with medical information systems, and provides backfilling and auditing interfaces to record the status of receiving and calling data.

[0057] The technical effects and advantages of the atlas reasoning and analysis system for addressing defects in cell antigen processing proposed in this invention are as follows:

[0058] This invention constructs a closed-loop reasoning system encompassing case data mapping, antigen assembly, presentation requirement determination, individual adjustment, and decision adjustment. This system achieves standardized integration, authenticity discrimination, and individualized correction of defective evidence in antigen processing and presentation pathways. During the assembly stage, the system introduces pattern gating and loading complex integrity rules to perform layered assembly and pre-emptive conflict resolution for direct and cross-presentation. In the decision stage, it generates a set of reachable paths and an observation consistency matrix in parallel, and uses a detection likelihood set to label undetected observations as pending, distinguishing between structural contradictions and detection limitations. This reduces the risk of misinterpreting non-random deletions as pathway breaks from the source, thereby stably identifying HLA-related structural defects and regulatory inhibition differences, improving the interpretability and consistency of defect localization, and reducing the probability of invalid validation and duplicate experiments.

[0059] Based on this, the system outputs defect nodes, defect paths, and evidence chains using the least-explanation path reasoning. A causal graph for verification and intervention is then constructed based on this. This graph integrates intervention costs, reverse measurements of expected improvement in presentation, observational contradiction penalties, and channel risk measurements to search for the least-cost path from the current presentation state to the target state that meets presentation requirements. A list of verification items and an intervention item list are generated, forming clinical decision support outputs. Execution results are backfilled with structured messages carrying data version, algorithm version, graph version, and evidence source identifiers. The accompanying edit logs and evidence chains ensure process traceability and auditability, improving the accuracy and stability of defect path identification while also considering interoperability and reusability in clinical workflows. This reduces unnecessary testing and intervention attempts, improving the timeliness and consistency of case analysis and strategy development. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the structure of a map reasoning and analysis system for addressing defects in cell antigen processing according to the present invention. Detailed Implementation

[0061] 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 embodiments of the present invention, and not all embodiments. 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.

[0062] In order to achieve the above objectives, Figure 1 A schematic diagram of the structure of a graph reasoning and analysis system for addressing defects in cell antigen processing is given in this invention. Specifically, it includes a case data mapping module, an antigen assembly module, a presentation requirement determination module, an individual strategy analysis module, and a decision adjustment module. The modules are connected by signals.

[0063] The case data mapping module is used to obtain the HLA locus typing, somatic cell variation and copy number, transcriptional expression, immunopeptidomics and surface HLA detection results of cases, and to complete the standardized mapping.

[0064] The antigen assembly module is used to assemble a knowledge graph of antigen processing and presentation, including lysis, transport, modification, loading, and display, and distinguishes between direct presentation and cross-presentation.

[0065] The presentation requirement determination module is used to calculate link connectivity and observation consistency on the assembled antigen processing and presentation knowledge graph, and determine whether the presentation requirement is met based on the preset presentation requirement conditions.

[0066] The individual strategy analysis module is used to determine the association between unavailable channels and presentation patterns based on HLA site copy deletion, inactivation mutation and domain destruction when the conditions are not met. It then uses constraint propagation to generate a map editing sequence, prunes nodes and shrinks links to obtain an individualized map.

[0067] The decision adjustment module is used to perform evidence propagation and minimum interpretation path reasoning on the individualized graph by combining the detection likelihood set, outputting defect nodes, defect paths and evidence chains, and constructing a verification and intervention causal graph. It searches for the minimum cost path from the current presentation state to the target state that meets the presentation requirements, and generates clinical decisions for output.

[0068] The case data mapping module is used to obtain HLA locus typing, somatic cell variation and copy number, transcriptomic expression, immunopeptidomics, and surface HLA detection results of cases, and to complete standardized mapping. The specific implementation is as follows:

[0069] HLA locus typing, somatic cell variation and copy number, transcriptional expression, immunopeptidomics and surface HLA detection results are accessed one by one according to the data source, and field standardization, type verification and missing value annotation are performed. Allele nomenclature standardization and heterozygosity consistency verification are performed on HLA locus typing. If typing results from different platforms exist for the same case, conflict mediation is performed according to typing resolution and sequencing batch quality markers. If the mediation conditions are not met, a source audit field is generated and dual-track records are maintained for subsequent judgment.

