Multi-source data fusion method and device, electronic equipment and storage medium

Abstract

The invention discloses a multi-source data fusion method and device, electronic equipment and a storage medium, and the method comprises the steps: collecting various omics data and phenotypic data of an observation object, carrying out the preprocessing of the various omics data and phenotypic data, and obtaining various target omics data and target phenotypic data, the method comprises the steps of respectively constructing similarity matrixes for various target omics data and target phenotype data, fusing all the similarity matrixes to generate a fusion network, and establishing an attribute graph of an observation object based on the fusion network, the various target omics data and the target phenotype data. The technical problem that in the related technology, omics data and phenotypic data with large data structure difference cannot be effectively fused, and consequently the fusion result has large deviation is solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a multi-source data fusion method and device, electronic equipment and storage medium. Background Art

[0002] With the continuous development of high-throughput experimental technology, massive omics data can be obtained in a relatively short time. These omics data make it possible to analyze the occurrence and development mechanism of complex system biology phenomena at the molecular level. The central dogma in genetics states that in organisms, genetic information in cells is transferred from DNA (deoxyribonucleic acid) to mRNA (messenger RNA), and then to proteins. The research on these three types of substances is called genomics, transcriptomics, and proteomics, respectively. In addition, although metabolomics, lipidomics, and glycomics are not part of the central dogma, they still provide valuable information about metabolites, lipids, and carbohydrates, which are intermediates of cellular information flow. In the context of precision medicine, any single omics data is not enough to explain the mechanism of complex system biology diseases, and the integration and analysis of multi-omics data is a hot topic and trend in current research. In recent years, multi-omics data fusion has gradually become the technical basis for biomedical research. The exploration of technologies and applications including machine learning analysis is still in its infancy. The development of such data fusion technology will provide a new technical basis for the occurrence and development of complex biological phenomena.

[0003] Currently, various bio-omics data (for example, genomics, transcriptomics, and proteomics) contain a large amount of key biological information. These data are characterized by redundancy, heterogeneity, high noise, and complex correlation patterns. Traditional biostatistics methods often have limitations such as shallow feature depth, and are often difficult to effectively process massive multi-dimensional omics data.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a multi-source data fusion method and device, electronic device and storage medium, so as to at least solve the technical problem in the related art that omics data and phenotypic data with large data structure differences cannot be effectively fused, resulting in large deviations in the fusion results.

[0006] According to one aspect of an embodiment of the present invention, a method for fusing multi-source data is provided, comprising: collecting a variety of omics data and phenotypic data of an observed object, and preprocessing the multiple omics data and the phenotypic data to obtain a variety of target omics data and target phenotypic data; constructing similarity matrices for the multiple target omics data and the target phenotypic data, respectively; fusing all the similarity matrices to generate a fusion network; and establishing a property graph of the observed object based on the fusion network, the multiple target omics data and the target phenotypic data.

[0007] Optionally, the step of preprocessing the multiple omics data and the phenotypic data to obtain multiple target omics data and target phenotypic data includes: determining the omics features of each sample in the omics data and the omics feature values ​​corresponding to the omics features; performing missing value processing on the omics data based on the omics features and the omics feature values ​​corresponding to the omics features to obtain initial omics data; performing logarithmic processing on the omics feature values ​​of each sample in the initial omics data to obtain initial omics feature values; normalizing the initial omics feature values ​​to obtain target omics feature values, wherein the samples indicated by the target omics feature values ​​constitute the target omics data; and performing preset embedding mapping on the phenotypic data to obtain the target phenotypic data.

[0008] Optionally, based on the omics feature and the omics feature value corresponding to the omics feature, the step of performing missing value processing on the omics data to obtain initial omics data includes: determining a target omics feature in which the missing ratio of the omics feature values ​​in all the samples is greater than a first preset ratio threshold; determining a target sample in which the missing ratio of all omics feature values ​​is greater than a second preset ratio threshold; deleting the target omics feature of each of the samples in the omics data, and deleting the target sample in the omics data; and using a preset mean algorithm to fill the remaining missing values ​​of each of the samples in the omics data after deletion to obtain the initial omics data, wherein the remaining missing values ​​refer to the missing values ​​of the omics feature of each of the samples in the omics data after deletion.

[0009] Optionally, the step of constructing a similarity matrix for a plurality of target omics data and the target phenotypic data, respectively, includes: determining a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data, wherein the observed data is the target omics data or the target phenotypic data, the first sample is any sample in the observed data, and the second sample is any sample in the observed data except the first sample; determining a similarity measurement index value between the first sample and the second sample based on a sample distance between the first sample and the second sample, the first sample diameter, the second sample diameter, a first weighting parameter, and a second weighting parameter; constructing a fully connected similarity matrix based on all the similarity measurement index values; and constructing the similarity matrix of the observed data based on the fully connected similarity matrix.

[0010] Optionally, the step of determining a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data includes: determining a first neighbor sample set of the first sample; determining a first sample distance between the first sample and each first neighbor sample in the first neighbor sample set, and determining the first sample diameter of the first sample based on all the first sample distances; determining a second neighbor sample set of the second sample; determining a second sample distance between the second sample and each second neighbor sample in the second neighbor sample set, and determining the second sample diameter of the second sample based on all the second sample distances.

[0011] Optionally, the step of fusing all the similarity matrices to generate a fused network includes: determining the total information corresponding to each of the similarity matrices and the information weight corresponding to each of the similarity matrices; and generating the fused network based on the total information corresponding to all the similarity matrices and the information weight.

[0012] Optionally, the step of establishing the property graph of the observed object based on the fusion network, multiple target omics data and the target phenotypic data includes: determining the data feature value corresponding to each data feature based on the multiple target omics data and the target phenotypic data; concatenating all the data feature values ​​corresponding to the data feature to obtain the feature attribute of the data feature; determining the data node corresponding to the data feature based on the fusion network; associating the feature attribute as the node attribute of the data node to generate the property graph.

[0013] Optionally, after establishing the attribute graph of the observed object based on the fusion network, multiple target omics data and the target phenotypic data, it also includes: using a preset learning model to process the attribute graph to obtain a vector value of each data node, wherein the preset learning model includes: an input layer, a hidden layer, and an output layer; based on all the vector values, a target attribute graph is generated; based on the target attribute graph, task analysis is performed on the observed object.

[0014] Optionally, a preset learning model is used to process the attribute graph to obtain a vector value of each data node, including: receiving the attribute graph through the input layer; performing preset neighbor node sampling processing on each data node in the attribute graph through the hidden layer to obtain an aggregated node set of each layer of the data node; performing preset neighbor aggregation processing on the aggregated node set to obtain an aggregated feature value; generating a target aggregated feature value based on all the aggregated features; representing the target aggregated feature value as the vector value, and outputting the vector value through the output layer.

