Dynamic quantile standardization method and system for cross-platform heterogeneous data and application of dynamic quantile standardization method and system

Abstract

The invention belongs to the technical field of bioinformatics and computational biology, and particularly relates to a dynamic quantile standardization method and system for cross-platform heterogeneous data and application of the dynamic quantile standardization method and system. Through a gene expression scoring algorithm, the method can effectively integrate RNA expression data of different platforms, batches and formats. Gene expression values are sorted and scored, and batch effects and background differences among different data sets are eliminated, so that more accurate and reliable integrated data are generated. The method is simple, convenient and high in adaptability, can process multiple types of gene expression data, does not need to depend on complex model hypothesis or deep learning technology, and remarkably improves the accuracy, reliability and universality of data analysis.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of bioinformatics and computational biology, and particularly relates to a dynamic quantile normalization method and system for cross-platform heterogeneous data, and application thereof. BACKGROUND

[0002] Existing RNA expression profile integration algorithms mainly include several common strategies: 1. Expression-based integration method This method relies on gene expression data of RNA expression profile data, and realizes the integration of data by calculating the change of gene expression in different data sets. For example, DESeq2, edgeR and other algorithms are widely used in differential expression analysis, and the extended version of these methods can be used for integrating different data sets.

[0003] 2. Model-based integration method This method integrates RNA expression profile data by constructing a statistical model. Specifically, the algorithm uses Bayesian models, mixed effect models or other multivariate statistical methods to combine the characteristics of different data sets for optimized integration. Common algorithms such as limma and ComBat can effectively control batch effects and technical errors.

[0004] 3. Deep learning method This method uses neural networks to learn the complex relationships between different data sets. These methods can automatically extract features from data, efficiently integrate, and compare between multiple data sets.

[0005] However, integrating public RNA expression profile data from different studies still faces significant challenges, mainly due to the interwoven effects of detection platform differences, batch effects and biological heterogeneity. Systematic biases exist in sequencing platforms used by different laboratories (such as Illumina HiSeq vs. NovaSeq), library preparation methods (such as polyA enrichment vs. rRNA removal), and sequencing depth, making it impossible to directly compare the absolute levels and detection sensitivity of gene expression. More importantly, batch effects - technical noise introduced by non-biological factors such as experimental time, operator, reagent batch, etc. - can severely distort the expression profile, making it difficult to completely distinguish and correct. In addition, the lack of standardized gene annotation of samples in public databases further increases the difficulty of data normalization and biological interpretation. Therefore, directly integrating raw public data without strict batch correction, standardization and heterogeneity assessment may lead to incorrect conclusions. The following technical problems exist: 1. Addressing batch effect issues: Existing expression-based integration methods (such as DESeq2, edgeR) usually rely on differential analysis of raw gene expression data, but these methods are susceptible to technical batch effects and sample source differences, which can lead to misinterpretation of expression differences between different datasets as biological differences, thereby affecting the accuracy of the analysis results. The gene expression scoring algorithm of the present invention eliminates the background differences between different datasets by independently sorting and scoring the samples in each dataset, effectively avoiding the influence of batch effects, thereby improving the accuracy of data integration.

[0006] 2. Reducing heterogeneity between datasets: Traditional model-based integration methods (such as limma, ComBat, etc.) perform well in controlling batch effects, but rely on hypothetical models (such as Bayesian models, mixed effect models), which may not fully meet the distribution of actual data, thereby affecting the accuracy of the model. In addition, such methods have high computational complexity and may be limited by data size and structure. The present invention converts gene expression data into relative expression values through simple sorting and scoring methods, effectively integrating different datasets without complex assumptions and model assumptions, significantly reducing the heterogeneity between datasets.

[0007] 3. Improve data availability and wide applicability: Existing deep learning methods (such as neural network-based integration methods) can extract potential features from complex data, but these methods usually require a large amount of labeled data and computing resources, and their application is still in the optimization stage. In addition, deep learning methods have poor interpretability and are not easy to understand the internal processing of the model. The algorithm of the present invention generates relative expression values through simple sorting and scoring of gene expression values, avoiding the "black box" problem in deep learning methods, and the calculation process is simple and easy to understand, with high operability, suitable for various researchers to integrate data.

[0008] 4. Support different types of gene expression data: Existing expression-based integration methods usually require data to have consistent processing methods (such as standardization, log conversion, etc.), otherwise data integration may appear inconsistent or biased. Unlike this, the present invention method allows multiple types of gene expression data input, including raw data, standardized data, log converted data, etc., making the algorithm more adaptable to different types of data and not being affected by the inconsistency of data preprocessing.

[0009] 5. Suitable for large-scale data sets: The method of the present application can handle a large amount of gene expression data through the ordering and scoring of gene expression, without significantly affecting the size or complexity of the data set as traditional algorithms do. In particular, in the training and validation of tumor diagnosis models, large-scale case data is crucial to improve the accuracy and generalization ability of the model, and the algorithm of the present application can provide efficient data integration to support the analysis of large-scale data sets. SUMMARY

[0010] The present application is directed to a computer software tool (R package) for integrating and analyzing transcriptome sequencing (RNA expression profile) data, aiming to solve the key technical problems of technical bias, batch effect and biological heterogeneity in the integration of cross-platform and cross-study RNA expression profile data sets. A dynamic gene expression standardization algorithm is proposed, which can effectively integrate RNA expression profile data from different sources.