[0070] Genome coordinate alignment and functional annotation were performed on somatic variation and copy number data. After unifying them to the same reference coordinates, copy deletion markers, inactivation mutation markers, and domain destruction markers were identified and labeled. The transcription expression matrix was normalized based on sequencing depth, gene length, and batch effect. Gene expression status labels were generated by combining expression distribution thresholds and the labels were bound to specific samples, tissue sites, and timestamps to form traceable mapping records.

[0071] Immunopeptidomics data underwent a spectrum identification confidence screening. The results were quality controlled using ion matching coverage, retention time bias, and identification confidence threshold. After removing low-confidence spectrum records, the retained peptides were mapped to presentation sites. The mapping process was constrained by the HLA locus typing of the cases. Candidate presentation sites were first selected by alleles, and then matched by peptide length and binding site characteristics. Peptides that could not meet the matching conditions but had high experimental confidence were marked as pending and the original evidence was retained.

[0072] Surface HLA detection results were numerically normalized based on antibody cloning, instrument gain, and label channel overflow correction parameters. The normalized results were then converted into surface expression tags and linked to corresponding cell populations and detection times. To ensure the usability of multi-source evidence in subsequent spectral inference, sample input volume, pretreatment method, and batch size were summarized as sample-level detection parameters; peptide mass-to-charge ratio range, retention time deviation, fragment ion coverage, identification confidence, and modification type were summarized as peptide-level detection parameters; and antibody cloning, label channel overflow correction, and instrument gain for surface HLA detection were summarized as surface expression detection parameters. These three types of parameters together constitute the detection likelihood set.

[0073] The detection likelihood set is used to describe the probability that each observation is actually detected. When the sample-level detection parameters or peptide-level detection parameters indicate a low detection probability, the module does not treat the non-detection as missing, but marks the node or edge corresponding to the observation as pending and adds a detection likelihood weight.

[0074] The final output consists of two parts. The first part is a tag set, which includes copy deletion markers, inactivation mutation markers, domain destruction markers, gene expression status tags, peptide-to-presentation site mappings, and surface expression tags. All tags have source identifiers, version identifiers, and timestamps, and are associated with specific nodes or edges. The second part is a detection likelihood set, which includes sample-level detection parameters, peptide-level detection parameters, and surface expression detection parameters. These parameters are mapped to observation confidence tags and bound to corresponding nodes or observation records.

[0075] The antigen assembly module is used to assemble a knowledge graph of antigen processing and presentation, including lysis, transport, modification, loading, and display, and distinguishes between direct presentation and cross-presentation. The specific implementation is as follows:

[0076] The system reads a tag set and a detection likelihood set. The tag set includes copy deletion markers, inactivation mutation markers, domain destruction markers, gene expression status tags, peptide-to-presentation site mappings, and surface expression tags. The detection likelihood set includes sample-level detection parameters, peptide-level detection parameters, and surface expression detection parameters.

[0077] The assembly process centers on layered mapping. First, in the direct presentation subgraph, nodes and directed edges are sequentially established for the lysis layer, transport layer, modification layer, loading layer, and display layer. The lysis layer contains proteasome activity nodes and associated substrate nodes; the transport layer contains TAP (antigen processing-associated transporter) transporter nodes; the modification layer contains ERAP (endoplasmic reticulum aminopeptidase) nodes 1 and 2; the loading layer contains loading complex nodes, case-specific HLA class 1 heavy chain nodes, and β2 microglobulin nodes; and the display layer contains surface HLA class 1 complex nodes. Subsequently, in the cross-presentation subgraph, exogenous uptake layers, endosome and lysosomal degradation layers, transmembrane translocation or vacuolar processing layers, loading layers, and display layers are established. The loading layer contains HLA class 2 molecules and pairing-invariant segments, as well as exchange regulatory nodes for HLA-DM and HLA-DO.

