Risk prediction method and system for new pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination, and storage medium
By employing a risk prediction method that combines pharyngeal microecology and immunity, and utilizing biomic and immunological data, core features are screened and identified. Combined with L1/L2 logistic regression and cross-validation, this approach addresses the shortcomings of existing technologies in identifying the risk of novel pneumonia and achieves accurate and interpretable prediction results.
Patent Information
- Application Number
- CN202511691544.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-18
- Publication Date
- 2026-02-06
AI Technical Summary
Existing technologies lack dedicated models for preoperative risk identification of new pneumonia after hematopoietic stem cell transplantation, and fail to adequately characterize the nonlinearity and feature interactions of preoperative multidimensional data, resulting in poor interpretability and a tendency to overfit.
A risk prediction method based on pharyngeal microecology and immunity was adopted. By acquiring biomic and immunological data, data preprocessing and second-order interaction term expansion were performed. L1 sparse constraint logistic regression was used to screen core features. Combined with L1/L2 hybrid ensemble logistic regression and five-fold cross-validation, a fixed feature locking and leakage prevention framework was established, and an interpretable module was generated to output the predicted risk.
It enables accurate risk prediction of new-onset pneumonia after hematopoietic stem cell transplantation, improves the interpretability and stability of the model, reduces the risk of information leakage, and provides repeatability and auditability.
Smart Images
Figure CN121483618A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of biomedical information processing, and in particular to a pharyngeal microecology-immune combined hematopoietic stem cell transplantation post-new pneumonia risk prediction method, system and storage medium. BACKGROUND
[0002] Pulmonary infection is the most common and highly threatening complication in patients with hematological diseases receiving allogeneic hematopoietic stem cell transplantation, significantly increasing non-relapse mortality and hospitalization burden. How to achieve accurate risk prediction at an early stage has always been a core problem in clinical and scientific research. The existing method lacks a preoperative special model for post-transplant new pneumonia in preoperative risk identification and is insufficient in describing the nonlinearity and feature interaction of preoperative multidimensional data and unstable feature selection, resulting in poor interpretability and easy overfitting. SUMMARY
[0003] The present application overcomes the shortcomings of the prior art and provides a pharyngeal microecology-immune combined hematopoietic stem cell transplantation post-new pneumonia risk prediction method, system and storage medium. The present application realizes the processing of prediction data of post-transplant new pneumonia and outputs the prediction risk through a new pneumonia risk prediction model by fusing various detection parameters.
[0004] In one preferred embodiment of the present application, a pharyngeal microecology-immune combined hematopoietic stem cell transplantation post-new pneumonia risk prediction method comprises: Obtaining sampling data, the sampling data comprising biological group data and immunological data; Performing risk assessment and solving according to the sampling data through a pre-trained interpretable prediction model of new pneumonia to obtain new pneumonia risk prediction; The establishment of the interpretable prediction model of new pneumonia comprises: Step S1, data preprocessing of sample data to obtain standardized sample data; Step S2, dividing the sample data into a training set and a test set using five-fold cross-validation; Step S3, processing the standardized sample data based on a two-stage "fixed feature locking" anti-leakage framework to lock the core feature list and the corresponding binomial interaction structure to obtain a fixed feature locking anti-leakage framework; Step S4, performing probability mean fusion based on the training set through the fixed feature locking anti-leakage framework and L1 / L2 mixed ensemble logistic regression to obtain new pneumonia risk prediction.
[0005] In one preferred embodiment of the present application, the two-stage "fixed feature locking" anti-leakage framework processing comprises the following steps: The L1 sparse constraint logic regression is used on the standardized sample data to expand the second-order interaction term feature, and a fixed number of features are screened out as core feature lists and frozen, and the core feature list and the corresponding binomial interaction structure are locked.
[0006] In a preferred solution of the present application, based on the obtained fixed feature locking anti-leakage framework, the standardized sample data is separated into a training set and a test set through five-fold cross-validation hierarchical cross-validation; the model is established through training set training, and the training effect of the model is optimized through test set.