[0011] By ordering and scoring the gene expression values, the batch effect and background difference between different data sets are eliminated, thereby generating more accurate and reliable integrated data. This method is simple, adaptable, can process different types of gene expression data, and does not rely on complex model assumptions or deep learning techniques, thereby improving the accuracy, reliability and applicability of data analysis.

[0012] The technical scheme adopted by the present application is: In a first aspect, the present application provides a dynamic quantile standardization method for cross-platform heterogeneous data, comprising: a) Collect gene expression matrices (data sets) from different detection platforms or experimental batches, with the number of samples in each data set being greater than 10, and any data format can be used; b) In each data set, sort all expression values of each gene, and divide them into 2-50 variable quantile intervals by a dynamic quantile algorithm; c) In each data set, for each gene of each sample, map the variable quantile interval to a uniformly distributed discrete level value to obtain a cross-platform comparable relative expression matrix; d) According to the state information of the samples (such as diagnosis information, survival time information, etc.), assist in the analysis process and construct the corresponding analysis model. For example, by inputting diagnosis information, a diagnosis model for disease classification can be constructed; when inputting survival time information, the survival rate of different samples can be compared and predicted through a survival analysis model (such as survival curve), further improving the interpretation and application value of cross-platform data; e) When integrating datasets, the pan-gene set present in any proportion (range 0-100%) of datasets can be specified, and if any dataset is missing any number of pan-genes, the average relative expression value is automatically imputed; f) The output is a standardized matrix that can be directly used for machine learning modeling or clinical decision support.

[0013] Preferably, the detection platform comprises at least one of microarray, RNA sequencing, digital PCR, real-time quantitative PCR, single-cell RNA sequencing, Nanostring technology, in situ hybridization, and gene chip.

[0014] Preferably, the data format comprises, but is not limited to, Count (count data), TPM (Transcripts Per Million), CPM (Counts Per Million), FPKM (Fragments Per Kilobase of transcript per Million mapped reads), RPKM (Reads Per Kilobase of exon model per Million mapped reads), and the like. Preferably, the state information of the sample comprises, but is not limited to, region, race, age, gender, whether or not suffering from a disease, whether or not a tumor sample, whether or not alive, and the like.

[0015] Preferably, after setting the number of quantile intervals, the dynamic quantile number ranking algorithm automatically determines the quantile interval expression value range according to all expression values of each gene of each dataset.

[0016] Preferably, the pan-gene set is composed of genes stably expressed in no less than a specified percentage (range 0-100%) of datasets.

[0017] The standardized matrix can be used to construct a diagnostic model for any disease.

[0018] In a second aspect, the present application provides a cross-disease generalized diagnostic system, comprising: a) a data receiving module for receiving the cross-platform expression matrix (dataset) standardized by the method of claim 1, and dividing into training dataset and validation dataset; b) a gene filtering module: specifying genes stably expressed in any percentage (range 0-100%) of datasets, or directly specifying a gene set of interest; c) a feature selection module to screen high diagnostic performance genes by receiver operating characteristic curve (ROC); d) a fusion modeling module to further screen high diagnostic performance genes with high correlation coefficients by algorithms including but not limited to logistic regression algorithm, and to construct disease diagnostic models based on training data sets; e) a model validation module to use the above disease diagnostic models to make diagnostic predictions on validation data sets; f) a result output module to output diagnostic performance and correlation coefficients of genes, diagnostic performance of models, visualized images, etc.

[0019] Further, the weights of the fusion modeling module are determined by Bayesian model averaging or stacked generalization methods.

[0020] According to claim 8 or 9, the result output module further comprises a visualization unit for real-time display of risk classification and threshold changes.

[0021] In a third aspect, the present application provides a companion diagnostic kit comprising: a) a computer readable medium storing the code of the method of any one of claims 1 or 2; b) a detection set of primer probes or antibodies for detecting the expression of the diagnostic genes of claim 2; c) an embedded microprocessor integrated with the system of claim 2; d) a visualization interface for outputting risk classification under dynamic threshold.

[0022] Further, one or more of the primer probe or antibody combinations for detecting the expression of the diagnostic genes.

[0023] Further, the primer probe combination is provided in a freeze-dried powder or liquid form packaged in a disposable microfluidic chip or a solid phase carrier.

[0024] Compared with the prior art, the present application has the following beneficial effects: 1. effectively eliminating batch effects and data heterogeneity In existing expression-based integration methods such as DESeq2, edgeR, etc., although differential expression analysis can be performed, these methods often have difficulty in completely eliminating batch effects and technical errors between different datasets, especially when facing multi-source data. This is because they often rely on the difference in absolute expression values, and fail to completely control the variation caused by sample source, technical processing, etc. The present application converts the expression data of each dataset into relative values by sorting and scoring the gene expression values of each sample in each dataset, thereby eliminating the bias caused by differences in experimental design and data processing methods. Through this sorting and scoring method, the heterogeneity between different datasets can be effectively reduced, thereby ensuring the accuracy of the data integration results.

[0025] 2. Simplify the calculation process and reduce model complexity Existing model-based integration methods such as limma and ComBat, although they can control batch effects to some extent, these methods often rely on complex statistical models, and the calculation process is tedious and requires high-performance computing resources. The scoring algorithm of the present application uses a simple sorting and scoring method without complex hypothesis conditions and model requirements, and the calculation process is intuitive and efficient. Such simplification not only improves the calculation efficiency, but also reduces the operation threshold, so that researchers can easily realize data integration without relying on advanced statistical models.