[0078] Gene expression status tags are attached to the corresponding coding gene nodes, copy deletion tags, inactivation mutation tags, and domain destruction tags are attached to the corresponding structural nodes, peptide-to-presentation site mappings are attached to the loading and display layers, and surface expression tags are attached to the display layer nodes. Each node generates an initial state field and an evidence citation field. The initial state value is available, pending, or unavailable. The evidence citation field records the tag source, timestamp, and version identifier. The detection likelihood set is used to generate observation confidence tags and write them into the evidence weight fields of edges and nodes. When the sample-level detection parameters or peptide-level detection parameters show a low detection probability, the weight fields of the corresponding nodes and edges are set to pending weights for subsequent observation consistency calculations in the presentation requirement determination module.

[0079] Fine-grained assembly and pre-emptive conflict resolution of bi-subgraphs are achieved through pattern gating and loading complex integrity rules. Pattern gating ensures that evidence only propagates in subgraphs consistent with cell type labels and presentation pattern markers. Directly presented subgraphs and cross-presented subgraphs have a convergence node set at the presentation layer for result verification. If two subgraphs give contradictory conclusions about the same presentation layer node, the module generates a presentation pattern conflict warning and marks the conflict path during assembly, which is then handled by the subsequent individual strategy analysis module.

[0080] When any of the key components of the loading layer—TAPBP (tapasin), calreticulin, ERP57, or β2 microglobulin—is determined to be unusable, the loading edges and display edges connected to that component are immediately marked as unusable. This process is performed during the assembly stage to directly solidify structural defects into graph structural constraints, rather than leaving them for unified judgment in subsequent reasoning stages. The allele existence rule means that once the HLA locus copy deletion marker or inactivation mutation marker of a case is confirmed, the corresponding HLA type I heavy chain node and its link to the display layer are marked as unusable, while not affecting feasible links associated with another allele. The module finally outputs the assembled antigen processing and presentation knowledge graph object, including the node and edge sets of the direct presentation subgraph and the cross-presentation subgraph, the node initial state field, the evidence weight field, the pattern label, and the assembly log.

[0081] The presentation requirement determination module is used to calculate link connectivity and observation consistency on the assembled antigen processing and presentation knowledge graph, and to determine whether the presentation requirement is met based on preset presentation requirement conditions. Specifically, the implementation is as follows:

[0082] The antigen processing and presentation knowledge graph, tag set, and detection likelihood set output by the antigen assembly module are read. The knowledge graph includes direct presentation subgraphs and cross-presentation subgraphs. The tag set includes copy deletion markers, inactivation mutation markers, domain destruction markers, gene expression status tags, peptide-to-presentation site mappings, and surface expression tags. The detection likelihood set includes sample-level detection parameters, peptide-level detection parameters, and surface expression detection parameters.

[0083] First, two traversals are performed on the knowledge graph: one from top to bottom and one from bottom to top. A set of reachable paths is generated in the order of fragmentation, transport, modification, loading, and presentation. Pattern gating is enabled during traversal, extending only within subgraphs that match the cell type label and presentation pattern label. When a node is marked as unavailable by the label set or all upstream edges of the node are determined to be unavailable, the extension is terminated and the reason for the breakpoint is recorded. Each path that completes the transition from the fragmentation layer to the presentation layer is registered as a reachable path, along with the source identifier, timestamp, and version identifier of each node and edge in the path, for subsequent comparison and auditing. Paths that cannot reach the presentation layer are registered as candidate paths and marked with interruption points.

[0084] Then, an observation consistency matrix is ​​constructed to compare the label set with the reachable path set one by one. The comparison rules include four categories:

[0085] The first category is structural consistency, which compares whether the copy missing marker, inactivation mutation marker, and structural domain destruction marker contradict the preservation of the corresponding node in the path. If there is a contradiction, the node and its dependent edges are marked as observed contradictions.

[0086] The second category is functional consistency, which compares gene expression status labels with the functional requirements of path nodes. When the essential components of the loading complex are marked as expression loss, the loading edges directly connected to them are marked as observational contradictions.