[0015] According to another aspect of an embodiment of the present invention, a multi-source data fusion device is also provided, including: a processing unit, used to collect multiple omics data and phenotypic data of an observed object, and pre-process the multiple omics data and the phenotypic data to obtain multiple target omics data and target phenotypic data; a construction unit, used to respectively construct similarity matrices for the multiple target omics data and the target phenotypic data; a fusion unit, used to fuse all the similarity matrices to generate a fusion network; and an establishment unit, used to establish an attribute graph of the observed object based on the fusion network, the multiple target omics data and the target phenotypic data.

[0016] Optionally, the processing unit includes: a first determination module, used to determine the omics features of each sample in the omics data and the omics feature values ​​corresponding to the omics features; a first processing module, used to perform missing value processing on the omics data based on the omics features and the omics feature values ​​corresponding to the omics features to obtain initial omics data; a second processing module, used to perform logarithmic processing on the omics feature values ​​of each of the samples in the initial omics data to obtain initial omics feature values; a third processing module, used to normalize the initial omics feature values ​​to obtain target omics feature values, wherein the samples indicated by the target omics feature values ​​constitute the target omics data; a first mapping module, used to perform preset embedded mapping on the phenotypic data to obtain the target phenotypic data.

[0017] Optionally, the first processing module includes: a first determination submodule, used to determine a target omics feature whose missing ratio of the omics feature values ​​in all the samples is greater than a first preset ratio threshold; a second determination submodule, used to determine a target sample whose missing ratio of all omics feature values ​​is greater than a second preset ratio threshold; a first deletion submodule, used to delete the target omics feature of each of the samples in the omics data, and delete the target sample in the omics data; a first filling submodule, used to fill the remaining missing values ​​of each of the samples in the omics data after deletion using a preset mean algorithm to obtain the initial omics data, wherein the remaining missing values ​​refer to the missing values ​​of the omics feature of each of the samples in the omics data after deletion.

[0018] Optionally, the construction unit includes: a second determination module, used to determine a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data, wherein the observed data is the target omics data or the target phenotypic data, the first sample is any sample in the observed data, and the second sample is any sample in the observed data except the first sample; a third determination module, used to determine a similarity measurement index value between the first sample and the second sample based on a sample distance between the first sample and the second sample, the first sample diameter, the second sample diameter, a first weighting parameter and a second weighting parameter; a first construction module, used to construct a fully connected similarity matrix based on all the similarity measurement index values; and a second construction module, used to construct the similarity matrix of the observed data based on the fully connected similarity matrix.

[0019] Optionally, the second determination module includes: a third determination submodule, used to determine a first neighbor sample set of the first sample; a fourth determination submodule, used to determine a first sample distance between the first sample and each first neighbor sample in the first neighbor sample set, and determine a first sample diameter of the first sample based on all the first sample distances; a fifth determination submodule, used to determine a second neighbor sample set of the second sample; and a sixth determination submodule, used to determine a second sample distance between the second sample and each second neighbor sample in the second neighbor sample set, and determine a second sample diameter of the second sample based on all the second sample distances.

[0020] Optionally, the fusion unit includes: a fourth determination module, used to determine the total information corresponding to each of the similarity matrices and the information weight corresponding to each of the similarity matrices; and a first generation module, used to generate the fusion network based on the total information corresponding to all of the similarity matrices and the information weights.

[0021] Optionally, the establishment unit includes: a fifth determination module, used to determine the data feature value corresponding to each data feature based on multiple target omics data and the target phenotypic data; a first splicing module, used to splice all the data feature values ​​corresponding to the data feature to obtain the characteristic attribute of the data feature; a sixth determination module, used to determine the data node corresponding to the data feature based on the fusion network; and a second generation module, used to associate the characteristic attribute as a node attribute of the data node to generate the attribute graph.

[0022] Optionally, the fusion device also includes: a fourth processing module, which is used to use a preset learning model to process the attribute graph after establishing the attribute graph of the observed object based on the fusion network, the multiple target genomics data and the target phenotypic data, so as to obtain the vector value of each data node, wherein the preset learning model includes: an input layer, a hidden layer, and an output layer; a third generation module, which is used to generate a target attribute graph based on all the vector values; and a first analysis module, which is used to perform task analysis on the observed object based on the target attribute graph.

[0023] Optionally, the fourth processing module includes: a first receiving submodule, used to receive the attribute graph through the input layer; a first processing submodule, used to perform preset neighbor node sampling processing on each data node in the attribute graph through the hidden layer to obtain an aggregate node set of each layer of the data node; a second processing submodule, used to perform preset neighbor aggregation processing on the aggregate node set to obtain an aggregate feature value; a first generating submodule, used to generate a target aggregate feature value based on all the aggregate features; and a first output submodule, used to represent the target aggregate feature value as the vector value, and output the vector value through the output layer.

[0024] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned multi-source data fusion method.

[0025] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned multi-source data fusion method.

[0026] In the present disclosure, a variety of omics data and phenotypic data of the observed object are collected, and the various omics data and phenotypic data are preprocessed to obtain a variety of target omics data and target phenotypic data, and similarity matrices are respectively constructed for the various target omics data and target phenotypic data, and all similarity matrices are fused to generate a fusion network. Based on the fusion network, the various target omics data and the target phenotypic data, an attribute graph of the observed object is established. In the present disclosure, the various omics data and phenotypic data collected can be preprocessed first, and then similarity matrices are respectively constructed for the various target omics data and the target phenotypic data that have been processed, and then all similarity matrices are fused to generate a fusion network, and then the attribute graph of the observed object is established according to the fusion network, the various target omics data and the target phenotypic data, and the fusion of the omics data and the phenotypic data with large data structure differences can be achieved under the premise of retaining the scale meaning of each single data, so as to obtain the attribute graph of the observed object, thereby solving the technical problem that the omics data and the phenotypic data with large data structure differences cannot be effectively fused in the related art, resulting in a large deviation in the fusion result. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0028] Figure 1 is a flowchart of an optional multi-source data fusion method according to an embodiment of the present invention;

[0029] Figure 2 is a flow chart of optional data collection and data preprocessing according to an embodiment of the present invention;

[0030] Figure 3 is a flow chart of an optional similarity matrix calculation and matrix fusion according to an embodiment of the present invention;

[0031] Figure 4 is a schematic diagram of an optional multi-source data fusion process based on a graph convolutional network according to an embodiment of the present invention;

[0032] Figure 5 is a schematic diagram of an optional multi-source data fusion device according to an embodiment of the present invention;

[0033] Figure 6 The present invention is a hardware structure block diagram of an electronic device (or mobile device) for a multi-source data fusion method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0037] In the present invention, many new data integration methods can be introduced from the fields of mathematics, statistics and computational science. These data integration methods can be roughly divided into three categories: integration methods based on series connection, integration methods based on models, and integration methods based on transformation.