[0026] 3. Strong adaptability, supporting multiple types of data input Traditional gene expression integration methods usually have strict requirements for input data, such as standardization or log conversion, which may limit or bias the data preprocessing stage. Unlike this, the algorithm of the present application can flexibly handle different types of gene expression data, including raw data, standardized data, and log-converted data. This adaptability makes the algorithm more versatile and able to meet diverse experimental needs. This means that regardless of the data preprocessing situation, the present application can maintain high integration effectiveness.

[0027] 4. Enhances the interpretability of data integration In existing deep learning methods, although these methods can extract potential patterns from complex data, their "black box" nature makes the results difficult to understand and interpret. The present application processes data through sorting and scoring of gene expression values, and the entire process is intuitive and easy to understand, and the results can be clearly explained. This transparency and interpretability allows researchers to better understand the data processing process when applying the present application, improving the credibility and practicality of the algorithm.

[0028] 5. Improving the integration effect of large-scale datasets With the dramatic increase in data volume in biological research, effectively integrating large-scale data has become a significant challenge. Existing methods may suffer from inefficiency or performance bottlenecks when processing large datasets due to high computational complexity. By sorting and scoring the gene expression values ​​of each sample, this invention provides a simple and efficient data integration scheme that not only reduces the computational burden but also enables the processing of larger datasets. This efficiency makes this invention particularly suitable for applications requiring substantial data support, such as tumor diagnostic models, thereby improving the training effect, accuracy, and generalization ability of the models.

[0029] The advantages of this invention over existing technologies are not only reflected in its simple and easy-to-implement calculation process, but also in its ability to solve problems such as batch effects, data heterogeneity, and computational complexity that cannot be completely eliminated in existing methods. These significant advantages enable this invention to provide a more accurate, reliable, and efficient solution in the field of biological data integration. Attached Figure Description

[0030] Figure 1 A flowchart of the method provided by the present invention; Figure 2 The ROC (AUC=0.981) trend chart of the training set provided by this invention; Figure 3 This is a schematic diagram of the training results provided by the present invention; Figure 4 This invention provides a visualization output of the model gene coefficients. Figure 5 The ROC (AUC=0.983) trend chart of the validation set provided for this invention; Figure 6 The polar coordinate diagram of the model gene coefficients provided by this invention; Figure 7 Molecular typing survival curves provided for this invention; Detailed Implementation

[0031] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0032] For better illustration of the content of this embodiment, the following professional terms are added for explanation: microarray, RNA sequencing, digital PCR, real-time quantitative PCR, single-cell RNA sequencing, Nanostring technology, in situ hybridization (FISH), gene chip.

[0033] Embodiment 1 As shown in Figure 1 , UniQuant (full name Universal Quantile-Based Transcriptome Integration) is an R language software package developed by Huanhou Su, designed for cross-platform heterogeneous transcriptome data integration analysis. This tool unifies the gene expression scale of different data sets through dynamic quantile standardization method, and combines gene screening, diagnostic model construction, molecular typing and survival analysis functions to provide a complete analysis process from raw data to disease mechanism analysis.

[0034] 1.1 Core functions a) Data standardization: eliminate technical differences between different data sets through dynamic quantile standardization method to unify the expression scale.

[0035] b) Gene screening: screen high variable genes and disease related genes, focus on key biological characteristics.

[0036] c) Diagnostic model construction: build diagnostic models based on screened genes and verify their performance on independent data sets.

[0037] d) Molecular typing: divide samples into different molecular subtypes according to gene expression patterns, and analyze the survival differences between subtypes.

[0038] e) Differential expression analysis: based on the molecular typing scheme, screen disease related or typing related differential expression genes, and analyze the molecular mechanism in depth.

[0039] 1.2 Application scenarios UniQuant is mainly used for: multi-data set integration analysis, disease diagnostic marker screening, molecular typing research, survival prognosis analysis.

[0040] II. System requirements 2.1 Hardware requirements The basic requirements of UniQuant for computer hardware are as follows: Processor: It is recommended to use at least dual-core processor, and multi-core processor is recommended when processing large-scale data.

[0041] Memory: At least 4GB of memory is required for the base configuration, and 8GB or more is recommended for handling large datasets (e.g., more than 10,000 samples).

[0042] Storage: At least 5GB of available hard disk space, depending on the size of the dataset.

[0043] 2.2 Software Requirements Operating System: Supports major operating systems such as Windows, Linux, and macOS.

[0044] R Environment: It is recommended to use R version 4.0 or later.

[0045] Dependency Packages: ggplot2: For visualizing analysis results edgeR: For gene expression data analysis pROC: For calculating and plotting ROC curves stringr: For string operations III. Installation and Configuration 3.1 Installing UniQuant UniQuant is not currently available on CRAN (Comprehensive R Archive Network), so it can be installed by following these steps: 1. Ensure that your system has R and RStudio installed (recommended).

[0046] 2. Install the necessary dependency packages, including tools for data visualization, gene expression analysis, and more.

[0047] 3. Install UniQuant from the GitHub repository (assuming it has been published): First, install the devtools package, then use it to obtain UniQuant from GitHub.

