Helicobacter pylori detection method based on mass spectrum recognition

Abstract

The invention relates to the technical field of helicobacter pylori detection, and particularly discloses a helicobacter pylori detection method based on mass spectrum recognition, which comprises the following steps: firstly, carrying out MALDI mass spectrum acquisition on a to-be-detected sample to obtain a protein fingerprint spectrum of the to-be-detected sample, and then carrying out pretreatment to remove noise and standardization treatment to obtain a protein fingerprint spectrum of the to-be-detected sample; further extracting to-be-detected feature visual mode features with discriminating ability, and extracting reference visual mode features of each known strain from the constructed strain reference database at the same time; a complex matching relation between a to-be-detected feature and a reference feature is learned by introducing protein fingerprint spectrum mode feature query response analysis based on a feature domain so as to obtain fingerprint spectrum visual mode feature query response coding representation, and a strain detection result is finally determined based on the query response. The strategy based on visual pattern feature extraction and network matching query aims at fully utilizing deep pattern information of mass spectrum data so as to realize more reliable helicobacter pylori mass spectrum identification.

Description

Technical Field

[0001] This application relates to the technical field of Helicobacter pylori detection, and more specifically, to a Helicobacter pylori detection method based on mass spectrometry identification. Background Art

[0002] Helicobacter pylori is a common Gram-negative bacterium that mainly colonizes the human gastric mucosa and is the main pathogenic factor causing various upper gastrointestinal diseases such as chronic gastritis, peptic ulcer, gastric mucosa-associated lymphoid tissue lymphoma, and even gastric cancer. Given its wide infection rate and serious health consequences, developing a rapid, accurate, and reliable Helicobacter pylori detection method is of crucial clinical significance and public health value for the early diagnosis, effective treatment of diseases, and prevention of serious complications such as gastric cancer.

[0003] In recent years, matrix-assisted laser desorption / ionization time-of-flight mass spectrometry (MALDI-TOF MS) technology has been widely used in the field of microbial identification due to its rapidity, high throughput, and relatively low cost. Existing microbial detection schemes based on mass spectrometry identification usually collect the protein fingerprint spectra of the microorganisms to be detected (after culturing or directly enriching from the sample), and then compare the obtained mass spectrometry peak list or the entire spectrum with the reference spectra of known bacterial species stored in the database to determine the species of the microorganisms to be detected. However, when applied to the detection of specific bacterial species such as Helicobacter pylori, these traditional methods based on peak matching or overall spectrum similarity calculation have certain limitations. For example, there may be subtle differences in protein expression between different Helicobacter pylori strains or between them and other symbiotic or pathogenic bacteria, resulting in overall similar mass spectrometry spectra but different key characteristic peaks; at the same time, factors such as sample preparation, culture conditions, and instrument status may also cause spectrum drift and peak intensity fluctuations, which may interfere with the accuracy of traditional comparison algorithms, especially when dealing with low-quality spectra or differentiating closely related bacterial species, easily leading to identification confusion or failure and affecting the sensitivity and specificity of the detection.

[0004] Therefore, an optimized Helicobacter pylori detection scheme based on mass spectrometry identification is desired. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. An embodiment of this application provides a Helicobacter pylori detection method based on mass spectrometry identification, which adopts a strategy based on visual pattern feature extraction and query network matching, aiming to make full use of the deep pattern information of mass spectrometry data to achieve more reliable mass spectrometry identification of Helicobacter pylori.

[0006] According to one aspect of the present application, a method for detecting Helicobacter pylori based on mass spectrometry identification is provided, including: mixing a sample to be tested with a MALDI matrix and spotting it on a MALDI target plate, and collecting a protein fingerprint spectrum using a MALDI-TOF mass spectrometer; extracting a set of protein reference fingerprint spectra of known strain types from a strain reference database; preprocessing the protein fingerprint spectrum to obtain a preprocessed protein fingerprint spectrum; extracting fingerprint spectrum visual pattern features from the preprocessed protein fingerprint spectrum to obtain visual pattern features of the protein fingerprint spectrum to be tested; extracting fingerprint spectrum visual pattern features from the set of protein reference fingerprint spectra of known strain types to obtain a set of reference visual pattern features of the protein fingerprint spectrum; performing a fingerprint spectrum pattern feature query on the visual pattern features of the protein fingerprint spectrum to be tested and the set of reference visual pattern features of the protein fingerprint spectrum to obtain a fingerprint spectrum visual pattern feature query response feature; and determining a strain detection result based on the fingerprint spectrum visual pattern feature query response feature.