[0007] In a preferred solution of the present application, the establishment of the interpretable prediction model of the new pneumonia also includes: generating a global explanation based on the obtained interpretable module for new pneumonia risk prediction, and obtaining interpretable prediction data.
[0008] In a preferred solution of the present application, the interpretable module includes a SHAP model interpretable framework.
[0009] In a preferred solution of the present application, the data preprocessing in step S1 includes: data standardization processing and second-order interaction term expansion to obtain feature data, and the obtained feature data is inhibited by L1 sparsification to suppress dimension expansion. The feature data includes microbiome features and immunology features.
[0010] In a preferred solution of the present application, the biological group features include pharyngeal microecological diversity features, C-reactive protein features, and preoperative pneumonia state features; and the immunology features include immunophenotype features.
[0011] In a preferred solution of the present application, a device for predicting the risk of new pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immune combination includes: A data acquisition module is configured to acquire sampling data, wherein the sampling data includes biological group data and immunology data. A data processing and analysis module is configured to perform risk assessment and solving based on the sampling data by using a pre-trained interpretable prediction model of new pneumonia, and obtain new pneumonia risk prediction. Steps for implementing a method for predicting the risk of new pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immune combination.
[0012] In a preferred solution of the present application, a system for predicting the risk of new pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immune combination includes a memory configured to store computer programs / instructions, and a processor configured to execute the computer programs / instructions to implement steps of a method for predicting the risk of new pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immune combination.
[0013] In a preferred embodiment of the present invention, a storage medium for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity is provided to implement the steps of a method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity.
[0014] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: This invention discloses a method, system, and storage medium for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on the combination of pharyngeal microecology and immunity; it realizes the processing of prediction data for new-onset pneumonia after transplantation, integrates multiple detection parameters, and outputs the predicted risk through a new-onset pneumonia risk prediction model. Attached Figure Description
[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0016] Figure 1 This is a flowchart of the model construction process in a preferred embodiment of the present invention; Figure 2 This is a bar chart of the first 15 fixed features in a preferred embodiment of the present invention; Figure 3 This is a global importance graph of SHAP in a preferred embodiment of the present invention; Figure 4 This is an analysis chart of the five-fold ROC curve and the average ROC (including the mean AUC ± SD) in a preferred embodiment of the present invention. Detailed Implementation
[0017] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.
[0018] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0019] Example 1, as Figures 1-3 As shown, a method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination includes: Acquire sampling data, which includes: biomic data and immunological data; Based on the sampled data, a risk assessment is performed using a pre-trained interpretable prediction model for emerging pneumonia to obtain a risk prediction for emerging pneumonia.
[0020] Specifically, the establishment of an interpretable predictive model for emerging pneumonia includes: Step S1: Perform data preprocessing on the sample data to obtain standardized sample data.
[0021] Specifically, the data preprocessing in step S1 includes: standardizing the data and expanding the second-order interaction terms (second-order polynomial and interaction term expansion) to obtain feature data, and then using L1 sparsification to suppress dimensionality expansion of the obtained feature data; the feature data includes microbiome features and immunological features. Further, the microbiome features include pharyngeal microecological diversity features, C-reactive protein features, and preoperative pneumonia status features; the immunological features include immunophenotypic features.
[0022] Step S2: Fivefold cross-validation (FivefoldCV) is used to divide the sample data into training and test sets; Step S3 involves a two-stage "fixed feature locking" leakage prevention framework processing based on standardized sample data. This process locks the core feature list and its corresponding binomial interaction structure, resulting in the fixed feature locking leakage prevention framework. Further, based on this framework, the standardized sample data is hierarchically cross-validated using five-fold cross-validation to separate the training and test sets. The model is trained using the training set, and its training effectiveness is validated using the test set.