[0048] Note: Since UniQuant may be in development, make sure to properly configure the R environment before installation and follow the official documentation or developer guidelines for operation.

[0049] 3.2 Configuring the Working Environment After installation, load UniQuant in the R environment and set the working directory: specify the main directory and the directory where the raw data is stored, and set the random seed to ensure that the results are reproducible.

[0050] Note: It is recommended to set the random seed before starting the analysis to ensure consistent results when running at different times.

[0051] IV. Data Preparation 4.1 Data Types and Format Requirements UniQuant accepts the following two types of data as input: 4.1.1 Raw Gene Expression Data Format: Matrix or data frame format, where: Rows represent genes, columns represent samples Row names should be gene symbols (e.g., "EEF1A1", "LOC643334") Column names should be sample IDs (e.g., "GSM890128", "GSM890129") Example: GSM890128 GSM890129 GSM890130 GSM890131 EEF1A1 14.054179 14.395189 14.349301 14.183636 LOC643334 6.543733 6.399822 6.391479 6.190396 SLC35E2 6.065169 6.172510 5.957617 5.980207 LOC642820 6.862825 6.699461 6.726557 6.664170 4.1.2 Sample Disease Status Data Format: Data frame format, where: The first column is the sample name (consistent with the column name in gene expression data) The second column is the disease status ("No" for non-disease samples, "Yes" for disease samples) The third column is the survival status (0 for survival, 1 for death) optional The fourth column is the follow-up time (in months) optional Example: Sample Disease Outcome Time GSM890128 No — — GSM890129 No — — GSM890130 No — — GSM890131 No — — Note: Non-disease samples do not need to fill in survival information.

[0052] 4.2 Data Loading and Preprocessing Before starting analysis, the original data needs to be loaded into the R environment: 1. Load individual datasets: read gene expression data and disease status data separately.

[0053] 2. Batch load multiple datasets: define a training set or validation set list, and load datasets one by one through a loop and subsequent processing.

[0054] Note: It is recommended to save the original data in RDS format for efficient loading and processing.

[0055] Five, Core Analysis Process 5.1 Quantile Score Quantile Score is the core function of UniQuant, which converts the original expression to a quantile score, unifying the expression scale of different datasets.

[0056] 5.1.1 Basic Usage Quantile normalization is applied to each dataset. The user needs to specify the dataset name, raw expression data, dataset type (training set or validation set), and disease status data.

[0057] 5.1.2 Batch processing multiple datasets Multiple datasets are processed separately: the expression data and disease status data of each dataset are read in a loop, and the quantile normalization function is called to complete the processing.

[0058] 5.1.3 Principle of quantile normalization Quantile normalization converts the original expression of each gene in the sample to an integer fraction of 1-10, where 1 represents the lowest 10% of the gene expression in the sample, and 10 represents the highest 10%. This conversion method can effectively eliminate technical differences between different datasets, making data from different sources comparable.

[0059] Example: The expression data format after quantile conversion is as follows: GSE36376@GSM890128 GSE36376@GSM890129 GSE36376@GSM890130 GSE36376@GSM890131 EEF1A1 4 9 8 6 LOC643334 10 8 8 4 SLC35E2 4 7 2 2 LOC642820 9 5 6 4 Note: Quantile normalization adds dataset identification (such as "GSE36376@GSM890128") to each sample ID to ensure the uniqueness of the sample when integrating multiple datasets.

[0060] 5.2 Hypervariable gene screening (UniQuant_Hypervariable_gene) Hypervariable gene screening is to identify genes with high expression variation in multiple datasets, which may be closely related to the occurrence and development of diseases.

[0061] 5.2.1 Basic usage When screening hypervariable genes, the number of genes to be screened (such as 1000) and the proportion of datasets that the genes need to cover (such as 70%) need to be set. The screening result is a list of hypervariable genes.

[0062] 5.2.2 Parameter interpretation n_hypervariable_gene: The number of hypervariable genes to be screened, with a value range of 1-3000.

[0063] dataset_threshold: The proportion of datasets that the gene needs to express on, with a value range of 0-1.0.

[0064] 5.2.3 Example of screening results The screening result is a character vector containing the symbols of the high-variable genes: "ACSL4", "AFP", "AKR1B10", "ALDH3A1", "APOA4", "APOF", "C7", "C9", "CCL19", "CCL20"; the variable gene screening calculates the coefficient of variation (CV) of the expression of each gene in each dataset and selects the genes with the highest CV.

[0065] 5.3 Disease-related gene screening (UniQuant_Disease_gene) Disease-related gene screening is a further screening of genes from high-variable genes that can effectively distinguish between diseased and non-diseased samples.

[0066] 5.3.1 Basic usage When screening disease-related genes from high-variable genes, the gene list to be screened and the minimum AUC threshold for diagnosing the disease need to be specified, and the result is a list of disease-related genes.

[0067] 5.3.2 Parameter interpretation ls_gene: the gene list to be screened, usually the result of high-variable gene screening.

[0068] AUC_threshold: the minimum AUC threshold for diagnosing the disease, with a value range of 0-1.0.

[0069] 5.3.3 Example of screening results As shown in Figure 3 and Figure 4 , the screening result is a character vector containing the symbols of the disease-related genes: "ACSL4", "AKR1B10", "APOA4", "APOF", "C7", "C9", "CCL19", "CCL20", "CTHRC1", "CYP1A2"; Note: Disease-related gene screening is based on the AUC value (Area Under the ROC Curve) of each gene in each dataset, and the closer the AUC value is to 1, the stronger the ability of the gene to distinguish between diseased and non-diseased samples.