[0038] The concatenation-based integration method considers developing models using a joint data matrix that is formed by combining multiple omics datasets. First, the raw data from the individual omics and the corresponding phenotypic information are directly concatenated without any preprocessing, and multiple omics data are spliced ​​into a single large matrix for supervised and unsupervised learning. Once the concatenation of the individual omics is completed, machine learning methods can be simply used for analysis. This type of method mainly includes traditional machine learning algorithms such as DT (Decision Tree), ANN (Approximate Nearest Neighbor), SVM (Support Vector Machine), LASSO (Least absolute shrinkage and selection operator), SVN (Support vector machine and upgrade algorithm), DNN (Deep Neural Network Algorithm), Joint NMF (Non-negative Matrix Decomposition), iCluster (Integrated Classification Algorithm Based on Multi-omics Data), JIVE (An open Java (a programming language) source code project), MoCluster (a clustering algorithm), etc. This type of method can use all connected features equally and can select the most discriminative features for a given phenotype.

[0039] Model-based integration methods create multiple intermediate models for different omics data, and then build the final model based on each intermediate model. The process is: first, multiple single-omics data matrices are constructed for the raw data of each individual omics and the corresponding phenotypic information, and a separate model is developed for each omics, and then these models are integrated into the joint model for analysis. This type of method mainly includes Majority-based voting, HierarchicalClassifiers, MOLI (a multi-omics late integration method based on deep neural networks), ATHENA (a formal analysis method), PSDF (Power Spectral Density-basedFault Diagnosis, a fault diagnosis method based on spectral density), MDI (Mean-Shift ClusteringAlgorithm), Lemon-Tree (a decision tree algorithm), etc. This type of method can be used to merge models based on different omics types, and each model can be developed from different sample groups with the same labels.

[0040] The transformation-based integration method first converts each omics dataset into a graph or kernel matrix, and then merges all the datasets into a graph or kernel matrix before building the model. The process is: first, establish a separate correspondence between the omics raw data and phenotypic information for each omics, then develop a separate graph or kernel matrix transformation method for each omics, and then integrate these transformations for single omics into a joint transformation, and use a machine learning algorithm to analyze it. This type of method mainly includes SDP-SVM (semidefinite programming support vector machine), FSMKL (feature set multi-kernel learning), RVM (relevance vector machine), Graphsharpening (image sharpening), PAMOGK (incomplete multi-kernel clustering analysis), NEMO (a moving object detector), etc. In this type of method, if there is some unique information in the dataset (such as sample feature number), it can be used to analyze more dimensions of omics information.

[0041] However, the integration method based on series cannot obtain the effective correlation between omics features, and because it ignores the large differences in the number of features and data distribution between different omics, the results often have large deviations. Deep learning-based technologies, such as Autoencoder (i.e., an unsupervised learning algorithm), require a large number of samples in order to better learn the features of multi-omics. In actual multi-omics applications, there are not too many samples, resulting in unsatisfactory results of the Autoencoder method. Subsequently, unsupervised multi-omics fusion methods emerged, such as I-Cluster (i.e., a hierarchical clustering algorithm), Similarity Network Fusion (SNF), MAUI (i.e., a large-scale automated omics information integration analysis tool) and Multi-Omics Factor Analysis (MOFA+). Although some accuracy improvements have been achieved, only the multi-omics data of the observed object are unsupervisedly learned to obtain a one-dimensional similarity network of the observed object and then analyzed, ignoring the multi-dimensional information of the observed object itself, resulting in poor results in the application process. In addition, most of the above data integration methods consider fusing multiple omics data with similar data structures, but are unable to fuse phenotypic data that are significantly different from omics data structures.

[0042] Based on this, the present invention proposes a multi-source data fusion analysis method based on graph convolutional networks, which can integrate multi-omics data and phenotypic data of the observed object, and can achieve data fusion under the premise of retaining the scale significance of each single data. The present invention also adopts a representation learning method based on a fusion network, which can capture shared complementary information from different data sources, deeply explore the intrinsic relationship between different omics data, and have good robustness to data noise and heterogeneity between data.

[0043] The present invention is described in detail below in conjunction with various embodiments.

[0044] Embodiment 1

[0045] According to an embodiment of the present invention, an embodiment of a method for fusing multi-source data is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0046] Figure 1 is a flowchart of an optional multi-source data fusion method according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0047] Step S101 , collecting a variety of omics data and phenotypic data of an observed object, and preprocessing the various omics data and phenotypic data to obtain a variety of target omics data and target phenotypic data.

[0048] Step S102 : constructing similarity matrices for the various target omics data and target phenotypic data respectively.

[0049] Step S103: fuse all similarity matrices to generate a fusion network.

[0050] Step S104, establishing an attribute graph of the observed object based on the fusion network, multiple target omics data, and target phenotypic data.

[0051] Through the above steps, a variety of omics data and phenotypic data of the observed object can be collected, and the various omics data and phenotypic data are preprocessed to obtain a variety of target omics data and target phenotypic data, and similarity matrices are respectively constructed for the various target omics data and target phenotypic data, and all similarity matrices are fused to generate a fusion network. Based on the fusion network, the various target omics data and the target phenotypic data, an attribute graph of the observed object is established. In an embodiment of the present invention, the various omics data and phenotypic data collected can be preprocessed first, and then similarity matrices are respectively constructed for the various target omics data and the target phenotypic data that have been processed, and then all similarity matrices are fused to generate a fusion network, and then the attribute graph of the observed object is established according to the fusion network, the various target omics data and the target phenotypic data, and the fusion of the omics data and the phenotypic data with large data structure differences can be achieved under the premise of retaining the scale meaning of each single data, so as to obtain the attribute graph of the observed object, thereby solving the technical problem that the omics data and the phenotypic data with large data structure differences cannot be effectively fused in the related art, resulting in a large deviation in the fusion result.

[0052] The embodiment of the present invention is described in detail below in combination with the above steps.

[0053] Step S101 , collecting a variety of omics data and phenotypic data of an observed object, and preprocessing the various omics data and phenotypic data to obtain a variety of target omics data and target phenotypic data.

[0054] In an embodiment of the present invention, multi-omics data information and phenotypic data (i.e., collecting various omics data and phenotypic data of the observed object) of the observed object (i.e., various organisms, for example, microorganisms, etc.) can be collected and sorted first. Among them, multi-omics data include, but are not limited to: genomics, transcriptomics, proteomics, metabolomics and other omics data. Phenotypic data include: quantitative data (such as weight, height, etc.), qualitative data (such as behavior), photos, clinical feature descriptions, medical history, physical examinations, and biochemical examinations, etc. Then, the collected multi-omics data can be preprocessed and converted, including standardization, normalization, feature selection and missing value processing of the omics data, and data preprocessing of the phenotypic data, including encoding conversion, data normalization, etc. (i.e., preprocessing of various omics data and phenotypic data to obtain various target omics data and target phenotypic data).