[0087] The third category is peptide consistency, which compares the mapping of peptides to presentation sites with the matching relationship between the loading layer and the presentation layer. When the peptide has mapping evidence and the peptide-level detection parameters meet the detection confidence requirements, the corresponding location of the path is marked as observation support. When it is not detected and the sample-level detection parameters or peptide-level detection parameters indicate a low detection probability, it is recorded as pending rather than contradictory, so as to avoid misjudging non-random missing parts as broken strands.

[0088] The fourth category is surface consistency. It compares the surface expression label with the display layer node. When the surface expression label is positive and the surface expression detection parameters meet the detection confidence requirements, it is judged as observation support. When the surface expression label is negative and there is sufficient observation support upstream, it is marked as observation contradiction. The cell values ​​in the matrix are limited to observation support, observation contradiction, and pending, and the corresponding source identifier, timestamp, and version identifier are retained. For cases where the conclusions of the directly presented subgraph and the cross-presented subgraph are inconsistent on the same display layer node, a presentation mode conflict entry is generated in the matrix and recorded.

[0089] Finally, based on the preset presentation requirement conditions, the satisfaction result is output. The presentation requirement conditions include three judgment rules: the path connectivity rule requires that there is at least one reachable path from the cleavage layer to the display layer on each existing HLA allele channel without any observational contradictions; the observation coverage rule requires that each case has at least one entry in the display layer marked by peptide consistency or surface consistency as observational support, and either one is sufficient to determine that there is evidence supporting the display end; the conflict prohibition rule requires that there are no unexplained presentation pattern conflicts or structural consistency contradictions.

[0090] The module substitutes the reachable path set and the observation consistency matrix into the above rules to obtain the satisfaction results and the non-satisfaction list. The non-satisfaction list includes the unavailable channel set, the pending channel set, the presentation mode conflict items, and the breakpoint reasons. These serve as inputs to the individual strategy analysis module for constraining propagation and generating the graph editing sequence. The satisfaction results and support items serve as inputs to the decision adjustment module for establishing the starting point and cost function in the causal graph. The innovation of this module lies in generating the reachable path set and the observation consistency matrix in parallel and labeling them as pending items using the detection likelihood set. This distinguishes between observation contradictions and detection limitations from the source, thereby reducing false negatives for truly feasible paths. At the same time, the pattern gating and conflict item mechanism exposes the discrepancies between direct presentation and cross-presentation in advance during the decision-making stage.

[0091] The individual strategy analysis module, when a condition is deemed unmet, identifies unusable channels and their associated presentation patterns based on HLA site copy deletions, inactivation mutations, and domain disruptions. It then uses constraint propagation to generate a map editing sequence, pruning nodes and shrinking links to obtain an individualized map. The specific implementation is as follows:

[0092] The satisfying results and the non-satisfying list are used as the starting conditions. At the same time, the copy deletion marker, inactivation mutation marker, domain destruction marker, gene expression status label, peptide-to-presentation site mapping, surface expression label and detection likelihood set output by the case data mapping module are read. The detection likelihood set consists of sample-level detection parameters, peptide-level detection parameters and surface expression detection parameters. First, a channel association mapping table is generated. A channel is defined as a set of directed links from the cleavage layer to the display layer with a specific HLA allele as the anchor point and a limited presentation mode. The allele existence rule marks the loading layer and display layer nodes corresponding to alleles with copy deletion marker, inactivation mutation marker or domain destruction marker as unavailable, and its upstream and downstream links are incorporated into the unavailable channel set.

[0093] The loading complex integrity rule requires that all necessary components of the loading complex be available simultaneously; otherwise, the loading edge and display edge connected to it will be included in the unavailable channel set. The pattern gating rule processes the evidence of directly presented subgraphs and cross-presented subgraphs separately, and records the two sets of paths corresponding to the presented pattern conflict entries as candidate edit objects.

