Quantitative analysis system, methods and applications based on dynamic metabolic fingerprinting

Abstract

This invention discloses a quantitative analysis system, method, and application based on dynamic metabolic fingerprinting, relating to the field of medical artificial intelligence. The system includes a dynamic metabolic fingerprinting extraction module and a deep learning quantification module. The dynamic metabolic fingerprinting extraction module integrates the metabolic fingerprints of collected body fluid samples from cancer patients into a dynamic metabolic fingerprint, where n is the total number of cancer patients, t is the total number of sampling time points, and m is the total number of metabolic biomarkers. The deep learning quantification module is connected to the dynamic metabolic fingerprinting extraction module and performs deep learning on the dynamic metabolic fingerprint, establishing a deep learning model to identify cancer-related metabolic biomarkers. This invention integrates multiple static sequence snapshots to establish a deep learning model that learns the temporal nature of tumor cell growth, enabling rapid identification of cancer prognosis and cancer-related metabolic biomarkers, providing a better solution for cancer prognosis and biomarker screening.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence, specifically to a quantitative analysis system, method, and application based on dynamic metabolic fingerprinting. Background Technology

[0002] Omics research characterizes target biological systems in a high-throughput and task-oriented manner. Compared to genomics and proteomics, metabolomics measures downstream products of biological processes and is more relevant to disease phenotypes. Although metabolomics has made significant progress in areas such as cancer diagnosis and biomarker discovery, most metabolic analyses are based on real-time diagnosis using a single static snapshot, limiting its application in dynamic cellular activities (e.g., prognosis of tumor cells).

[0003] Therefore, those skilled in the art are dedicated to developing a quantitative analysis system and method based on dynamic metabolic fingerprinting to address the problem of its limited application in dynamic cell activity (e.g., prognosis of tumor cells). Summary of the Invention

[0004] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is to provide a quantitative analysis system and method based on dynamic metabolic fingerprinting to solve the problem of its limited application in dynamic cell activity (e.g., prognosis of tumor cells).

[0005] To achieve the above-mentioned technical objectives, the present invention mainly adopts the following technical solutions:

[0006] In the first aspect, this application provides a quantitative analysis system based on dynamic metabolic fingerprinting, including a dynamic metabolic fingerprinting extraction module and a deep learning quantization module.

[0007] The dynamic metabolic fingerprint extraction module integrates the metabolic fingerprints of collected body fluid samples from cancer patients into a dynamic metabolic fingerprint based on the sampling time. n represents the total number of cancer patients, t represents the total number of sampling time points, and m represents the total number of metabolic markers.

[0008] The deep learning quantization module is connected to the dynamic metabolic fingerprint extraction module, and performs deep learning on the dynamic metabolic fingerprint to establish a deep learning model and identify cancer-related metabolic biomarkers.

[0009] In a preferred embodiment of the present invention, the deep learning quantization module includes a data augmentation module, a time-series feature processing module, and a quantitative analysis module. The input end of the data augmentation module is connected to the output end of the dynamic metabolic fingerprint extraction module, the output end of the data augmentation module is connected to the input end of the time-series feature processing module, and the output end of the time-series feature processing module is connected to the input end of the quantitative analysis module.

[0010] In another preferred embodiment of the present invention, the data augmentation module performs data augmentation on the dynamic metabolic fingerprint spectrum input by the dynamic metabolic fingerprint spectrum extraction module; the temporal feature processing module performs temporal analysis on the dynamic metabolic fingerprint spectrum and extracts temporal features for subsequent modeling; the quantitative analysis module uses the output of the temporal feature processing module as input and employs a quantitative analysis network to calculate the probability of cancer recurrence.

[0011] In another preferred embodiment of the present invention, the data augmentation module performs data augmentation on the dynamic metabolic fingerprint spectrum input by the dynamic metabolic fingerprint spectrum extraction module, specifically including the following steps: by using a random number of masks and a random mask probability, all metabolic fingerprint spectra at random time points are set to zero, thereby achieving the purpose of amplifying the training data.

[0012] Num auged =n×N mask ×Prob mask

[0013] Num auged N represents the number of samples to be amplified, where n is the original number of samples. mask For the number of random masks, Prob mask represents the probability of a random mask.