[0055] Optionally, the steps of preprocessing multiple omics data and phenotypic data to obtain multiple target omics data and target phenotypic data include: determining the omics characteristics of each sample in the omics data and the omics characteristic values ​​corresponding to the omics characteristics; performing missing value processing on the omics data based on the omics characteristics and the omics characteristic values ​​corresponding to the omics characteristics to obtain initial omics data; performing logarithmic processing on the omics characteristic values ​​of each sample in the initial omics data to obtain initial omics characteristic values; normalizing the initial omics characteristic values ​​to obtain target omics characteristic values, wherein the samples indicated by the target omics characteristic values ​​constitute the target omics data; and performing preset embedding mapping on the phenotypic data to obtain target phenotypic data.

[0056] In an embodiment of the present invention, the omics features of each sample in the omics data and the omics feature values ​​corresponding to the omics features can be first determined (that is, each sample in each type of omics data has multiple omics features (for example, features such as accurate values ​​and mean values), and each omics feature has a corresponding omics feature value in the sample). Then, according to the omics features and the omics feature values ​​corresponding to the omics features, the omics data is processed for missing values ​​to obtain initial omics data. Afterwards, the omics feature values ​​of each sample in the initial omics data can be logarithmically processed to obtain the initial omics feature values. For example, all numerical values ​​of the omics features are log2(x+1) transformed, where x represents the original omics feature value. This operation can narrow the range of the data, thereby making the omics feature data more stable. The initial omics feature values ​​are then normalized to obtain the target omics feature values. For example, in order to ensure that the omics feature values ​​are within a certain range, all omics feature values ​​are normalized and standardized. The Z-score (i.e., standardized score) standardization method can be used to normalize the omics feature values. The Z-score standardization formula is: μ is the mean of x, σ is the variance of x, and z is taken as the new omics feature value (i.e., the target omics feature value) after processing the original values ​​of all omics features x. Then, the samples indicated by the target omics feature value constitute the target omics data. In addition, the text data in the phenotypic data can be subjected to an embedding mapping based on a transformer (a neural network model) (i.e., a preset embedding mapping is performed on the phenotypic data) to obtain the target phenotypic data.

[0057] Optionally, based on the omics features and the omics feature values ​​corresponding to the omics features, the step of performing missing value processing on the omics data to obtain initial omics data includes: determining a target omics feature whose missing ratio of omics feature values ​​in all samples is greater than a first preset ratio threshold; determining a target sample whose missing ratio of all omics feature values ​​is greater than a second preset ratio threshold; deleting the target omics feature of each sample in the omics data, and deleting the target sample in the omics data; using a preset mean algorithm to fill the remaining missing values ​​of each sample in the deleted omics data to obtain the initial omics data, wherein the remaining missing values ​​refer to the missing values ​​of the omics features of each sample in the deleted omics data.

[0058] In the embodiment of the present invention, since there are still some missing values ​​in the omics features, and these missing values ​​have a great impact on the results of the entire process, it is necessary to process the missing values ​​of the omics data, specifically: the target omics features whose missing ratio of omics feature values ​​in all samples is greater than the first preset ratio threshold (for example, 20%) can be determined, and the target samples whose missing ratio of all omics feature values ​​is greater than the second preset ratio threshold (for example, 20%) can be determined, and then the target omics features of each sample in the omics data are deleted, and the target samples in the omics data are deleted. For example, if the missing ratio of a certain omics feature in all sample expression values ​​is greater than 20%, the omics feature is directly deleted, and when the missing ratio of all omics features of a certain sample is greater than 20%, the sample is deleted. Afterwards, for the retained features and samples, the remaining missing values ​​can be filled by the k nearest neighbor mean (that is, the preset mean algorithm is used to fill the remaining missing values ​​of each sample in the deleted omics data to obtain the initial omics data, and the remaining missing values ​​refer to the missing values ​​of the omics features of each sample in the deleted omics data).

[0059] Figure 2 is a flow chart of optional data collection and data preprocessing according to an embodiment of the present invention, such as Figure 2 As shown, the following process is included:

[0060] (1) Collect multi-omics and phenotypic data of the subjects;

[0061] (2) missing value processing, logarithmic transformation, standardization and feature selection were performed on the collected data;

[0062] (3) Integrate multi-omics data using similarity network fusion algorithm;

[0063] (4) Use the fused similarity matrix and multi-omics splicing data to construct the attribute graph of the observed object.

[0064] Step S102 : constructing similarity matrices for the various target omics data and target phenotypic data respectively.

[0065] Optionally, the step of constructing a similarity matrix for multiple target omics data and target phenotypic data respectively includes: determining a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data, wherein the observed data is the target omics data or the target phenotypic data, the first sample is any sample in the observed data, and the second sample is any sample in the observed data except the first sample; determining a similarity measurement index value between the first sample and the second sample based on a sample distance, a first sample diameter, a second sample diameter, a first weighting parameter, and a second weighting parameter between the first sample and the second sample; constructing a fully connected similarity matrix based on all similarity measurement index values; and constructing a similarity matrix of the observed data based on the fully connected similarity matrix.

[0066] In an embodiment of the present invention, similarity matrices can be constructed for multiple omics data and phenotypic data respectively (i.e., similarity matrices can be constructed for multiple target omics data and target phenotypic data respectively), and then the multiple networks (i.e., multiple similarity matrices) can be fused using a similarity network fusion algorithm.

[0067] In this embodiment, a single similarity matrix t=1,2,…,T, T represents the number of omics, i represents sample i, j represents sample j, and N represents the number of samples. The constructed relationship is as follows:

[0068]

[0069] Among them, N k (i) represents the k nearest neighbors of sample i; ε represents a very small value, which can be set to 0 as an optional implementation of the present application. When ε = 0, formula (1) can be interpreted as only when sample j is the nearest neighbor of sample i, their edges will be retained, and the remaining edges will be "corrected", so the obtained More robust.

[0070] is the fully connected similarity matrix.

[0071] In this embodiment, the fully connected similarity matrix S (t) Normalize the matrix by rows get, It is a similarity metric. The smaller the distance between samples i and j, The larger the value, the greater the similarity. The Gaussian kernel function is used to calculate the sample i and sample j. The calculation relationship is as follows:

[0072]

[0073] in, is the local diameter of sample i, is the local diameter of sample j, is the distance between sample j and sample i, α and β are weighting parameters, and as an optional implementation of the present application, they can be set to 1 / 6.

[0074] In this embodiment, the local diameter of the sample is obtained by the following relationship:

[0075]

[0076] In this embodiment, the distance between samples It can be calculated in many ways. As an optional implementation of the present application, Euclidean distance calculation can be used.