[0094] The observation preservation rules identify nodes and edges that have a peptide-to-presentation site mapping and meet the confidence threshold of peptide-level detection parameters as priority observation objects. For cases where no detection is detected and the detection probability is low according to sample-level or peptide-level detection parameters, the module does not include the link in the unavailable channel set, but instead assigns it to the pending channel set and records the pending weight. The channel association mapping table generates the unavailable channel set, pending channel set, presentation mode marker, and evidence source identifier accordingly, providing boundaries for subsequent map editing.

[0095] After obtaining the channel association mapping table, the module uses constraint propagation to generate the graph editing sequence. The variables are the node state and the edge channel state, with values ​​of available, pending, and unavailable; the constraints include node predecessor availability constraints, loading complex integrity constraints, allele existence constraints, presentation pattern consistency constraints, and observation preservation constraints.

[0096] The set of unavailable channels is used as the initial seed to assign the value as unavailable, and the set of pending channels is assigned the value as pending. Constraint propagation is performed on the topological order of the knowledge graph to shrink the feasible domain of variables. When all upstream edges of a node are unavailable or conflict with the integrity constraint of the loading complex, the node is marked as unavailable and propagated downstream. When a presentation mode conflict occurs, the two mode subgraphs are solved separately. The subgraph that is consistent with the observation preservation constraint and has higher observation support is retained, and the corresponding link of the other subgraph is set as unavailable or pending.

[0097] After propagation stabilizes, the minimal cut sets causing inconsistencies are identified. The order of these minimal cut sets follows a priority rule for inconsistency resolution: first, loading and display layer nodes directly triggered by allele presence constraints are deleted; second, essential components of the loading complex are removed; third, mode-specific transport or modification links are removed; and finally, objects marked as pending only due to insufficient detection likelihood are downweighted rather than deleted. The graph editing sequence consists of two types of atomic actions: node pruning removes nodes deemed unusable and not subject to observation preservation constraints; and link contraction removes breakpoint edges caused by mode separation or allele loss from the channels and establishes redirections between substitutable edges at the same level, keeping the remaining channels connected. After executing the complete graph editing sequence, an individualized graph is output, along with an editing log recording the unique identifiers of the edited nodes and links, corresponding allele and presentation mode markers, the source and timestamp of triggering evidence, the type and execution order of the editing action, and the state before and after editing. This log serves as the starting point for causal searches in the audit and decision adjustment modules.

[0098] In summary, by using channels as the basic editing unit and introducing two layers of protection—undetermined weight reduction and observation preservation—structural defects are rigidly eliminated while detection limitations are flexibly handled. This avoids misjudging non-random missing data as structural fractures, ultimately resulting in an individualized atlas consistent with case molecular evidence and usable for intervention design.

[0099] The decision adjustment module is used to perform evidence propagation and minimum interpretation path reasoning on a personalized graph by combining the detection likelihood set. It outputs defective nodes, defective paths, and evidence chains, and constructs a verification and intervention causal graph. It searches for the minimum cost path from the current presentation state to the target state that meets the presentation requirements, and generates clinical decisions for output. Specifically, the implementation is as follows:

[0100] The module receives individualized graphs and detection likelihood sets as inputs and first performs evidence propagation on the individualized graphs: the mapping from peptides to presentation sites and surface expression labels are used as observation sources, and the observation credibility labels corresponding to the detection likelihoods are written into the evidence weight fields of nodes and edges; when an observation comes from sample-level detection parameters or peptide-level detection parameters indicating a low detection probability, the module treats the observation as pending and participates in propagation with reduced weight.

[0101] The propagation proceeds in the order of fragmentation, transport, modification, loading, and display. The interpretation score of a node is composed of three parts: the weighted support of upstream observational evidence directly connected to it, whether a downstream link consistent with the observation can be formed, and the structural penalty imposed by copy deletion markers, inactivation mutation markers, and domain destruction markers related to the node.

[0102] Using the explanatory score as the criterion, the minimum explanatory path reasoning is employed to select the set of paths that satisfy observation consistency among all reachable paths, prioritizing paths that can explain the same observation with fewer nodes and fewer conflicts. This yields defective nodes, defective paths, and chains of evidence. Defective nodes are those whose explanatory scores are significantly biased towards anomalies due to structural penalties and downstream inconsistencies. Defective paths are those containing one or more defective nodes and having contradictory entries with the observation consistency matrix. The chain of evidence is an ordered record of the observation source, source identifier, timestamp, and version identifier supporting each conclusion, used for auditing and verification.