[0007] In a possible implementation manner, the preprocessing includes baseline correction, noise smoothing, peak identification, and alignment processing.

[0008] In a possible implementation manner, extracting fingerprint spectrum visual pattern features from the preprocessed protein fingerprint spectrum to obtain visual pattern features of the protein fingerprint spectrum to be tested includes: passing the preprocessed protein fingerprint spectrum through a fingerprint spectrum visual pattern feature extractor based on a dilated convolutional neural network model to obtain an encoded vector of visual pattern features of the protein fingerprint spectrum to be tested as the visual pattern features of the protein fingerprint spectrum to be tested.

[0009] In a possible implementation manner, extracting fingerprint spectrum visual pattern features from the set of protein reference fingerprint spectra of known strain types to obtain a set of reference visual pattern features of the protein fingerprint spectrum includes: passing each protein reference fingerprint spectrum of known strain types in the set of protein reference fingerprint spectra of known strain types through the fingerprint spectrum visual pattern feature extractor based on the dilated convolutional neural network model to obtain a set of encoded vectors of reference visual pattern features of the protein fingerprint spectrum as the set of reference visual pattern features of the protein fingerprint spectrum.

[0010] In a possible implementation, a fingerprint spectrum pattern feature query is performed on the set of the visual pattern features of the protein fingerprint spectrum to be measured and the reference visual pattern features of the protein fingerprint spectrum to obtain a fingerprint spectrum visual pattern feature query response feature, including: after enhancing the feature of the encoded vector of the visual pattern features of the protein fingerprint spectrum to be measured, performing a monomer semantic query encoding on it respectively with each encoded vector of the reference visual pattern features of the protein fingerprint spectrum in the set of the encoded vectors of the reference visual pattern features of the protein fingerprint spectrum to obtain a set of encoded vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured; calculating the monomer semantic weights of each encoded vector of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured in the set of the encoded vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured respectively to obtain a set of self-attention weights of the monomer query semantics of the visual pattern of the protein fingerprint spectrum to be measured; aggregating the set of the encoded vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured based on the set of the self-attention weights of the monomer query semantics of the visual pattern of the protein fingerprint spectrum to be measured to obtain the encoded vector of the fingerprint spectrum visual pattern feature query response.

[0011] In a possible implementation, after enhancing the feature of the encoded vector of the visual pattern features of the protein fingerprint spectrum to be measured, performing a monomer semantic query encoding on it respectively with each encoded vector of the reference visual pattern features of the protein fingerprint spectrum in the set of the encoded vectors of the reference visual pattern features of the protein fingerprint spectrum to obtain a set of encoded vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured, including: performing a feature enhancement based on deconvolution encoding on the encoded vector of the visual pattern features of the protein fingerprint spectrum to be measured to obtain an enhanced encoded vector of the visual pattern features of the protein fingerprint spectrum to be measured, and the enhanced encoded vector of the visual pattern features of the protein fingerprint spectrum to be measured and each encoded vector of the reference visual pattern features of the protein fingerprint spectrum in the set of the encoded vectors of the reference visual pattern features of the protein fingerprint spectrum have the same feature scale; performing a monomer semantic query encoding on the enhanced encoded vector of the visual pattern features of the protein fingerprint spectrum to be measured and each encoded vector of the reference visual pattern features of the protein fingerprint spectrum in the set of the encoded vectors of the reference visual pattern features of the protein fingerprint spectrum respectively to obtain the set of the encoded vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured.

[0012] In a possible implementation, the monomer semantic weights of each of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors in the set of protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured are calculated respectively to obtain a set of self-attention weights of the protein fingerprint spectrum visual pattern monomer query semantics, including: determining the monomer semantic matching degrees of each of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors in the set of protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured based on the self-distribution characteristics of the feature set of the set of protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured to obtain a set of monomer semantic matching degrees of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured; inputting the set of monomer semantic matching degrees of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured into a relational gating proxy module to obtain a set of self-attention weights of the protein fingerprint spectrum visual pattern monomer query semantics.

[0013] In a possible implementation, determining the monomer semantic matching degrees of each of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors in the set of protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured based on the self-distribution characteristics of the feature set of the set of protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured to obtain a set of monomer semantic matching degrees of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured, including: performing inherent alignment calibration on each of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors in the set of protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured to obtain a set of calibrated protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors; calculating the monomer semantic matching degree of each calibrated protein fingerprint spectrum visual pattern monomer semantic query score encoding vector based on the distribution characteristics of the set of calibrated protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to obtain the set of monomer semantic matching degrees of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to be measured.