[0077] In this embodiment, the first sample diameter of the first sample (i.e., any sample in the observation data, such as sample i) and the second sample diameter of the second sample (i.e., any sample in the observation data except the first sample, such as sample j) in the observation data (i.e., target omics data or target phenotypic data) can be determined first, and then the sample diameter of the first sample and the second sample can be determined according to the sample distance between the first sample and the second sample (e.g., ), the first sample diameter (such as ), the second sample diameter (such as ), a first weighting parameter (such as α) and a second weighting parameter (such as β), determine a similarity measurement index value (such as ) (i.e., the similarity metric value can be calculated by formula (2) (3)). Then, based on all similarity metric values, a fully connected similarity matrix (such as S (t) ), then construct the similarity matrix of the observed data based on the fully connected similarity matrix (such as W (t) )(that is, the similarity matrix can be obtained through formula (1).

[0078] Optionally, the step of determining a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data includes: determining a first neighbor sample set of the first sample; determining a first sample distance between the first sample and each first neighbor sample in the first neighbor sample set, and determining the first sample diameter of the first sample based on all first sample distances; determining a second neighbor sample set of the second sample; determining a second sample distance between the second sample and each second neighbor sample in the second neighbor sample set, and determining the second sample diameter of the second sample based on all second sample distances.

[0079] In the embodiment of the present invention, the local diameter of the sample (i.e., the sample diameter) can be obtained by formula (3), specifically: the neighboring sample set of the sample can be determined first (e.g., ) (i.e., determining the first nearest neighbor sample set of the first sample, determining the second nearest neighbor sample set of the second sample), and then determining the sample distance between the sample and each nearest neighbor sample in the nearest neighbor sample set (e.g., ) (i.e., determining the first sample distance between the first sample and each first neighbor sample in the first neighbor sample set, and determining the second sample distance between the second sample and each second neighbor sample in the second neighbor sample set), and then determining the sample diameter of the sample based on all sample distances (e.g., ) (i.e., based on all first sample distances, a first sample diameter of the first sample is determined, and based on all second sample distances, a second sample diameter of the second sample is determined).

[0080] Step S103: fuse all similarity matrices to generate a fusion network.

[0081] In an embodiment of the present invention, after constructing the similarity matrices of all samples in a single-omics dimension respectively, multiple similarity matrices can be further fused through a similarity network fusion algorithm to obtain a sample similarity matrix (i.e., a fusion network) containing information of each omics data (i.e., all similarity matrices are fused to generate a fusion network).

[0082] In an embodiment of the present invention, the similarity network fusion algorithm utilizes a method of constructing a matrix using multi-omics data to transform and then integrate the detection information matrices of different omics, which can accelerate the calculation process of the research and provide valuable information.

[0083] Optionally, the step of fusing all similarity matrices to generate a fused network includes: determining the total information corresponding to each similarity matrix and the information weight corresponding to each similarity matrix; and generating a fused network based on the total information corresponding to all similarity matrices and the information weight.

[0084] In the embodiment of the present invention, when different omics association matrices are fused (i.e., all similarity matrices are fused), When it represents a fully connected or non-bipartite graph, random walks can be used to integrate information from different omics, and it is believed that a stable state can be reached after two random walks (r=2).

[0085] In the embodiment of the present invention, W (t) Represents the total information of omics t after integrating other omics information, where represents the sum of the information of (T-1) groups except group t, k represents any group except group t, ω k represents the information weight of omics k. W integrating other omics information is obtained by random walk (t) . (t)The multi-omics similarity matrix W (i.e., fusion network) is obtained by adding the weights (i.e., the total information W corresponding to each similarity matrix can be determined first (t) And the information weight ω corresponding to each similarity matrix t , and then generate the fusion network W) according to the total information and information weight corresponding to all similarity matrices. The formula is as follows:

[0086]

[0087]

[0088] Among them, α represents the weighting parameter.

[0089] Step S104, establishing an attribute graph of the observed object based on the fusion network, multiple target omics data, and target phenotypic data.

[0090] Optionally, the step of establishing an attribute graph of the observed object based on a fusion network, multiple target omics data and target phenotypic data includes: determining a data feature value corresponding to each data feature based on multiple target omics data and target phenotypic data; concatenating all data feature values ​​corresponding to the data feature to obtain a feature attribute of the data feature; determining a data node corresponding to the data feature based on the fusion network; and associating the feature attributes as node attributes of the data node to generate an attribute graph.

[0091] In an embodiment of the present invention, an attribute graph of the observed object can be established according to the obtained fusion network, and the filtered data features can be spliced ​​as the attributes of each node in the attribute graph (that is, the attribute graph of the observed object is established based on the fusion network, multiple target omics data and target phenotypic data). Specifically, the data feature value corresponding to each data feature can be determined according to the multiple target omics data and the target phenotypic data, and then all the data feature values ​​corresponding to the data feature are spliced ​​to obtain the characteristic attribute of the data feature (that is, all the data feature values ​​corresponding to the data feature are spliced ​​to obtain the characteristic attribute of the data feature), and then based on the fusion network, the data node corresponding to the data feature is determined (that is, each data node on the attribute graph represents a data feature), and then the characteristic attribute is associated with the node attribute of the data node (that is, the characteristic attribute is used as the node attribute of the corresponding data node) to generate an attribute graph.

[0092] Figure 3 is a flowchart of an optional similarity matrix calculation and matrix fusion according to an embodiment of the present invention, such as Figure 3 As shown, the following process is included:

[0093] (1) Construct similarity matrices for multiple omics data sets;

[0094] (2) Use the similarity network fusion algorithm to integrate multiple similarity matrices into a single matrix;

[0095] (3) Construct an undirected graph by fusing similarity matrices;

[0096] (4) Multi-omics data and phenotypic data are spliced ​​as attributes of graph nodes to obtain the observed object attribute graph.

[0097] Optionally, after establishing the attribute graph of the observed object based on the fusion network, multiple target omics data and target phenotypic data, it also includes: using a preset learning model to process the attribute graph to obtain the vector value of each data node, wherein the preset learning model includes: an input layer, a hidden layer, and an output layer; based on all vector values, generating a target attribute graph; based on the target attribute graph, performing task analysis on the observed object.

[0098] In an embodiment of the present invention, after establishing the attribute graph of the observed object, the attribute graph can be subjected to graph neural network representation learning to learn the embedding (i.e., vector value) of each data node (i.e., using a preset learning model (the preset learning model includes: input layer, hidden layer, output layer) to process the attribute graph and obtain the vector value of each data node), and then generate a target attribute graph based on all vector values. The attribute graph after representation learning (i.e., the target attribute graph) can be used for subsequent tasks (i.e., task analysis of the observed object based on the target attribute graph). For example, labeled data can be used for classification tasks, and unlabeled data can be used for clustering tasks.

[0099] Alternatively, before using a preset learning model to process the attribute graph to obtain the vector value of each data node, it also includes: determining a preset learning framework; constructing an initial learning model based on the preset learning framework, wherein the initial learning model includes: an input layer, a hidden layer, and an output layer; training the initial learning model to determine target parameter values ​​of the preset parameters; setting the parameter values ​​of the preset parameters in the initial learning model as the target parameter values ​​to obtain the preset learning model.