[0070] 5.4 Diagnosis model construction and verification (UniQuant_Model_Training and UniQuant_Model_Validation) Diagnosis model construction is to establish a multi-gene diagnosis model based on disease-related genes to distinguish between diseased and non-diseased samples.

[0071] 5.4.1 Model construction Build a diagnostic model based on disease-related genes, input is a list of disease-related genes, and get the model-related results.

[0072] 5.4.2 Model validation Validate the performance of the diagnostic model on the validation set, without additional input parameters, directly call the function to get the validation results.

[0073] 5.4.3 Result analysis, as shown in Figure 5 to Figure 6

[0074] The results of model construction and validation include multiple components: Model_gene: The final gene list used to build the model.

[0075] AUC: The AUC value of the model on the training set or validation set.

[0076] Cutoff: The best threshold and its corresponding sensitivity, specificity, and accuracy.

[0077] Coefficient: The coefficient of each gene in the model.

[0078] ROC_Model_plot: ROC curve of the model.

[0079] ROC_Gene_plot: ROC comparison chart of each gene and the model.

[0080] 5.4.4 Model performance visualization Extract the model data of the training set, draw the ROC curve: add a diagonal reference line, set the chart title, coordinate axis style, and label the model AUC value, etc. Note: The model is built using the logistic regression method, combining the expression levels of multiple disease-related genes to build a multi-gene diagnostic model.

[0081] 5.5 Molecular typing (UniQuant_Class) Molecular typing is to divide the disease samples into different molecular subtypes based on gene expression patterns, which may have different clinical characteristics and prognosis.

[0082] 5.5.1 Basic usage Molecular typing based on high-variable genes, need to specify the gene list used for typing, the number of classes divided (such as 3 classes) and the typing method (such as "NMF"), the result contains the sample typing table.

[0083] 5.5.2 Parameter interpretation ls_gene: The gene list used for typing, usually the result of high-variable gene screening.

[0084] ​n_Class: Number of classes to be partitioned.

[0085] method: Partitioning method, can be "CCP" or "NMF".

[0086] 5.5.3 Result Analysis The partitioning result is a data frame containing samples and corresponding classes: Sample Class GSE102079@GSM2723193 Class_1 GSE102079@GSM2723195 Class_2 GSE102079@GSM2723197 Class_2 GSE102079@GSM2723198 Class_1 GSE102079@GSM2723199 Class_3 GSE102079@GSM2723200 Class_2 5.5.4 Partitioning Quantity Statistics Sample quantity statistics of different classes in the partitioning result: Class_1 Class_2 Class_3 570 690 708 Note: Molecular partitioning methods include constrained clustering (CCP) and non-negative matrix factorization (NMF), with NMF as the default method.

[0087] 5.6 Survival Analysis Survival analysis is an important method to evaluate the relationship between molecular partitioning and patient prognosis.

[0088] 5.6.1 Basic Usage 1. Load the survival analysis related package.

[0089] 2. Set the working directory, specify the quantile data and result storage path.

[0090] 3. Load the disease state data and molecular partitioning results of the dataset.

[0091] 4. Filter samples containing survival information, remove samples without follow-up time.

[0092] 5. Merge the disease state data and molecular partitioning results to obtain a dataset containing survival information and partitioning.

[0093] 6. Perform survival analysis, build a survival model with partitioning as the grouping variable.

[0094] 7. Draw survival curves, set chart parameters such as display P value, legend position, color, etc.

[0095] 5.6.2 Result Interpretation As Figure 7 shown, the survival analysis result is usually presented in a survival curve graph. If the survival curves of different molecular partitions are significantly separated and the P value is less than 0.05, it indicates that the partitioning is significantly related to prognosis.

[0096] Example: The survival analysis result based on the GSE14520 dataset shows that the survival curves of patients in three molecular partitions have significant differences (P=0.0054).

[0097] Note: Survival analysis is only applicable to diseased samples, non-diseased samples do not need to fill in survival information.

[0098] 5.7 Differential gene expression analysis (DEG Analysis) Differential gene expression analysis is to identify genes with significant expression difference between different groups (e.g. diseased vs non-diseased, or different molecular subtypes).

[0099] 5.7.1 Disease status-based differential expression analysis Disease-related differential expression analysis is performed, and the gene needs to be significantly different in at least a certain proportion of the dataset (e.g. 70%), and the disease-related differential gene list is obtained.

[0100] 5.7.2 Differential expression analysis based on molecular subtypes Molecular subtype-related differential expression analysis is performed, and the subtype result, target category and control category (e.g. comparing Class_3 with the combination of Class_1+Class_2) need to be specified, and the dataset proportion threshold is set to obtain the subtype-related differential gene.

[0101] 5.7.3 Parameter interpretation dataset_threshold: The differential gene needs to be significantly different in at least a certain proportion of the dataset, with a value range of 0-1.0.

[0102] input_Class: Molecular subtype result data frame.

[0103] Class_test: Target category to be compared.

[0104] Class_control: List of control groups.

[0105] 5.7.4 Result interpretation The differential expression analysis result contains the list of differential genes and their expression fold change and significance level, which can be further used for enrichment analysis and subsequent research.