[0014] In a possible implementation, determining the strain detection result based on the fingerprint spectrum visual pattern feature query response feature, including: passing the fingerprint spectrum visual pattern feature query response encoding vector through a strain detector based on a classifier to obtain a strain detection result, and the strain detection result is used to represent the strain category label.

[0015] The present application provides a method for detecting Helicobacter pylori based on mass spectrometry recognition, which first performs MALDI mass spectrometry acquisition on the sample to be tested to obtain its protein fingerprint spectrum, then removes noise and standardizes it through preprocessing, further extracts the visual pattern features of the tested features with discrimination ability, and extracts the reference visual pattern features of each known strain from the constructed strain reference database, and finally, learns the complex matching relationship between the tested features and the reference features by introducing the protein fingerprint spectrum pattern feature query response analysis based on the feature domain to obtain the fingerprint spectrum visual pattern feature query response encoding representation, and finally determines the strain detection result based on this query response. This strategy based on visual pattern feature extraction and query network matching aims to make full use of the deep pattern information of mass spectrometry data to achieve more reliable Helicobacter pylori mass spectrometry recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0017] Figure 1 This is a schematic flow chart of a Helicobacter pylori detection method based on mass spectrometry recognition according to an embodiment of the present application.

[0018] Figure 2 Schematic diagram of data flow of the Helicobacter pylori detection method based on mass spectrometry recognition according to an embodiment of the present application.

[0019] Figure 3 This is a schematic flow chart of step S6 in the Helicobacter pylori detection method based on mass spectrometry recognition according to an embodiment of the present application.

[0020] Figure 4 This is a schematic flow chart of step S61 in the Helicobacter pylori detection method based on mass spectrometry recognition according to an embodiment of the present application.

[0021] Figure 5 This is a schematic flow chart of step S62 in the Helicobacter pylori detection method based on mass spectrometry recognition according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0023] In view of the above technical problems, in the technical solution of the present application, a method for detecting Helicobacter pylori based on mass spectrometry identification is proposed, aiming to improve the accuracy and robustness of mass spectrometry detection of Helicobacter pylori. The core lies in no longer relying solely on discrete peak positions and peak intensity information, but focusing on extracting fingerprint spectrum visual pattern features from the protein fingerprint spectrum after preprocessing such as baseline correction, noise smoothing, peak identification, and alignment, which can better reflect the overall morphology and structural information of the spectrum. That is to say, the protein fingerprint spectrum, like an image, contains rich pattern information. Extracting its visual pattern features can capture the macroscopic structure of the spectrum and the feature distribution in the key regions. Such features may be more resistant to interference caused by noise and subtle changes than the presence or exact position of a single peak, and thus can more stably represent the identity of the bacterial species. Correspondingly, the same operation is also performed on the known bacterial species reference fingerprint spectra in the bacterial species reference database to extract reference visual pattern features and construct a reference database based on visual pattern features. Further, in order to achieve efficient and accurate matching, this solution learns the complex matching relationship between the to-be-detected features and the reference features by introducing protein fingerprint spectrum pattern feature query response analysis based on the feature domain. This query network can go beyond simple distance calculation or similarity scoring and more intelligently determine which reference bacterial species has the most similar visual pattern to the to-be-detected sample, thereby outputting a fingerprint spectrum visual pattern feature query response code representation, and finally determining the bacterial species detection result based on this query response. This strategy based on visual pattern feature extraction and query network matching aims to make full use of the deep pattern information of mass spectrometry data to achieve more reliable mass spectrometry identification of Helicobacter pylori.

[0024] In view of the above technical problems, in the technical solution of the present application, a method for detecting Helicobacter pylori based on mass spectrometry identification is proposed. Figure 1 It is a schematic flowchart of the method for detecting Helicobacter pylori based on mass spectrometry identification according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of the method for detecting Helicobacter pylori based on mass spectrometry identification according to an embodiment of the present application, as Figure 1 and Figure 2As shown, the Helicobacter pylori detection method based on mass spectrometry identification includes: S1, mixing a sample to be tested with a MALDI matrix and spotting the mixture on a MALDI target plate, and using a MALDI-TOF mass spectrometer to collect a protein fingerprint spectrum; S2, extracting a set of protein reference fingerprint spectra of known strain types from a strain reference database; S3, preprocessing the protein fingerprint spectrum to obtain a preprocessed protein fingerprint spectrum; S4, extracting fingerprint spectrum visual pattern features from the preprocessed protein fingerprint spectrum to obtain visual pattern features of the protein fingerprint spectrum to be tested; S5, extracting fingerprint spectrum visual pattern features from the set of protein reference fingerprint spectra of known strain types to obtain a set of reference visual pattern features of the protein fingerprint spectrum; S6, performing a fingerprint spectrum pattern feature query on the visual pattern features of the protein fingerprint spectrum to be tested and the set of reference visual pattern features of the protein fingerprint spectrum to obtain a fingerprint spectrum visual pattern feature query response feature; S7, determining a strain detection result based on the fingerprint spectrum visual pattern feature query response feature.