[0100] In an embodiment of the present invention, a preset learning framework (i.e., a representation learning framework) may be determined first. The representation learning framework may adopt a framework based on a graph autoencoder. The graph algorithm used in the framework may adopt a GraphSAGE model (i.e., a graph neural network model based on a spatial domain). The architecture of the GraphSAGE model is mainly composed of an input layer, a hidden layer, and an output layer (i.e., based on a preset learning framework, an initial learning model is constructed, and the initial learning model includes: an input layer, a hidden layer, and an output layer). In this embodiment, the input of the input layer is the obtained attribute graph, and the hidden layer is composed of a multi-layer neighbor node feature aggregation layer. As an optional implementation of the present application, the number of layers may be set to two layers, and the aggregation layer may be mainly divided into two parts: random sampling of neighbor nodes and neighbor aggregation.

[0101] In an embodiment of the present invention, the initial learning model can be trained to determine the target parameter values ​​of preset parameters (i.e., predetermined hyperparameters, such as learning rate lr, batch sample size batch_size, number of iterations epoch, etc.), and then the parameter values ​​of the preset parameters in the initial learning model are set to the target parameter values ​​to obtain a trained preset learning model.

[0102] In an embodiment of the present invention, the purpose of designing hyperparameters (including: learning rate lr, batch sample size batch_size, number of iterations epoch, etc.) for the constructed model is to speed up the training speed and converge to the global optimal solution. These parameters affect the convergence speed and performance of the algorithm during the training process. Improper settings will increase the training time. If the learning rate is too small, the convergence speed will be reduced; if the learning rate is too large, the model will not converge. Improper settings of the number of iterations will lead to overfitting, and the batch sample size batch_size will also affect the convergence of the algorithm.

[0103] In the embodiment of the present invention, GraphSAGE can also transform the full-graph training method of GCN (i.e., graph convolutional neural network) into a node-centric small-batch training method through random combination training of subgraphs, so that the model can be trained on large-scale graph data and can make predictions on new graph structures, improving the applicability of industrial scenarios. For neighbor aggregation operations, other methods of centralized extraction of neighbor features can be replaced by convolution operations.

[0104] Optionally, a preset learning model is used to process the attribute graph to obtain a vector value of each data node, including: receiving the attribute graph through an input layer; performing preset neighbor node sampling processing on each data node in the attribute graph through a hidden layer to obtain an aggregate node set for each layer of the data node; performing preset neighbor aggregation processing on the aggregate node set to obtain an aggregate feature value; generating a target aggregate feature value based on all aggregate features; representing the target aggregate feature value as a vector value, and outputting the vector value through an output layer.

[0105] In the embodiment of the present invention, the attribute graph may be received through the input layer, and then the preset neighbor node sampling (i.e., random neighbor node sampling) may be performed on each data node in the attribute graph through the hidden layer to obtain an aggregate node set of each layer of the data node. The random neighbor node sampling process is as follows:

[0106] For the target node v, we can assume that the depth of feature aggregation is K, with B k , 1≤k≤K represents the nodes contained in the neighborhood of node v in the kth layer, N k (B k ) represents the result of random sampling of the k-th layer neighborhood, that is, a certain number of nodes are randomly selected as aggregation nodes in the k-th layer neighborhood. Thus, the aggregation node set of each layer neighborhood of the target node v can be obtained, and the number of neighbors of the node in each layer is different.

[0107] Then, the preset neighbor aggregation (i.e., neighbor aggregation) is performed on the aggregated node set to obtain the aggregated feature value, and then the target aggregated feature value is generated based on all the aggregated features. The process of neighbor aggregation is as follows:

[0108] Based on the obtained randomly sampled nodes, by aggregating their feature representations, the final aggregate representation of the target node can be obtained. The specific learning strategy is as follows:

[0109]

[0110] in, and (k=1,2,3,…,K) is the k-th layer feature extracted by GraphSAGE model feature; is N(u),N(u) ′ The features obtained by aggregation at the t layer through the Aggregate aggregation method; Represents the set of neighbor nodes of node u, N(u) ′ represents the neighbor node set of node u′; CONCAT is feature concatenation; σ l ,σ2 activation function; The weight parameters that the model needs to train.

[0111] In this embodiment, the Aggregate aggregation method generally includes mean aggregation (Mean aggregator), GCN (i.e., graph convolutional neural network) aggregation (GCN aggregator), LSTM (i.e., long short-term memory network) aggregation (LSTMaggregator) and maximum pooling aggregation (Pooling aggregator). As an optional implementation of the present application, the present application adopts two aggregation functions, namely mean aggregation and maximum pooling aggregation.

[0112] Mean aggregation is based on the average of each dimension of the neighborhood nodes in the graph, and the mean is used to express the correlation distribution of the node neighbors. The mean aggregation function is expressed as follows:

[0113]

[0114] in, represents the k-th layer feature of the model, mean is the mean operator, Represents the set of neighbor nodes of node u.

[0115] The maximum pooling aggregation can capture the most prominent performance between neighbor nodes in time and space. After nonlinear transformation of neighbor nodes, the maximum value of each dimension of neighbor nodes is selected. The maximum pooling aggregation function is expressed as follows:

[0116]

[0117] in, represents the k-th layer feature of the model, max is the maximum value operator, W pool represents the pooling weight parameter, b represents the bias parameter, represents the set of neighbor nodes of node u, and σ represents the activation function.

[0118] In this embodiment, the activation function is generally a nonlinear function, which can enable the network to obtain richer expression capabilities. As an optional implementation of the present application, the present application selects the ReLU activation function, which is expressed as follows:

[0119]

[0120] Among them, z represents the variable of the activation function. When z is a negative number, the value of the function is 0, which is used to ensure that the neural network meets a nonlinear activation condition.

[0121] Figure 4 is a schematic diagram of an optional multi-source data fusion process based on a graph convolutional network according to an embodiment of the present invention, such as Figure 4As shown in the figure, multi-source data such as gene expression data, DNA methylation data, copy number variation data, phenotypic data, etc. are preprocessed (Preprocess), and then graph data is obtained through association network fusion (ANF) and concatenation of multi-source data (Concatenated). Then, graph representation learning is performed on the graph data. The learned graph data can be applied to task analysis of downstream applications.

[0122] In this embodiment, microscopic biogenomics data and macroscopic phenotypic data can be fused. In order to test the effectiveness of the scheme, the method can be applied to two different fields of supervised learning and unsupervised learning, respectively. Among them, the accuracy of the supervised learning method based on graph convolution can reach 85.3%, and the clustering results of the graph-based representation learning also have a good performance, which can optimally cluster the data into different categories, and the clustering Rand coefficient reaches 0.7653. As shown in Table 1, in supervised learning applications, the framework proposed in this embodiment has been experimentally proved to be superior to the three indicators of accuracy (Accuracy), F1 (i.e., an indicator for evaluating the performance of binary classifiers) and AUC (Area Under Curve, i.e., area under the curve) in the related technology. Based on AutoEncoder (i.e., self-encoder) fusion, based on SNF (i.e., similarity network) fusion and based on GBDT (i.e., gradient boosting model) fusion.