[0106] Differential expression analysis is performed using the edgeR package, and the exact test method is used to detect differentially expressed genes.

[0107] Six, advanced applications and extensions 6.1 Custom gene set analysis In addition to using high variable genes and disease-related genes, users can also customize gene sets for analysis according to research interests: 6.1.1 Using custom gene sets for diagnostic model construction Define a custom gene set (e.g., immune-related genes) and input it into the diagnostic model construction function to obtain the diagnostic model results based on this gene set.

[0108] 6.1.2 Molecular typing using custom gene sets Molecular typing based on custom gene sets, set the number of categories and typing method, get the corresponding molecular typing results.

[0109] Note: Custom gene set analysis can be used to verify the role of genes in specific biological pathways or functional categories in disease.

[0110] 6.2 Dataset removal and reanalysis If a dataset has quality problems or abnormalities, it can be removed from the analysis: 6.2.1 Remove specific datasets Specify the dataset name to be removed, call the dataset removal function to exclude it from the analysis.

[0111] 6.2.2 Reanalysis after removal After removing the dataset, re-perform high-variable gene screening and diagnostic model construction to obtain updated results.

[0112] Note: The dataset removal function allows users to exclude datasets with poor quality or abnormalities, improving the reliability of the analysis results.

[0113] 6.3 Performance evaluation of diagnostic models on each dataset The performance of the diagnostic model on each dataset can be evaluated to identify datasets that may have heterogeneity: 6.3.1 AUC values of models on each dataset Evaluate the AUC values of the diagnostic model on all datasets to obtain a list of AUC values corresponding to each dataset.

[0114] 6.3.2 Visualize AUC values for each dataset Draw a bar chart of AUC values for each dataset, set chart title, axis labels, and color parameters, and visually display the performance of the model on different datasets. The AUC values of the model on each dataset can be used to identify datasets that may have heterogeneity, and further investigate the reasons.

[0115] Seven, result output and visualization 7.1 Key result files The main result files generated by UniQuant analysis include: 1. Quantile-normalized data: stored under the directory "UniQuant / dataset" as "@Expression.rds" and "@Disease.rds".

[0116] 2. Hypervariable gene list: stored under the directory "UniQuant / result" as "Hypervariable_gene.rds".

[0117] 3. Disease-associated gene list: included in the result of disease-associated gene screening.

[0118] 4. Diagnosis model result: stored under the directory "UniQuant / result" as "Diagnosis_Model.rds", including model genes ("Model_gene.rds"), coefficients, and AUC values, etc.

[0119] 5. Molecular typing result: stored as "df_Class.rds" under the directory "UniQuant / result".

[0120] 6. Differential expression analysis result: included in the variables "rst_DEG_Disease" or "rst_DEG_Class".

[0121] 7.2 Visualization result: the main visualization results generated by UniQuant are stored under the directory "UniQuant / image", as shown in Figure 7 .

[0122] 1. ROC curve: shows the performance of the diagnosis model on the training set and the validation set.

[0123] 2. Gene coefficient plot: shows the coefficients of each gene in the model.

[0124] 3. Survival curve: shows the survival difference of different molecular typing.

[0125] 4. Differential expression volcano plot: shows the distribution of differentially expressed genes.

[0126] 7.2.1 Gene coefficient visualization 1. Extract the model coefficient data, and arrange the gene names and coefficient values (retain three decimal places).

[0127] 2. Draw the polar coordinate plot of gene coefficients: draw a column chart with genes as the horizontal axis and coefficients as the vertical axis, add reference lines, and convert to polar coordinates.

[0128] 3. Add color gradient and beautify: Set color gradient range, adjust legend position and style, beautify chart theme and axis style.

[0129] Note: Gene coefficient plot is displayed in polar form, which is intuitive to understand the contribution of each gene to the diagnostic model.

[0130] More specifically, the model building experiment environment 1. System architecture

[0131] 1. Module details (Module Details) include: 1) Quantile score module Input: Expression matrix E[gene × sample], disease label D(sample, Disease, Outcome, Time) steps: ① Sort the expression value of each gene in all samples; ② Map to 0-10 integer rank (0=lowest, 10= highest); ③ Concatenate dataset name as column name (e.g. GSE14520@GSM362958).

[0132] Output: Quantile matrix Q[gene × sample_q].

[0133] 2) Hypervariable gene module (UniQuant_Hypervariable_gene) Definition: In at least dataset_threshold proportion of datasets, the expression variance of the gene is in the top n_hypervariable_gene.

[0134] Algorithm: for each dataset d var_g = variance(E_d[g,]) top_g = rank(var_g) <= n_ hypervariable_gene keep g if sum(top_g across d) / N_datasets >= dataset_ threshold 3) Disease gene voting module (UniQuant_Disease_gene): Calculate the AUC and average AUC of each high-variability gene for diagnosing the disease in the training set, obtain the behavior gene, and list the AUC matrix of the dataset, then retain the genes with average AUC≥AUC_threshold.

[0135] 4) Diagnostic model module Training: Use the retained genes to build a multivariate logistic regression Internal validation: 5-fold cross-validation; External validation: independent validation set directly predict and plot ROC.

[0136] 5) Molecular subtyping module Input: quantile matrix of high-variant genes (only disease samples); Method: CCP (Consensus Cluster Plus): K-means 50 iterations, 80% resampling; NMF: Brunet algorithm, 200 runs; Output: optimal K subtypes labels and heatmap.