[0025] Specifically, in step S1, a sample to be tested is mixed with a MALDI matrix and spotted on a MALDI target plate, and a MALDI-TOF mass spectrometer is used to collect a protein fingerprint spectrum. It should be understood that in proteomic analysis, a MALDI matrix (such as α-cyano-4-hydroxycinnamic acid, etc.) can effectively promote the ionization of polypeptides or proteins in a sample and has the essential functions of suppressing interference and enhancing signals. A biological sample to be detected (such as wet bacteria of Helicobacter pylori cultured) is suspended in a suitable solution at a specified concentration and then thoroughly mixed with a proportionally prepared MALDI matrix to ensure sufficient contact between the protein and the matrix crystals. Subsequently, an appropriate amount of the mixed solution (for example, 1 μL) is dispensed into a designated well of the MALDI target plate and naturally volatilized and dried at room temperature, so that the proteins / peptides are uniformly embedded in the matrix crystals. This solid-phase spotting can significantly improve the signal-to-noise ratio and the reproducibility of the spectrum in subsequent analysis. After the sample preparation is completed, the MALDI target plate is loaded into the MALDI-TOF mass spectrometer. The MALDI-TOF mass spectrometer scans the sample point by point according to preset parameters (such as laser energy, pulse frequency, etc.). The laser pulse irradiates the sample matrix mixture point, and the matrix molecules absorb the excitation energy to desorb and ionize the proteins / peptides embedded therein, obtaining the flight time distribution information of the charged ions. The spectrum is calibrated inside the instrument and recorded by an external data acquisition system, and finally the fingerprint mass spectrum of each sample to be tested is output, typically in the form of a two-dimensional spectrum of m / z (mass-to-charge ratio) and intensity. For example, after specific culture, Helicobacter pylori is collected by centrifugation, diluted, and then mixed with a MALDI matrix solution in a 1:1 ratio. 1 μL of the mixed solution is accurately dropped onto a designated well of the MALDI target plate and naturally dried to form a uniform film. After loading the target plate, the protein fingerprint spectrum is gradually collected according to the preset target points.

[0026] Specifically, in step S2, a set of protein reference fingerprint spectra of known bacterial species types is extracted from the bacterial species reference database. It should be understood that in clinical or actual microbial detection scenarios, Helicobacter pylori not only has diverse strain spectra, but is also easily interfered by multiple factors such as sample processing, culture conditions, and experimental instruments, resulting in a certain degree of spectral drift, peak intensity fluctuations, and even the disappearance and addition of characteristic peaks in the protein fingerprint spectra between different batches or different laboratories. Traditional comparison algorithms based on peak values or simple full-spectrum similarity are difficult to handle the potential complex changes and microdifferences in the protein spectra between Helicobacter pylori and its closely related bacterial species due to their reliance on discrete peak table information and surface feature distributions. Therefore, in this application, the set of protein fingerprint spectra of known bacterial species (including standard strains of all levels of Helicobacter pylori) is standardized and incorporated into the reference database.

[0027] Specifically, in step S3, the protein fingerprint spectrum is preprocessed to obtain a preprocessed protein fingerprint spectrum. It should be understood that since the original MALDI-TOF mass spectrometry data is often interfered by multiple factors, for example, factors such as sample preparation, culture conditions, and instrument status may cause baseline drift in the spectrum, contain a large amount of random noise, exhibit fluctuations in peak intensity, and small offsets in peak positions (i.e., spectral drift). These defects in the original data will seriously interfere with subsequent analysis. Whether it is a traditional peak-matching-based method or a visual pattern feature-based method, relatively pure and standardized data input is required. Specifically, in the technical solution of this application, the preprocessing includes baseline correction, noise smoothing, peak identification, and alignment processing. Baseline correction aims to eliminate the slowly changing background signal caused by the matrix or other non-peptide substances to ensure that the peak height can truly reflect the peptide abundance; noise smoothing aims to reduce the high-frequency random fluctuations in the spectrum, improve the signal-to-noise ratio, and make the true protein peaks easier to be detected; the purpose of peak identification (or peak extraction) is to accurately find the characteristic peaks representing proteins or polypeptide fragments from the smoothed spectrum and determine their mass-to-charge ratio positions and intensity information; and the alignment processing is to correct the systematic peak position offsets caused by subtle instrument differences or calibration problems between different samples or multiple measurements of the same sample to ensure that the peaks of the same biomarker appear at comparable positions in different spectra.