[0014] In another preferred embodiment of the present invention, the time-series feature processing module performs time-series analysis on the dynamic metabolic fingerprint and extracts time-series features for subsequent modeling, specifically including the following steps:

[0015] (1) Calculate the query, key, and value vectors using three fully connected networks:

[0016] q i =m i W q ;

[0017] k i =m i W k ;

[0018] v i =m i W v ;

[0019] These represent the parameters of a fully connected network. The metabolic fingerprint represents the metabolic fingerprint at time point i, where d is the number of metabolic biomarkers.

[0020] (2) Calculate the attention matrix:

[0021]

[0022] d k Attn represents the feature dimension after transformation by the self-attention mechanism. ij Let the attention weights be the values ​​at time points i and j.

[0023] (3) Feature update:

[0024]

[0025] o i This represents the metabolic characteristics at the i-th time point after the final update via the self-attention mechanism.

[0026] In a preferred embodiment of the present invention, multiple time-series feature processing modules are stacked together to extract dynamic metabolic features.

[0027] In another preferred embodiment of the present invention, the quantitative analysis network consists of a linear transformation layer and a sigmoid activation layer. The linear transformation layer transforms the output of the time series feature processing module into a 2D vector space of cancer recurrence and non-recurrence, and maps the predicted values ​​of cancer recurrence and non-recurrence to the range of 0-1 through the sigmoid function.

[0028] Secondly, this application provides a quantitative analysis method based on dynamic metabolic fingerprinting, characterized by the following steps:

[0029] S1: Collect metabolic fingerprint profiles from body fluid samples of cancer patients and integrate them into a dynamic metabolic fingerprint profile according to sampling time. n represents the total number of cancer patients, t represents the total number of sampling time points, and m represents the total number of metabolic markers.

[0030] S2: Perform deep learning on the dynamic metabolic fingerprint profile to establish a deep learning model and identify cancer-related metabolic biomarkers.

[0031] Specifically, S1 includes the following steps:

[0032] S91: Collect sequential body fluid samples from cancer patients at different stages of treatment;

[0033] S92: Metabolic fingerprints of body fluid samples from S91 were extracted using matrix-assisted laser desorption / ionization mass spectrometry.

[0034] S93: Integrate metabolic fingerprint profiles into dynamic metabolic fingerprint profiles based on sampling time. n represents the total number of cancer patients, t represents the total number of sampling time points, and m represents the total number of metabolic markers.

[0035] Thirdly, this application also provides the application of the quantitative analysis system based on dynamic metabolic fingerprinting as described in the first aspect in the preparation of a device for cancer detection or identification.

[0036] Compared with the prior art, this application has the following main advantages:

[0037] This invention integrates multiple static sequence snapshots using dynamic metabolic analysis to model and analyze metabolic biomarkers in time series. As an extension of static metabolic analysis, dynamic metabolic analysis, combined with machine learning, explores the temporal nature of tumor cell growth, playing an important role in adjusting and designing cancer treatment plans and improving survival rates.

[0038] This invention establishes a deep learning model that learns the temporal sequence of tumor cell growth to quickly identify cancer prognosis and determine cancer-related metabolic biomarkers, providing a better solution for cancer prognosis and biomarker screening.

[0039] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the structure of a dynamic metabolic fingerprint extraction model and a deep learning quantization model according to a preferred embodiment of the present invention. Detailed Implementation

[0041] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0042] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0043] like Figure 1 As shown, this invention provides a quantitative analysis system based on dynamic metabolic fingerprinting, including a dynamic metabolic fingerprinting extraction module and a deep learning quantization module; wherein, the dynamic metabolic fingerprinting extraction module integrates the metabolic fingerprints of collected body fluid samples from cancer patients into a dynamic metabolic fingerprinting system according to the sampling time. n represents the total number of cancer patients, t represents the total number of sampling time points, and m represents the total number of metabolic markers.

[0044] Specifically, the steps for extracting this dynamic fingerprint are as follows:

[0045] 1) Collect sequential body fluid samples from cancer patients at different stages of treatment;

[0046] 2) Matrix-assisted laser desorption / ionization mass spectrometry (MADS) was used to extract metabolic fingerprints from the body fluid samples in step 1), obtaining specific m / z ratios in the fingerprints and identifying characteristic signal peaks for each type of cancer patient. Generally, during the experiment, each cancer patient was individually numbered, and then fingerprint samples collected from different treatment stages of each cancer patient were compiled and organized to form a metabolic fingerprint dataset.