[0123] Table 1

[0124]

[0125] In an embodiment of the present invention, in order to enable various multi-source data to serve lower-level applications more effectively, a framework for fusing multi-source data into an attribute graph is constructed. The graph convolutional neural network algorithm used in this embodiment can also be replaced by other similar graph algorithms (for example, graph algorithms such as graph convolutional neural networks with self-attention mechanisms).

[0126] In the embodiment of the present invention, multi-source data can be fused efficiently and meaningfully, and the constructed observation object attribute graph can be better applied to the analysis and prediction tasks of downstream observation objects after graph representation learning. The effectiveness of the framework can also be verified in experimental results.

[0127] The following is a detailed description in conjunction with another embodiment.

[0128] Embodiment 2

[0129] A multi-source data fusion device provided in this embodiment includes multiple implementation units, each of which corresponds to each implementation step in the above-mentioned embodiment 1.

[0130] Figure 5is a schematic diagram of an optional multi-source data fusion device according to an embodiment of the present invention, such as Figure 5 As shown, the fusion device may include: a processing unit 50, a construction unit 51, a fusion unit 52, and a building unit 53, wherein:

[0131] A processing unit 50 is used to collect a variety of omics data and phenotypic data of the observed object, and pre-process the various omics data and phenotypic data to obtain a variety of target omics data and target phenotypic data;

[0132] A construction unit 51 is used to construct similarity matrices for a plurality of target omics data and target phenotypic data respectively;

[0133] A fusion unit 52, used for fusing all similarity matrices to generate a fusion network;

[0134] The establishing unit 53 is used to establish a property graph of the observed object based on the fusion network, multiple target omics data and target phenotypic data.

[0135] The above-mentioned fusion device can collect multiple omics data and phenotypic data of the observed object through the processing unit 50, and pre-process the multiple omics data and phenotypic data to obtain multiple target omics data and target phenotypic data, respectively construct similarity matrices for the multiple target omics data and target phenotypic data through the construction unit 51, fuse all similarity matrices through the fusion unit 52 to generate a fusion network, and establish an attribute graph of the observed object based on the fusion network, multiple target omics data and target phenotypic data through the establishment unit 53. In an embodiment of the present invention, the collected various omics data and phenotypic data can be preprocessed first, and then similarity matrices can be constructed for the processed various target omics data and target phenotypic data respectively, and then all similarity matrices can be fused to generate a fusion network. Thereafter, an attribute graph of the observed object is established based on the fusion network, the various target omics data and the target phenotypic data. This can achieve the fusion of omics data and phenotypic data with large data structure differences while retaining the scale meaning of each single data, so as to obtain the attribute graph of the observed object, thereby solving the technical problem in the related art that omics data and phenotypic data with large data structure differences cannot be effectively fused, resulting in large deviations in the fusion results.

[0136] Optionally, the processing unit includes: a first determination module, used to determine the omics features of each sample in the omics data and the omics feature values ​​corresponding to the omics features; a first processing module, used to perform missing value processing on the omics data based on the omics features and the omics feature values ​​corresponding to the omics features to obtain initial omics data; a second processing module, used to perform logarithmic processing on the omics feature values ​​of each sample in the initial omics data to obtain initial omics feature values; a third processing module, used to normalize the initial omics feature values ​​to obtain target omics feature values, wherein the samples indicated by the target omics feature values ​​constitute the target omics data; a first mapping module, used to perform preset embedded mapping on the phenotypic data to obtain target phenotypic data.

[0137] Optionally, the first processing module includes: a first determination submodule, used to determine a target omics feature whose missing ratio of omics feature values ​​in all samples is greater than a first preset ratio threshold; a second determination submodule, used to determine a target sample whose missing ratio of all omics feature values ​​is greater than a second preset ratio threshold; a first deletion submodule, used to delete the target omics feature of each sample in the omics data, and delete the target sample in the omics data; a first filling submodule, used to use a preset mean algorithm to fill the remaining missing values ​​of each sample in the deleted omics data to obtain initial omics data, wherein the remaining missing values ​​refer to the missing values ​​of the omics features of each sample in the deleted omics data.

[0138] Optionally, the construction unit includes: a second determination module, used to determine a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data, wherein the observed data is target omics data or target phenotypic data, the first sample is any sample in the observed data, and the second sample is any sample in the observed data except the first sample; a third determination module, used to determine a similarity measurement index value between the first sample and the second sample based on a sample distance, a first sample diameter, a second sample diameter, a first weighting parameter, and a second weighting parameter between the first sample and the second sample; a first construction module, used to construct a fully connected similarity matrix based on all similarity measurement index values; and a second construction module, used to construct a similarity matrix of the observed data based on the fully connected similarity matrix.

[0139] Optionally, the second determination module includes: a third determination submodule, used to determine a first neighbor sample set of the first sample; a fourth determination submodule, used to determine a first sample distance between the first sample and each first neighbor sample in the first neighbor sample set, and determine a first sample diameter of the first sample based on all first sample distances; a fifth determination submodule, used to determine a second neighbor sample set of the second sample; and a sixth determination submodule, used to determine a second sample distance between the second sample and each second neighbor sample in the second neighbor sample set, and determine a second sample diameter of the second sample based on all second sample distances.

[0140] Optionally, the fusion unit includes: a fourth determination module, used to determine the total information corresponding to each similarity matrix and the information weight corresponding to each similarity matrix; a first generation module, used to generate a fusion network based on the total information and information weights corresponding to all similarity matrices.

[0141] Optionally, the establishment unit includes: a fifth determination module, used to determine the data feature value corresponding to each data feature based on multiple target omics data and target phenotypic data; a first splicing module, used to splice all data feature values ​​corresponding to the data feature to obtain the characteristic attributes of the data feature; a sixth determination module, used to determine the data node corresponding to the data feature based on the fusion network; a second generation module, used to associate the characteristic attributes as node attributes of the data node to generate an attribute graph.

[0142] Optionally, the fusion device also includes: a fourth processing module, which is used to process the attribute graph using a preset learning model after establishing the attribute graph of the observed object based on the fusion network, multiple target omics data and target phenotypic data, so as to obtain the vector value of each data node, wherein the preset learning model includes: an input layer, a hidden layer, and an output layer; a third generation module, which is used to generate a target attribute graph based on all vector values; and a first analysis module, which is used to perform task analysis on the observed object based on the target attribute graph.

[0143] Optionally, the fourth processing module includes: a first receiving submodule, used to receive the attribute graph through the input layer; a first processing submodule, used to perform preset neighbor node sampling processing on each data node in the attribute graph through the hidden layer to obtain an aggregate node set of each layer of the data node; a second processing submodule, used to perform preset neighbor aggregation processing on the aggregate node set to obtain an aggregate feature value; a first generating submodule, used to generate a target aggregate feature value based on all aggregate features; a first output submodule, used to represent the target aggregate feature value as a vector value, and output the vector value through the output layer.