[0028] Specifically, first of all, baseline correction, as the first step in mass spectrometry data processing, aims to eliminate the background interference signals in the spectra. This background signal mainly originates from matrix residues, ion suppression effects, and the accumulation of non-specific ions during the analysis process, which can lead to the appearance of a "false bottom" phenomenon of the spectrum shifting upward or even slowly fluctuating in the low-frequency region. The algorithm selection for baseline correction mainly includes the rolling ball algorithm, TopHat filtering, least squares fitting, etc. The rolling ball algorithm slides along the laser time axis with a specific window size to simulate the lower bound of the background curve, thereby achieving baseline removal. For example, the window range around the m / z spectral line is set to 200 Da, and the local minimum points are fitted to the baseline by simulating the lower edge of the rolling ball closely adhering to the spectral line. Then, the above baseline profile is subtracted from the original signal point by point to restore the true peak height. TopHat filtering suppresses the slowly changing components of the spectrum by constructing a morphological structure element. The least squares method forms a background parabola by polynomial fitting, and then the background function is differentiated from the original spectrum for correction. High-quality baseline correction can ensure that the signal intensity only reflects the true protein peak concentration and effectively suppresses the interference of false signals.

[0029] After completing baseline correction, it is necessary to further perform noise smoothing on the remaining signals. The high-frequency noise in mass spectrometry data mainly manifests as random spikes, noise bands, and periodic microfluctuations. Effective noise smoothing can not only retain the sharpness and contrast of the true peak shape but also eliminate signal fluctuations unrelated to biology. Commonly used noise smoothing methods include moving average, Gaussian filtering, Savitzky-Golay smoothing, etc. Moving average calculates the mean value by sliding a fixed window (such as 5 - 11 points), which is suitable for high-frequency noise reduction but with a slightly reduced resolution. Gaussian filtering weights and averages the original signal with a Gaussian kernel, which can filter out sudden noise while ensuring that the peak shape remains unchanged. Savitzky-Golay filtering fits the signal with a low-order polynomial within a local sliding window, taking into account both noise suppression and preservation of peak shape details. Taking MALDI-TOF proteomics data as an example, the window width can usually be set to 7 - 15 points, and the filtering order is set to 2 - 3 orders. The parameters need to be adjusted according to the specific spectrum quality and expected resolution balance. Through multiple experiments and empirical evaluations, a standardized signal trajectory with obvious peak structures, low noise backgrounds, and natural and smooth peak shapes is finally obtained.

[0030] Subsequently, based on the spectrogram after noise smoothing, peak identification processing needs to be implemented. The purpose is to accurately locate the protein / polypeptide characteristic peaks in the spectrogram and extract their key information parameters (such as m / z value, peak height, peak area, peak width, etc.). The process of peak identification needs to be dynamically and adaptively adjusted in combination with features such as the overall signal-to-noise ratio, local noise level, and peak spacing of the spectrogram. Commonly used methods include threshold determination based on local maxima, peak shape discrimination based on the first derivative and second derivative, and multi-scale detection combined with prior biological information. In specific operations, the spectrogram data points are traversed in a window sliding manner, and the value differences of its adjacent points are calculated for each data point. When this point is higher than the nearest data points on the left and right and reaches the preset signal-to-noise ratio threshold (such as S / N > 3), and its locally fitted peak width and peak area meet the protein peak shape distribution characteristics, then this point can be determined as a legal peak position. For complex samples (such as Helicobacter pylori proteome spectra), signal processing techniques such as breakpoint detection and continuous wave peak analysis can be introduced to merge weak peaks and false peaks in multiple detections and improve the accuracy of peak identification. Peak identification parameters such as minimum peak height, minimum peak width, and minimum peak area need to be flexibly set according to the performance of the mass spectrometer and the specific experimental system.