[0023] Furthermore, the two-stage "fixed feature locking" leakage prevention framework process includes the following steps: On standardized sample data, logistic regression with L1 sparse constraints is applied to the entire dataset to screen out a fixed list of top features for second-order interaction terms, which is then frozen and locked to ensure the core feature list and its corresponding binomial interaction structure. This achieves the extraction and locking of core features, resulting in a fixed feature locking framework to prevent leakage. Furthermore, unified column order and end-to-end consistency are ensured; the "fixed feature list + mapping relationship + column order" is reused as metadata throughout the entire process. Standardization is only used for training iteration estimation and applied within the current iteration. This ensures reproducibility and traceability across different environments / batches, reducing engineering and compliance risks. Even further, only the core feature list and binomial / interaction structure are locked on the entire dataset (without passing any parameters), and then preprocessing is performed using only the training set in hierarchical cross-validation, continuing the fixed structure. This prevents information leakage from the source and ensures the consistency and auditability of feature column order and processing flow during validation and training.
[0024] Step S4 involves using a fixed-feature-locked anti-leakage framework and an L1 / L2 hybrid ensemble logistic regression to perform probability mean fusion based on the training set, thus obtaining a prediction of the risk of novel coronavirus pneumonia. Further, a lightweight ensemble consisting of L1 and L2 logistic regressions is trained at each training fold, and probability mean fusion is performed on the validation output. This balances the sparse interpretability of L1 with the stable generalization of L2, improving robustness and deployment availability.
[0025] Example 2, based on Example 1, such as Figures 1-4 As shown, the establishment of an interpretable prediction model for emerging pneumonia also includes: generating a global interpretation based on the acquired interpretability module for emerging pneumonia risk prediction, thus obtaining interpretable prediction data. Specifically, the interpretability module includes the SHAP model interpretability framework. Furthermore, in data preprocessing, the data needs to be anonymized, and the data source must comply with regulatory requirements.
[0026] Example 3, building upon Example 1 or Example 2, demonstrates a core feature list obtained through "fixed feature locking" and its sources (original, derived binomial, or interaction terms), providing a unified column order and traceable feature mapping relationships. This list is reused uniformly in cross-validation, final training, and online deployment, ensuring no leakage, reproducibility, and auditability. ROC / AUC for each fold is obtained based on hierarchical cross-validation, and the average ROC is calculated. During the evaluation process, class imbalance robustness and consistency with the processing flow are maintained, ensuring consistency between offline evaluation and online deployment.
[0027] Specifically, features are extracted and locked to obtain a fixed feature locking and leakage prevention framework. Based on this framework, L1 sparse screening is used to suppress dimensionality expansion. The validation output is then fused using probability mean fusion via L1 / L2 hybrid integrated logistic regression. Furthermore, this invention highlights the upper respiratory tract (pharynx) microbiota as a core source of preoperative, repeatable information closely related to lung infection risk, using "pharyngeal microecological diversity" as the starting point to achieve proactive risk identification and intervention. Further, the second-order interaction term expansion uses L1 sparse constraint-based logistic regression on the entire dataset to screen out the top 15 fixed features and freeze them, locking only the "feature list and binomial interaction structure." Subsequently, in each fold cross-validation, only the training set is used to fit the normalizer, and isomorphic processing is performed according to the fixed structure to strictly avoid information leakage. Furthermore, the L1 sparse screening explicitly introduces the nonlinearity and interaction effects of clinical, microecological, and immune features, while simultaneously suppressing dimensionality expansion through L1 sparsification. While maintaining linear interpretability, the numerical features are extended using binomial and interaction methods, and sparse selection is used to suppress dimensionality expansion; nonlinearity and synergistic effects are captured, while maintaining a concise and reproducible model structure. Furthermore, L1 / L2 hybrid ensemble logistic regression trains multiple base learners per fold, half L1 and half L2, with hyperparameters randomly sampled and the validation outputs integrated using probability mean, balancing sparse interpretability and stable generalization. Further, the Shannon index is a continuous indicator measuring community diversity; this invention also uses its high / low grouping (ShannonGroup). CRP: C-reactive protein, a commonly used inflammatory marker. Second-order polynomial features: containing first- to second-order terms of the original features and their interaction terms. L1 / L2 regularization: imposing sparsity / smoothing constraints on coefficients; L1 aids feature selection, while L2 improves stability. Hierarchical K-fold cross-validation: a K-fold evaluation method that maintains consistent class distribution. Data leakage: improper use of validation / test information during evaluation leads to performance overestimation. AUC / ROC: performance metrics and curves for binary classification. SHAP: Feature Contribution Interpretation Method. Isomorphic Processing: Based on a fixed list of features and mapping relationships, a consistent feature generation / column extraction / sorting / standardization process is performed on any data. Fixed Feature Locking: Used only to structurally freeze the core feature set and mapping relationships on the entire dataset, and reused uniformly in subsequent stages (without passing parameters). Hybrid Ensemble Logistic Regression: A lightweight ensemble consisting of L1 and L2 logistic regression base learners, fusing their predicted probabilities to improve robustness.