[0137] Key parameter design Parameter Value Range Recommended Default Effect n_hypervariable_gene 500–3000 1000 Control feature dimension dataset_threshold 0.5–1.0 0.7 Cross dataset consistency disease_threshold 0.5–1.0 0.7 Cross disease consistency n_Class 2–10 3 Number of molecular subtypes method "CCP” / "NMF” "CCP” Clustering algorithm Running example Scenario: user has 19 liver cancer datasets, 9 of which constitute the training dataset, and the other 10 constitute the validation dataset. Based on the training dataset, build a liver cancer diagnosis model, and perform molecular subtyping on the liver cancer samples of the entire dataset.

[0138] Step 1: directory structure project / ├─0.raw / │ ├─GSE1001_Expression.rds│ ├─GSE1002_ Expression.rds│ ├─GSE1003_Expression.rds│ ├─*Disease.rds├─1.Quantile / ├ ─2.UniQuant / └─run.R Step 2: run script run.R (<30 lines) library(UniQuant) dir_main<- "project_path"; setwd(dir_main) dir_in <- "rawdata_path" ① Quantile transformation Training set Training setup ls_datassets = c('GSE6764', 'GSE14520', 'GSE17856', 'GSE57957', ' GSE36376','GSE102079', 'GSE54236', 'GSE36411', 'GSE22058') for (ls_datasset in ls_datassets){ df_raw = readRDS(file.path(dir_in, paste0(ls_datasset, '_ Expression.rds'))) df_Disease = readRDS(file.path(dir_in, paste0(ls_datasset, '_ Disease.rds'))) Quantile_score(ls_datasset, df_raw, "Training", df_Disease) } Validation set setup ls_datassets = c('PMID31585088', 'PMID35382356', 'GSE77314', ' GSE76427','GSE25097', 'GSE63898', 'GSE39791', 'GSE144269','GSE114564',' GSE14811') for (ls_datasset in ls_datassets){ df_raw = readRDS(file.path(dir_in, paste0(ls_datasset, '_ Expression.rds'))) df_Disease = readRDS(file.path(dir_in, paste0(ls_datasset, '_ Disease.rds'))) Quantile_score(ls_datasset, df_raw, "Validation", df_Disease) }②–⑤ Five-step method gene <- UniQuant_Hypervariable_gene(1000, 0.7)$Gened gene <- UniQuant_Disease_gene(gene, 0.7)$Disease_gene model <- UniQuant_Model_Training(dgene) val <- UniQuant_Model_Validation() subtype<- UniQuant_Class(gene, 3, "NMF") Visualize output model$ROC_Model_plot ggsave("training_roc.pdf", width=5,height=5) Step 3: Results training_roc.pdf: AUC=0.985 df_Class.csv: Sample subtype labels volcano_DEG.pdf: Subtype differential gene volcano plot 2.5 Results Validation Validation dimension Metrics Results Explanation Internal cross-validation AUC 0.981 ± 0.006 5-fold CV External independent validation AUC 0.983 10 non-overlapping datasets Survival analysis Log-rank p 0.0054 Significant difference among 3 subtypes A. Quantile normalization: Map raw expression to 0-10 integer ranks, remove sequencing depth, platform, batch differences.

[0139] B. Generalization threshold: dataset_threshold, disease_threshold ensure genes are consistently valid across ≥70% datasets.

[0140] C. Cross-platform or study validation: AUC>0.98 across 9 training sets + 10 independent validation sets.

[0141] D. One-click R package: Pure R language, few dependencies, supports Windows / macOS / Linux.

[0142] According to the above experimental environment, a medical record diagnosis model instance is constructed, Example 2: Hepatocellular Carcinoma (HCC) Diagnosis Training: GSE6764, GSE14520... 9 datasets (n=2,167); Validation: PMID31585088, GSE77314... 10 datasets (n=1,538); Results: Model AUC_training = 0.981, AUC_validation = 0.983; Core genes AKR1B10, DCN, CYP1A2, etc. 12.

[0143] Example 3: Breast cancer diagnosis Datasets: GSE44272, GSE32489, GSE124817, GSE46141, GSE124647, GSE138813, GSE113890, GSE147995, GSE291822, GSE26338. Set disease_threshold = 0.7, found 8 genes CCL19, FOS, etc. significantly differentially expressed in multiple breast cancer datasets, AUC 0.75-0.92.

[0144] Example 4: Intestinal cancer chemotherapy sensitivity gene analysis Datasets: GSE203071, GSE41568, GSE72968, GSE72969, GSE100480, GSE189212, GSE200407, GSE62080. Set disease_threshold = 0.7, found genes GSTP1, XRCC1, Survivin, PinX1, WBSCR22, etc. are highly related to intestinal cancer chemotherapy sensitivity.

[0145] Example 5: Integration of bulk RNA expression profile and single-cell RNA transcriptome After converting the bulk RNA expression profile to quantile, it is combined with the single-cell RNA transcriptome data; using UniQuant_Class() to obtain 3 microenvironment subtypes (Immune-high, Metabolic, Proliferative), Kaplan-Meier p=0.0054.

[0146] It should be noted that the information interaction, execution process, etc. between the above devices / units, since the same concept as the method embodiments of the present application, its specific functions and the technical effects brought about, specific can refer to the method embodiment part, here will not be repeated.