[0047] 3) Integrate metabolic fingerprint profiles into dynamic metabolic fingerprint profiles based on sampling time. n represents the total number of cancer patients, t represents the total number of sampling time points, and m represents the total number of metabolic markers.

[0048] The deep learning quantization module is connected to the dynamic metabolic fingerprint extraction module, and performs deep learning on the dynamic metabolic fingerprint to establish a deep learning model to identify cancer-related metabolic biomarkers.

[0049] In this embodiment, the deep learning quantization structure includes three modules: a data augmentation module, a temporal feature processing module, and a quantitative analysis module. The input of the data augmentation module is connected to the output of the dynamic metabolic fingerprint extraction module, the output of the data augmentation module is connected to the input of the temporal feature processing module, and the output of the temporal feature processing module is connected to the input of the quantitative analysis module, forming a connected signal path.

[0050] The data augmentation module augments the input dynamic metabolic fingerprint profile by setting all metabolic fingerprint profiles at random time points to zero using a random number of masks and a random mask probability, thereby amplifying the training data.

[0051] Num auged =n×N mask ×Prob mask

[0052] Num auged N represents the number of samples to be amplified, where n is the original number of samples. mask For the number of random masks, Prob mask represents the probability of a random mask.

[0053] The temporal feature processing module employs a self-attention mechanism to perform temporal analysis on dynamic metabolic fingerprints, extracting temporal features for subsequent modeling. The specific steps are as follows:

[0054] 1) Calculate the query, key, and value vectors using three fully connected networks:

[0055] q i =m i Q q ;

[0056] k i =m i W k ;

[0057] v i =m i W v ;

[0058] These represent the parameters of a fully connected network. The denoted time point represents the metabolic fingerprint, where d is the number of metabolic biomarkers.

[0059] 2) Calculate the attention matrix:

[0060]

[0061] d k Attn represents the feature dimension after transformation by the self-attention mechanism. ij Let be the attention weights between the i-th time point and the j-th time point.

[0062] 3) Feature update:

[0063]

[0064] o i This represents the metabolic characteristics at the i-th time point after the final update via the self-attention mechanism.

[0065] It should be noted that the time-series feature processing module can be set to two, three, four, or more. Depending on the actual situation, multiple time-series feature processing modules can be stacked to extract dynamic metabolic features. No further restrictions are imposed here.

[0066] The quantitative analysis module employs a fully connected network for quantitative analysis: taking the output of the time-series feature processing module as input, the quantitative analysis network calculates the probability of cancer recurrence. The quantitative analysis network consists of a linear transformation layer and a sigmoid activation layer. The linear transformation layer transforms the output of the feature processing module into a 2D vector space representing cancer recurrence and non-recurrence, and the sigmoid function maps the predicted values ​​for cancer recurrence and non-recurrence to the range of 0-1.

[0067] The embodiments of this application also provide a method for analysis using the above-described quantitative analysis system based on dynamic metabolic fingerprinting, which mainly includes the following steps:

[0068] S1: Collect sequential body fluid samples from cancer patients at different stages of treatment;

[0069] Metabolic fingerprints of body fluid samples were extracted using matrix-assisted laser desorption / ionization mass spectrometry to obtain specific m / z ratios in the fingerprints and to identify characteristic signals for each type of cancer patient.

[0070] 3) Integrate metabolic fingerprint profiles into dynamic metabolic fingerprint profiles based on sampling time. n represents the total number of cancer patients, t represents the total number of sampling time points, and m represents the total number of metabolic markers.

[0071] The sampling time can be set on the first, third, seventh, fifteenth, and thirtieth days of the treatment process, depending on the patient's actual treatment course and method. No further restrictions are imposed here.

[0072] S2: Perform deep learning on dynamic metabolic fingerprint profiles to establish a deep learning model and identify cancer-related metabolic biomarkers.