[0028] Furthermore, the upper respiratory tract (pharyngeal) microecology serves as a core information source and a deployable entry point: community diversity indicators obtainable from pharyngeal swabs are incorporated into preoperative standardized features and prioritized as key features in fixed feature locking, forming crucial interactions with immunity / inflammation. This enables non-invasive, low-cost, and standardized preoperative data collection and model deployment, transforming "pharyngeal microecological diversity" into interpretable and actionable risk factors, directly targeting intervention pathways such as pharyngeal microecological management.
[0029] Example 4: A device for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology and immune system integration, comprising: The data acquisition module is used to acquire sampled data, which includes: biomic data and immunological data. The data processing and analysis module performs risk assessment and solves the problem using a pre-trained interpretable prediction model for emerging pneumonia based on the sampled data, thereby obtaining a risk prediction for emerging pneumonia. Steps for implementing the method for predicting the risk of new pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination in Example 1.
[0030] Example 5: A risk prediction system for new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination, comprising a memory for storing computer programs / instructions; and a processor for executing the computer programs / instructions to implement the steps of the risk prediction method for new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination in Example 1.
[0031] Example 6: A storage medium for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination, used to implement the steps of the method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination in Example 1.
[0032] Working principle: A method, system, and storage medium for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination; the method enables the processing of predictive data on new-onset pneumonia after transplantation, and integrates multiple detection parameters to output the predicted risk through a new-onset pneumonia risk prediction model.
[0033] like Figure 1As shown, this invention provides data on pharyngeal microecological diversity (Shannon continuous / grouped), C-reactive protein (CRP), preoperative pneumonia status, and immunity. After data processing and standardization, second-order interaction terms are extended to the numerical features. Subsequently, a two-stage "fixed feature locking" leakage prevention framework is adopted: first, a set of core features and corresponding binomial / interaction structures are locked on the entire dataset (only the structure is locked, no parameters are passed); then, in hierarchical cross-validation, preprocessing is performed only on the training set, and isomorphic processing is carried out using the fixed structure; each fold trains a hybrid ensemble of L1 / L2 logistic regression and performs probability mean ensemble on the validation results; the average ROC and performance indicators are summarized; and the final model is trained on the entire dataset based on fixed features, and a global interpretation is generated using interpretable methods (such as SHAP). Figure 2 (Fixed Feature Set) Displays the list of core features obtained through "Fixed Feature Locking" and their sources (original, derived binomial or interaction terms), and provides a unified column order and traceable feature mapping relationship; this list is reused uniformly in cross-validation, final training and online deployment to ensure that it is not leaked, reproducible and auditable. Figure 3 (Interpretability diagram) The final model adopts an interpretation method based on contribution value (such as SHAP) to provide an explanation of global importance; at the same time, it outputs the 15 most important influencing factors to support the decision explanation. Figure 4 (ROC and Evaluation) The ROC / AUC of each fold is obtained based on hierarchical cross-validation, and the average ROC is calculated; the class imbalance robustness and processing flow are kept consistent during the evaluation process to ensure consistency between offline evaluation and online deployment; class imbalance robustness and consistency are guaranteed.
[0034] This invention integrates multiple detection parameters (including pharyngeal microecological diversity, C-reactive protein, preoperative pneumonia status, and immunophenotype) and outputs a predicted risk through a novel pneumonia risk prediction model. The novel pneumonia risk prediction model includes: obtaining feature parameters from sampled data through robust feature engineering based on interactive multinomial expansion and leakage prevention; performing data analysis based on these feature parameters combined with hierarchical cross-validation, class-imbalanced robust strategies, and regularized lightweight logistic regression; and outputting global and individual-level evidence based on the data analysis through an interpretability module, using this evidence as the predicted risk.