[0147] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction, and do not limit the protection scope of the present application. The specific working process of the unit and module in the system can refer to the corresponding process in the foregoing method embodiment, which will not be described here.

[0148] The embodiment of the present application also provides a network device, comprising at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor implements the steps in any of the foregoing method embodiments when executing the computer program.

[0149] The embodiment of the present application provides a computer program product, which, when running on a mobile terminal, enables the mobile terminal to implement the steps in any of the foregoing method embodiments.

[0150] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the embodiment of the present application can implement all or part of the processes in the foregoing method embodiments by a computer program to instruct related hardware to complete, and the computer program can be stored in a computer-readable storage medium. When the processor executes the computer program, the steps of the foregoing method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer-readable medium at least includes any entity or device capable of carrying the computer program code to the photographing apparatus / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium. For example, U disk, mobile hard disk, magnetic disk or optical disk, etc. In some jurisdictions, according to legislation and patent practice, the computer-readable medium cannot be an electrical carrier signal and a telecommunication signal.

[0151] In the above embodiments, the description of each embodiment focuses on different aspects, and the parts not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments.

[0152] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0153] In the embodiments provided in the present application, it should be understood that the disclosed apparatus / network device and method can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely schematic, for example, the division of the modules or units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0154] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0155] The above described embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A dynamic quantile normalization method for cross-platform heterogeneous data, characterized in that, include: a) Collect gene expression matrices from different detection platforms or experimental batches. The number of samples in a single dataset should be greater than 10, and any data format can be used. b) Within each dataset, all expression values ​​for each gene are sorted and divided into 2–50 variable quantile intervals using a dynamic quantile ranking algorithm; c) Within each dataset, for each gene in each sample, the variable quantile intervals are mapped to uniformly distributed discrete rank values ​​to obtain a cross-platform comparable relative expression matrix; d) Based on the state information of the samples, assist the analysis process and construct corresponding analysis models; e) When integrating datasets, you can specify to retain ubiquitous gene sets that exist in any proportion of the dataset, and if a dataset is missing any number of ubiquitous genes, the average relative expression value will be automatically imputed. f) The output is a standardized matrix that can be directly used for machine learning modeling or clinical decision support.

2. The dynamic quantile normalization method for cross-platform heterogeneous data according to any one of claims 1, characterized in that, The detection platform includes at least one of microarray, RNA sequencing, digital PCR, real-time quantitative PCR, single-cell RNA sequencing, Nanostring technology, Insitu hybridization, and gene chip.

3. A dynamic quantile normalization method for cross-platform heterogeneous data according to any one of claims 1, characterized in that, The data formats mentioned include, but are not limited to, Count, TPM, CPM, FPKM, and RPKM.

4. A dynamic quantile normalization method for cross-platform heterogeneous data according to any one of claims 1, characterized in that, The status information of the sample includes, but is not limited to, region, race, age, gender, whether it is diseased, whether it is a tumor sample, and whether it is alive.

5. A dynamic quantile normalization method for cross-platform heterogeneous data according to claim 1, characterized in that... After setting the number of quantile intervals, the dynamic quantile ranking algorithm automatically determines the range of expression values ​​for each quantile interval based on all expression values ​​of each gene in each dataset.

6. The dynamic quantile normalization method for cross-platform heterogeneous data according to claim 1, characterized in that... The ubiquitous gene set consists of genes that are stably expressed in the dataset at a percentage not less than a specified percentage, ranging from 0% to 100%.

7. A dynamic quantile normalization method for cross-platform heterogeneous data according to any one of claims 1-6, characterized in that, The standardized matrix can be used to construct diagnostic models for any disease.

8. A cross-disease generalized diagnostic system, characterized in that, include: The data receiving module is used to receive the cross-platform representation matrix standardized by the method described in claim 1, and divide it into a training dataset and a validation dataset; Gene filtering module: Specify the percentage of genes stably expressed in the dataset, or directly specify the set of genes of interest; The feature selection module screens genes with high diagnostic performance based on receiver operating characteristic curves. The fusion modeling module further screens genes with high diagnostic performance and high correlation coefficients through algorithms including but not limited to logistic regression, and builds disease diagnostic models based on the training dataset; The model validation module uses the aforementioned disease diagnosis model to perform diagnostic predictions on the validation dataset. The results output module outputs the diagnostic performance and correlation coefficient of the gene, the diagnostic performance of the model, and a visualized image.

9. The diagnostic system according to claim 8, characterized in that, The weights of the fusion modeling module are determined by Bayesian model averaging or stacking generalization methods.

10. The diagnostic system according to claim 8 or 9, characterized in that, The result output module also includes a visualization unit for real-time display of risk classification and threshold changes.

11. A companion diagnostic kit, characterized in that, include: A computer-readable medium storing the method encoded according to any one of claims 1–10; Detection combination of primer probes or antibodies for detecting the expression of the diagnostic gene as described in claim 8; An embedded microprocessor integrating the system of claim 8; A visual interface for outputting risk classification under dynamic thresholds.

12. The kit according to claim 11, characterized in that, The combination of one or more of the primers, probes or antibodies used to diagnose gene expression.

13. The reagent kit according to any one of claims 9-11, characterized in that, The primer-probe combination is encapsulated in a disposable microfluidic chip or solid carrier in the form of lyophilized powder or liquid.

Images