[0031] After the accurate identification of peaks is completed, for cross-batch sample data or multiple measurements of a single sample, due to ion source energy drift, time base adjustment error, etc., the same protein peak often has a slight shift on the x-axis (m / z dimension). To ensure that the peak information can be effectively compared between spectrograms and between the reference database and the sample to be measured, all spectrograms need to be subjected to peak alignment processing. The methods of peak alignment include global linear mapping, local non-linear registration, and dynamic time normalization based on markers. Linear mapping assumes that the full-spectrum drift is constant, and the m / z deviation of the internal standard peak or strong characteristic peak in different samples is calculated, and simple linear regression is used to assist in calibrating the entire spectral line. For complex drift patterns, non-linear registration can be used. First, a group of co-occurring peaks (i.e., homologous marker peak groups) are identified in each spectrogram, and the dynamic time warping algorithm is used to match the m / z sequences to achieve global and local non-uniform mapping. In addition, existing advanced spectrogram alignment toolkits all provide an automated multi-spectrogram batch alignment function, and interpolation weighting, time domain normalization, waveform hierarchical clustering and other techniques can be used at the bottom layer to ensure the consistency of the corresponding relationship of each identified peak in all samples or all database reference spectrograms.

[0032] In a specific embodiment, after the protein fingerprint spectrum is collected, the original profile-type data is first loaded into professional data processing software, rolling ball baseline correction with a window size of 15 Da is applied, followed by denoising with a 7-point-width Savitzky-Golay filter, effective peaks are identified using maximum value detection with a signal-to-noise ratio threshold of 3, and then 3-5 internal standard peaks are selected for linear / non-linear alignment mapping to obtain the preprocessed protein fingerprint spectrum.

[0033] Specifically, in step S4, fingerprint spectrum visual pattern features are extracted from the preprocessed protein fingerprint spectrum to obtain the visual pattern features of the protein fingerprint spectrum to be measured. It should be understood that since traditional Helicobacter pylori detection methods overly rely on discrete peak positions, peak intensity information, or overall spectrum similarity, they are easily affected by factors such as sample preparation and instrument status, resulting in spectral drift, intensity fluctuations, and subtle differences between strains, leading to insufficient identification accuracy and stability. Based on this, in order not to be limited to these variable information but to explore fingerprint spectrum visual pattern features that can better reflect the stable internal structure of the spectrum, in the technical solution of this application, fingerprint spectrum visual pattern features are further extracted from the preprocessed protein fingerprint spectrum to obtain the visual pattern feature coding vector of the protein fingerprint spectrum to be measured. Such fingerprint spectrum visual pattern features are considered to contain rich macroscopic structure and key region distribution information like images, and can better resist noise and small changes, thus more stably representing the species identity. Therefore, by extracting fingerprint spectrum visual pattern features, features that can capture deep structural information such as the overall morphology, peak distribution pattern, and relationship between peaks can be extracted from the preprocessed protein fingerprint spectrum.

[0034] Specifically, in one embodiment, extracting fingerprint spectrum visual pattern features from the preprocessed protein fingerprint spectrum to obtain the visual pattern features of the protein fingerprint spectrum to be measured includes: passing the preprocessed protein fingerprint spectrum through a fingerprint spectrum visual pattern feature extractor based on a dilated convolutional neural network model to obtain the visual pattern feature coding vector of the protein fingerprint spectrum to be measured as the visual pattern features of the protein fingerprint spectrum to be measured. The use of models such as dilated convolutional neural networks is to utilize their ability to effectively expand the receptive field, capture long-range dependencies, and multi-scale features when processing fingerprint spectrum-like image data, so as to more comprehensively capture complex global and local pattern features in the mass spectrum, rather than just focusing on isolated peaks. The ultimate goal is to transform these complex visual pattern features into a compact and information-rich mathematical representation, that is, the visual pattern feature coding vector of the protein fingerprint spectrum to be measured. This coding vector is a high-level generalization and abstraction of the original spectrum visual features, aiming to retain the discriminatory information to the greatest extent while filtering out irrelevant variations.