[0035] This invention overcomes the problems of existing methods, such as single data source, insufficient characterization of nonlinearity and feature interaction, unrobust feature selection, and lack of interpretability. It achieves predictive data processing for post-transplant pneumonia based solely on preoperative data and provides a reference for oropharyngeal microecological management and imaging assessment.
[0036] Based on the preferred embodiments of the present invention, and through the above description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.
Claims
1. A method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination, characterized in that, include: Acquire sampling data, which includes: biomic data and immunological data; Based on the sampled data, a risk assessment is performed using a pre-trained interpretable prediction model for emerging pneumonia to obtain a risk prediction for emerging pneumonia. The establishment of the interpretable predictive model for the newly emerging pneumonia includes: Step S1: Perform data preprocessing on the sample data to obtain standardized sample data; Step S2: Use five-fold cross-validation to divide the sample data into a training set and a test set; Step S3: Based on standardized sample data, perform a two-stage "fixed feature locking" leakage prevention framework process to lock the core feature list and the corresponding binomial interaction structure to obtain the fixed feature locking leakage prevention framework. Step S4: Based on the training set, a probability mean fusion is performed using a fixed feature locking anti-leakage framework and L1 / L2 hybrid ensemble logistic regression to obtain a prediction of the risk of novel coronavirus pneumonia.
2. The method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination as described in claim 1, characterized in that: The two-stage "fixed feature locking" leakage prevention framework process includes the following steps: On standardized sample data, logistic regression with L1 sparse constraints is applied to the entire data to screen out the top few fixed features as the core feature list and freeze them. The core feature list and the corresponding binomial interaction structure are locked. The extraction and locking of core features are realized, and a fixed feature locking and leakage prevention framework is obtained.
3. The method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination as described in claim 2, characterized in that: Based on the acquired fixed features, a leakage prevention framework is established. The standardized sample data is hierarchically cross-validated in a five-fold cross-validation process to separate the training set and the test set. The model is trained and built using the training set, and the training effect of the model is optimized and verified using the test set.
4. The method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination as described in claim 1, characterized in that: The establishment of an interpretable prediction model for emerging pneumonia also includes: generating a global interpretation based on the interpretability module of the acquired emerging pneumonia risk prediction to obtain interpretable prediction data.
5. The method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination as described in claim 4, characterized in that: The interpretability module includes the SHAP model interpretability framework.
6. The method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination as described in claim 3, characterized in that: The data preprocessing in step S1 includes: data standardization and second-order interaction term expansion to obtain feature data, and L1 sparsification to suppress dimensionality expansion of the obtained feature data; The characteristic data includes microbiome characteristics and immunological characteristics.
7. The method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination as described in claim 6, characterized in that: Biological characteristics include pharyngeal microecological diversity, C-reactive protein characteristics, and preoperative pneumonia status characteristics; Immunological characteristics include immunophenotypic characteristics.
8. A device for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination, characterized in that, include: The data acquisition module is used to acquire sampled data, which includes: biomic data and immunological data. The data processing and analysis module performs risk assessment and solves the problem using a pre-trained interpretable prediction model for emerging pneumonia based on the sampled data, thereby obtaining a risk prediction for emerging pneumonia. The steps are for implementing the method for predicting the risk of new pneumonia after hematopoietic stem cell transplantation based on the combination of pharyngeal microecology and immunity as described in any one of claims 1-7.
9. A system for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology and immune system integration, characterized in that, Memory, used to store computer programs / instructions; A processor for executing the computer program / instructions to implement the steps of the method for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on the pharyngeal microecology-immunity combination as described in any one of claims 1-7.
10. A storage medium for predicting the risk of new-onset pneumonia after hematopoietic stem cell transplantation based on pharyngeal microecology-immunity combination, characterized in that, The steps are for implementing the method for predicting the risk of new pneumonia after hematopoietic stem cell transplantation based on the combination of pharyngeal microecology and immunity as described in any one of claims 1-7.