[0144] The above-mentioned fusion device may also include a processor and a memory. The above-mentioned processing unit 50, construction unit 51, fusion unit 52, establishment unit 53, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0145] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the attribute graph of the observed object is established based on the fusion network, multiple target omics data and target phenotypic data by adjusting the kernel parameters.

[0146] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one storage chip.

[0147] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes the following method steps: collecting multiple omics data and phenotypic data of an observed object, and preprocessing the multiple omics data and phenotypic data to obtain multiple target omics data and target phenotypic data, respectively constructing similarity matrices for the multiple target omics data and target phenotypic data, fusing all similarity matrices to generate a fusion network, and establishing an attribute graph of the observed object based on the fusion network, the multiple target omics data and the target phenotypic data.

[0148] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned multi-source data fusion method.

[0149] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement the above-mentioned multi-source data fusion method.

[0150] Figure 6 1 is a hardware structure block diagram of an electronic device (or mobile device) for a multi-source data fusion method according to an embodiment of the present invention. Figure 6 As shown, the electronic device may include one or more ( Figure 6602a, 602b, ..., 602n are used to illustrate) processor 602 (processor 602 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), and memory 604 for storing data. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a keyboard, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 6 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 6 More or fewer components as shown, or with Figure 6 Different configurations are shown.

[0151] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0152] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0154] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0155] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0156] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0157] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for fusing multi-source data, characterized in that: include: Collecting a variety of omics data and phenotypic data of the observed object, and preprocessing the various omics data and phenotypic data to obtain a variety of target omics data and target phenotypic data; constructing similarity matrices for the target omics data and the target phenotypic data respectively; Fusion of all similarity matrices to generate a fusion network; Based on the fusion network, the multiple target omics data and the target phenotypic data, a property graph of the observed object is established.

2. The fusion method according to claim 1, characterized in that: The step of preprocessing the plurality of omics data and the phenotypic data to obtain a plurality of target omics data and target phenotypic data comprises: Determining the omics feature of each sample in the omics data and the omics feature value corresponding to the omics feature; Based on the omics features and the omics feature values ​​corresponding to the omics features, performing missing value processing on the omics data to obtain initial omics data; Performing logarithmic processing on the omics feature value of each sample in the initial omics data to obtain an initial omics feature value; Normalizing the initial omics feature value to obtain a target omics feature value, wherein the sample indicated by the target omics feature value constitutes the target omics data; The phenotypic data is subjected to a preset embedding mapping to obtain the target phenotypic data.

3. The fusion method according to claim 2, characterized in that: The step of performing missing value processing on the omics data based on the omics features and the omics feature values ​​corresponding to the omics features to obtain initial omics data includes: Determine a target omics feature whose missing ratio of the omics feature value in all the samples is greater than a first preset ratio threshold; Determine the target sample whose missing ratio of all omics feature values ​​is greater than a second preset ratio threshold; Deleting the target omics feature of each of the samples in the omics data, and deleting the target sample in the omics data; A preset mean algorithm is used to fill the remaining missing values ​​of each of the samples in the omics data after deletion to obtain the initial omics data, wherein the remaining missing values ​​refer to the missing values ​​of the omics features of each of the samples in the omics data after deletion.

4. The fusion method according to claim 1, characterized in that: The steps of constructing similarity matrices for the target omics data and the target phenotypic data respectively include: Determine a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data, wherein the observed data is the target omics data or the target phenotypic data, the first sample is any sample in the observed data, and the second sample is any sample in the observed data except the first sample; Determine a similarity measurement index value between the first sample and the second sample based on a sample distance between the first sample and the second sample, the first sample diameter, the second sample diameter, a first weighting parameter, and a second weighting parameter; Based on all the similarity measurement index values, construct a fully connected similarity matrix; Based on the fully connected similarity matrix, the similarity matrix of the observation data is constructed.

5. The fusion method according to claim 4, characterized in that: The step of determining a first sample diameter of a first sample and a second sample diameter of a second sample in the observed data comprises: Determine a first neighbor sample set of the first sample; Determine a first sample distance between the first sample and each first neighbor sample in the first neighbor sample set, and determine a first sample diameter of the first sample based on all the first sample distances; Determine a second neighbor sample set of the second sample; A second sample distance between the second sample and each second neighbor sample in the second neighbor sample set is determined, and a second sample diameter of the second sample is determined based on all the second sample distances.

6. The fusion method according to claim 1, characterized in that: The step of fusing all the similarity matrices to generate a fusion network includes: Determine the total information corresponding to each of the similarity matrices and the information weight corresponding to each of the similarity matrices; The fusion network is generated based on the total information corresponding to all the similarity matrices and the information weights.

7. The fusion method according to claim 1, characterized in that: The step of establishing a property graph of the observed object based on the fusion network, the plurality of target omics data and the target phenotypic data comprises: Based on the plurality of target omics data and the target phenotypic data, determining a data feature value corresponding to each data feature; Concatenate all the data feature values ​​corresponding to the data feature to obtain a feature attribute of the data feature; Based on the fusion network, determining a data node corresponding to the data feature; The characteristic attributes are associated with the node attributes of the data nodes to generate the attribute graph.

8. The fusion method according to claim 1, characterized in that: After establishing the property graph of the observed object based on the fusion network, the plurality of target omics data and the target phenotype data, the method further includes: Using a preset learning model to process the attribute graph to obtain a vector value of each data node, wherein the preset learning model includes: an input layer, a hidden layer, and an output layer; Based on all the vector values, generating a target attribute graph; Based on the target attribute graph, task analysis is performed on the observed object.

9. The fusion method according to claim 8, characterized in that: The step of using a preset learning model to process the attribute graph to obtain a vector value of each data node includes: Receiving the attribute graph through the input layer; Perform preset neighbor node sampling processing on each data node in the attribute graph through the hidden layer to obtain an aggregate node set in each layer of the data node; Performing preset neighbor aggregation processing on the aggregation node set to obtain an aggregation feature value; Based on all the aggregate features, generating a target aggregate feature value; The target aggregate feature value is represented as the vector value, and the vector value is output through the output layer.

10. A multi-source data fusion device, characterized in that: include: A processing unit, used for collecting a variety of omics data and phenotypic data of the observed object, and preprocessing the various omics data and phenotypic data to obtain a variety of target omics data and target phenotypic data; A construction unit, used to construct similarity matrices for the target omics data and the target phenotypic data respectively; A fusion unit, used for fusing all the similarity matrices to generate a fusion network; An establishing unit is used to establish a property graph of the observed object based on the fusion network, the multiple target omics data and the target phenotypic data.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the multi-source data fusion method according to any one of claims 1 to 9.

12. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the multi-source data fusion method described in any one of claims 1 to 9.

Images