[0035] Specifically, dilated convolution introduces an interval sampling mechanism, which can significantly expand the receptive field and take into account both local peak regions and global trends while keeping the kernel parameters unchanged. The specific network structure is generally set as a multi-layer stack, and different dilation rates are used in each layer to capture multi-scale feature information. In a specific embodiment, the fingerprint spectrum visual pattern feature extractor based on the dilated convolutional neural network model first sets three convolutional layers in sequence, namely the first convolutional layer (64 channels, kernel width 3, dilation rate 1), the second convolutional layer (128 channels, kernel width 3, dilation rate 2), and the third convolutional layer (256 channels, kernel width 3, dilation rate 4). Batch normalization and ReLU activation are inserted between each layer, and global average pooling is attached as the output layer. The spatial feature vectors of all channels are averaged in the channel direction and compressed into a 256-dimensional fingerprint spectrum visual pattern feature encoding vector of the protein to be tested. During the training phase, a large number of strain fingerprint spectra and their class labels in the reference database are required to participate together. Through supervised learning, the cross-entropy loss or triplet loss function is defined to guide the classification and separation of the feature encoding space. Parameter optimization uses adaptive gradient descent methods such as Adam, with the learning rate set from 1e-3 to 1e-5, the batch size from 32 to 128, and the number of training epochs from 10 to 100, which is specifically determined according to the size of the training set and the convergence properties of the model. In the model training process, mechanisms such as Dropout and L2 regularization can be introduced to prevent overfitting, and data augmentation strategies such as spectrogram translation and peak width perturbation can also be used to enrich data diversity. Finally, the parameters after training convergence are frozen and used as the fingerprint spectrum visual pattern feature extractor in the inference phase.

[0036] Specifically, in step S5, fingerprint spectrum visual pattern features are extracted from the set of protein reference fingerprint spectra of the known strain types to obtain a set of protein fingerprint spectrum reference visual pattern features. It should be understood that in order to construct a high-quality reference benchmark that matches the feature representation of the sample to be tested and overcome the drawback of the traditional method being vulnerable to interference due to relying on discrete peak information, not only such robust features need to be extracted from the sample to be tested, but also the same feature transformation process must be performed on the known strain reference database used for comparison. If the reference database still uses the traditional spectrogram or peak list form, it will not be possible to effectively compare and match with the visual pattern feature encoding vector of the sample to be tested, rendering the subsequent query network ineffective. Therefore, in the technical solution of this application, fingerprint spectrum visual pattern features are further extracted from the set of protein reference fingerprint spectra of the known strain types to obtain a set of protein fingerprint spectrum reference visual pattern feature encoding vectors. That is to say, the original or preprocessed protein fingerprint spectra of a large number of known strains (including different Helicobacter pylori strains and other related strains) stored in the strain reference database are transformed into a standardized, visual pattern feature-based database through the same feature extraction mechanism as that for processing the sample to be tested (for example, also using the dilated convolutional neural network model). Specifically, a corresponding protein fingerprint spectrum reference visual pattern feature encoding vector is generated for each reference strain spectrogram, and these vectors are combined to form a new, feature-domain-based reference database. The goal of this database is to accurately and stably encode the characteristic identities of each known strain, making its representation consistent with the feature vector of the sample to be tested, laying the foundation for subsequent pattern matching.

[0037] In one embodiment, extracting fingerprint spectrum visual pattern features from the set of protein reference fingerprint spectra of the known strain types to obtain a set of protein fingerprint spectrum reference visual pattern features includes: respectively passing each protein reference fingerprint spectrum of the known strain types in the set of protein reference fingerprint spectra of the known strain types through the fingerprint spectrum visual pattern feature extractor based on the dilated convolutional neural network model to obtain a set of protein fingerprint spectrum reference visual pattern feature encoding vectors as the set of protein fingerprint spectrum reference visual pattern features.

[0038] Specifically, in step S6, a fingerprint spectrum pattern feature query is performed on the set of the visual pattern features of the protein fingerprint spectrum to be measured and the reference visual pattern features of the protein fingerprint spectrum to obtain a fingerprint spectrum visual pattern feature query response feature. It should be understood that traditional methods based on peak list alignment or overall spectrum similarity calculation often stay at the comparison of surface features and are difficult to effectively capture complex pattern changes caused by protein expression differences, sample processing, or instrument fluctuations, and are prone to failure when differentiating closely related species or processing variant spectra. Therefore, simply extracting more robust visual pattern features is not enough, and an intelligent engine that can understand and compare these complex feature patterns is also needed. Therefore, in the technical solution of this application, the set of the encoded vectors of the visual pattern features of the protein fingerprint spectrum to be measured and the encoded vectors of the reference visual pattern features of the protein fingerprint spectrum is further input into a fingerprint spectrum pattern feature query network to obtain an encoded vector of the fingerprint spectrum visual pattern feature query response. In particular, the fingerprint spectrum pattern feature query network can achieve high-level and in-depth comparison between the visual pattern features of the sample to be measured and numerous reference pattern features in the reference database, and finally condense a discriminative signal that can accurately reflect the attribution of the sample to be measured. This fingerprint spectrum pattern feature query network can perform the functions of monomer semantic query and self-attention aggregation. First, it can use the monomer semantic query unit to carefully evaluate the semantic correlation between the feature vector to be measured and each reference feature vector. Here, the semantic meaning refers to the strain-specific biological information contained in the mass spectrometry visual pattern. This is not just calculating a simple similarity score, but generating an encoded vector of the monomer semantic query score of the visual pattern of the protein fingerprint spectrum to be measured that contains multi-dimensional matching information. More importantly, the network will then use the self-attention aggregation strategy and consider the self-distribution characteristics of the feature set to achieve context-aware matching degree evaluation. This means that the network does not view the matching results of each to-be-measured-reference pair in isolation, but considers the sample to be measured in the context of the entire reference database, dynamically determines which reference patterns are the most significant and consistent with the sample to be measured globally, and adjusts the weights accordingly to effectively aggregate information. This encoded vector of the fingerprint spectrum visual pattern feature query response realizes the transition from scattered local feature matching to unified and global semantic determination, providing a solid and highly confident basis for determining the final strain detection result based on this vector, thus strongly supporting the more reliable and accurate Helicobacter pylori mass spectrometry identification goal aimed at by the entire technical solution.