[0103] After obtaining the defective nodes and defective paths, the module constructs a verification and intervention causal graph and searches for the minimum cost path from the current presentation state to the target state that meets the presentation requirements. The nodes of the causal graph consist of interventionable points and observed outputs, and the edges represent regulatory or functional dependencies confirmed by the individualized graph. The module sets a cost structure for each intervention action, including intervention cost, a reverse metric of the expected improvement in presentation, a penalty related to inconsistent observations, and a risk metric for potential adverse effects on other channels. All four metrics are derived from a comprehensive quantification of the individualized graph, the set of reachable paths, and the observation consistency matrix.

[0104] Heuristic strategies are employed during the search process:

[0105] Prioritize expanding intervention nodes that can simultaneously reduce the number of observational contradictions and add observational support items to the presentation layer, and assign lower priority to branches that only depend on pending observations; when there are presentation pattern conflicts, the module evaluates the path cost on the two pattern subgraphs respectively and retains the one with the smaller total cost. The final output includes the minimum cost path and its corresponding evidence chain, a list of defective nodes and a set of defective paths, reversibility labels, a list of validation items and a list of intervention items; at the same time, it generates structured messages for clinical decision support output. The structured messages include data version, algorithm version and graph version identifiers, evidence source identifiers and timestamps, as well as message headers and field mapping relationships for medical information system integration.

[0106] In summary, this module internalizes the detection likelihood set into propagation weights and conflict penalties, enabling flexible handling of non-random missing values ​​during the decision-making stage. It first completes the attribution of anomalies using the least explanatory path reasoning, and then completes the path search for executable interventions on the causal graph using a cost function, thereby achieving the integration of explanation and intervention and forming a closed loop with the individualized graph and the presentation requirement determination module.

[0107] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0108] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein 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 implementation should not be considered beyond the scope of this application.

[0109] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0110] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0111] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cell antigen processing deficiency oriented mapping inference analysis system, characterized in that: The system comprises a case data mapping module, an antigen assembly module, a presentation demand determination module, an individual strategy analysis module, and a decision adjustment module, and the modules are connected through signals; The case data mapping module is used for obtaining HLA locus typing, somatic variation and copy number, transcription expression, immunopeptidomics and surface HLA detection results of a case, and completing standardized mapping; The antigen assembly module is used for assembling an antigen processing and presentation knowledge graph, including cleavage, transport, modification, loading, display, and distinguishing direct presentation and cross-presentation; The presentation demand determination module is used for calculating link connectivity and observation consistency on the assembled antigen processing and presentation knowledge graph, and determining whether the presentation demand is met according to a preset presentation demand condition; The individual strategy analysis module is used for determining unavailable channels and presentation mode association according to HLA locus copy loss, inactivation mutation and domain damage when it is determined that the presentation demand is not met, generating a graph editing sequence by using constraint propagation, trimming nodes and shrinking links to obtain an individualized graph; The decision adjustment module is used for performing evidence propagation and minimum explanation path reasoning on the individualized graph in combination with a detection likelihood set, outputting defect nodes, defect paths and evidence chains, constructing a verification and intervention causal graph, searching for a minimum cost path from a current presentation state to a target state meeting the presentation demand, and generating a clinical decision for output; The presentation demand determination module comprises: calculating a set of reachable paths from input to output on the assembled antigen processing and presentation knowledge graph in the order of cleavage, transport, modification, loading and display; The labels output by the case data mapping module are compared with corresponding nodes to form an observation consistency matrix; The set of reachable paths and the observation consistency matrix are jointly determined according to a preset presentation demand condition, and a satisfaction result is output; The detection likelihood set is established based on sample input quantity, peptide segment type and experimental process parameters, and the detection likelihood set and the labels are output together; The detection likelihood set established by the case data mapping module comprises: Sample-level detection parameters are generated according to sample input quantity, pretreatment method and on-machine batch; Peptide segment-level detection parameters are generated according to peptide segment mass-to-charge ratio range, retention time deviation, fragment ion coverage, identification confidence and modification type; Surface expression detection parameters are generated according to antibody clones of surface HLA detection, label channel overflow correction and instrument gain; Sample-level detection parameters, peptide segment-level detection parameters and surface expression detection parameters are mapped into observation confidence labels and bound to corresponding nodes or observation records.