[0073] The quantitative analysis system or method based on dynamic metabolic fingerprinting described in this application can integrate multiple static sequence snapshots for dynamic metabolic analysis, enabling modeling and analysis of metabolic biomarkers over time. As an extension of static metabolic analysis, dynamic metabolic analysis, combined with machine learning to mine the temporal nature of tumor cell growth, will play a crucial role in adjusting and designing cancer treatment plans and improving survival rates. It can be applied to cancer detection or identification devices, providing better solutions for rapidly identifying cancer prognosis, determining cancer-related metabolic biomarkers, and advancing cancer prognosis and biomarker screening.

[0074] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A quantitative analysis system based on dynamic metabolic fingerprinting, characterized in that: Includes a dynamic metabolic fingerprint extraction module and a deep learning quantization module; The dynamic metabolic fingerprint extraction module integrates the metabolic fingerprints of collected body fluid samples from cancer patients into a dynamic metabolic fingerprint based on the sampling time. Where n is the total number of cancer patients, t is the total number of sampling time points, and d is the total number of metabolic biomarkers; the metabolic fingerprint extraction steps are as follows: 1) Collect sequential body fluid samples from cancer patients at different stages of treatment; 2) Metabolic fingerprints of body fluid samples from 1) were extracted using matrix-assisted laser desorption / ionization mass spectrometry to obtain specific m / z ratios in the fingerprints and to identify characteristic signal peaks for each type of cancer patient. Each cancer patient was individually numbered, and then the fingerprint samples collected from different treatment stages of each cancer patient were summarized and organized to form a metabolic fingerprint set. 3) Integrate metabolic fingerprint profiles into dynamic metabolic fingerprint profiles based on sampling time. , where n is the total number of cancer patients, t is the total number of sampling time points, and d is the total number of metabolic markers; The deep learning quantization module is connected to the dynamic metabolic fingerprint extraction module, and performs deep learning on the dynamic metabolic fingerprint to establish a deep learning model and identify cancer-related metabolic biomarkers. The deep learning quantization module includes a data augmentation module, a temporal feature processing module, and a quantitative analysis module. The input of the data augmentation module is connected to the output of the dynamic metabolic fingerprint extraction module, the output of the data augmentation module is connected to the input of the temporal feature processing module, and the output of the temporal feature processing module is connected to the input of the quantitative analysis module. The data augmentation module performs data augmentation on the dynamic metabolic fingerprint spectrum input by the dynamic metabolic fingerprint spectrum extraction module; specifically, it includes the following steps: by using a random number of masks and a random mask probability, all metabolic fingerprint spectra at random time points are set to zero, thereby amplifying the training data. The number of samples to be amplified. Original sample size The number of random masks. For random mask probability; The temporal feature processing module employs a self-attention mechanism to perform temporal analysis on the dynamic metabolic fingerprint, extracting temporal features for subsequent modeling; specifically, it includes the following steps: (1) Calculate the query, key, and value vectors using three fully connected networks: , , These represent the parameters of a fully connected network. The metabolic fingerprint represents the metabolic fingerprint at time point i. The number of metabolic markers; (2) Calculate the attention matrix: This represents the feature dimension after transformation by the self-attention mechanism. The attention weights are defined for the i-th and j-th time points; multiple temporal feature processing modules are superimposed to extract dynamic metabolic features. (3) Feature update: This represents the metabolic characteristics at the i-th time point after being updated by the self-attention mechanism. The quantitative analysis module takes the output of the time-series feature processing module as input and uses a quantitative analysis network to calculate the probability of cancer recurrence. The quantitative analysis network consists of a linear transformation layer and a sigmoid activation layer. The linear transformation layer transforms the output of the feature processing module into a 2-dimensional vector space of cancer recurrence and non-recurrence, and maps the predicted values ​​of cancer recurrence and non-recurrence to the range of 0-1 through the sigmoid function.

2. A quantitative analysis method for dynamic metabolic fingerprint profiles based on the quantitative analysis system of claim 1, characterized in that, Includes the following steps: S1: Collect metabolic fingerprint profiles from body fluid samples of cancer patients and integrate them into a dynamic metabolic fingerprint profile according to sampling time. , where n is the total number of cancer patients, t is the total number of sampling time points, and d is the total number of metabolic markers; S2: Perform deep learning on the dynamic metabolic fingerprint profile to establish a deep learning model and identify cancer-related metabolic biomarkers.

3. The application of the quantitative analysis system based on dynamic metabolic fingerprinting as described in claim 1 in the preparation of a device for cancer detection or identification.

Images