2. The system for mapping and reasoning for cell antigen processing defects according to claim 1, wherein: The case data mapping module comprises: Field specification and alias checking are performed on HLA locus typing, somatic variation and copy number, transcription expression, immunopeptidomics and surface HLA detection results based on a unified time reference and a term table; Allele naming standardization and heterozygosity consistency checking are performed on HLA locus typing; Genomic coordinate alignment and functional annotation are performed on somatic variation and copy number, and copy loss, inactivation mutation and domain damage markers are extracted; Gene expression state labels are generated for transcription expression; The confidence screening of the spectrum identification of immunopeptidomics and the mapping of the peptide segment to the presentation site are completed, the numerical normalization of the surface HLA detection is performed, and the surface expression tag is generated.

3. The system for mapping and reasoning for cell antigen processing defects according to claim 2, wherein: The content of the individual strategy analysis module when determining that the presentation requirement is not met includes: Extracting HLA site copy loss, inactivation mutation, and domain damage markers from case data mapping results; Projecting the markers to the nodes and links associated with the alleles, determining the set of unavailable channels, and establishing a channel association mapping table; Generating a spectrum editing sequence using constraint propagation based on the channel association mapping table, performing pruning on associated nodes, and performing contraction on dependent links to obtain an individualized graph.

4. The system for mapping and reasoning for cell antigen processing defects according to claim 3, wherein: Generating a spectrum editing sequence using constraint propagation includes: Constructing a constraint satisfaction network with node states and edge channel states as variables; Injecting node predecessor availability, loading complex integrity, and allele existence as constraint conditions into the network; Performing constraint propagation to reduce the feasible region of variables and identify the minimal cut set that causes inconsistency; According to the topological order and inconsistency resolution priority rules, the editing actions are sorted, and the spectrum editing sequence is output.

5. The system for mapping and reasoning for cell antigen processing defects according to claim 4, wherein: The decision adjustment module includes: Performing evidence propagation on the individualized graph in combination with the detection likelihood set and calculating the explanation score of each node and each path; Building a verification and intervention causal graph based on the intervention points and observation outputs, with nodes representing intervention points and observation outputs, and edges representing regulation or functional dependence; Constructing a path cost function based on intervention cost, expected presentation improvement amplitude, and evidence inconsistency penalty; Using heuristic search to search for the minimum cost path from the current presentation state to the target state that meets the presentation requirement in the causal graph; Output the minimum cost path and the corresponding evidence chain, and generate a clinical decision support output.

6. The system as claimed in claim 1, wherein the system is directed to antigen processing deficiency of cells. The antigen assembly module, when distinguishing between direct presentation and cross-presentation, performs the following: According to the cell type tag and the presentation mode marker, evidence is propagated only within the subgraph matching it; Using display layer nodes as convergence nodes, the results from the direct presentation subgraph and the cross-presentation subgraph are checked for consistency; When the check is inconsistent, mark the conflict path and output a presentation mode conflict prompt for graph editing by the individual strategy analysis module.

7. The system as claimed in claim 3, wherein the system is directed to antigen processing defects in cells. The editing log includes: Unique identification of edited nodes and links, corresponding alleles and presentation mode markers; Evidence source, timestamp, and version identification that triggered the edit; Editing action type and execution order, node state and link state before and after editing; Rollback markers and traceable checksums for recovery.

8. The cell antigen processing deficiency mapping and inference analysis system of claim 5, wherein: The structured message includes: Defect node list, defect path set, and corresponding evidence chain; Reversibility tag, verification item list, and intervention item list; Data version, algorithm version, and graph version identification, evidence source identification, and timestamp; Message header and field mapping relationship for medical information system interface, and provides backfill and audit interface to record receiving and calling status.

Images