[0039] In one embodiment, as Figure 3As shown, a fingerprint spectrum pattern feature query is performed on the set of visual pattern features of the protein fingerprint spectrum to be measured and the reference visual pattern features of the protein fingerprint spectrum to obtain a fingerprint spectrum visual pattern feature query response feature, including: S61, after enhancing the feature of the encoding vector of the visual pattern feature of the protein fingerprint spectrum to be measured, performing a monomer semantic query encoding on it and each encoding vector of the reference visual pattern feature of the protein fingerprint spectrum in the set of encoding vectors of the reference visual pattern feature of the protein fingerprint spectrum to obtain a set of encoding vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured; S62, calculating the monomer semantic weight of each encoding vector of the monomer semantic query score of the visual pattern of the protein fingerprint spectrum to be measured in the set of encoding vectors of the monomer semantic query score of the visual pattern of the protein fingerprint spectrum to be measured to obtain a set of self-attention weights of the monomer query semantics of the visual pattern of the protein fingerprint spectrum to be measured; S63, aggregating the set of encoding vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured based on the set of self-attention weights of the monomer query semantics of the visual pattern of the protein fingerprint spectrum to be measured to obtain the encoding vector of the fingerprint spectrum visual pattern feature query response.

[0040] In one embodiment, as Figure 4 shown, after enhancing the feature of the encoding vector of the visual pattern feature of the protein fingerprint spectrum to be measured, performing a monomer semantic query encoding on it and each encoding vector of the reference visual pattern feature of the protein fingerprint spectrum in the set of encoding vectors of the reference visual pattern feature of the protein fingerprint spectrum to obtain a set of encoding vectors of the monomer semantic query scores of the visual pattern of the protein fingerprint spectrum to be measured, including: S611, performing feature enhancement on the encoding vector of the visual pattern feature of the protein fingerprint spectrum to be measured based on deconvolution encoding to obtain an enhanced encoding vector of the visual pattern feature of the protein fingerprint spectrum to be measured, and the enhanced encoding vector of the visual pattern feature of the protein fingerprint spectrum to be measured has the same feature scale as each encoding vector of the reference visual pattern feature of the protein fingerprint spectrum in the set of encoding vectors of the reference visual pattern feature of the protein fingerprint spectrum. Specifically, this process can be represented by the formula: ; Among them, is the encoding vector of the visual pattern feature of the protein fingerprint spectrum to be measured, is deconvolution encoding, is the deconvolution weight matrix, is the first norm of the vector, is the enhanced encoding vector of the visual pattern feature of the protein fingerprint spectrum to be measured.

[0041] Specifically, the visual pattern feature encoding vector of the protein fingerprint spectrum to be measured is subjected to feature enhancement based on deconvolution encoding, and its core effect lies in greatly improving the expression ability of features and the richness of dimension semantics. The deconvolution encoding (i.e., inverse convolution or transposed convolution) mechanism can realize the reconstruction and enhancement of features in the high-dimensional space by learning the data-driven feature upsampling weights while retaining the original key spectral structure information. Specifically, this operation can not only improve the spatial resolution of the features of the sample to be measured, strengthen the representation ability of the microscopic patterns and global structures therein, but also adaptively enhance the content in the original encoding vector through a trainable weight system, effectively restoring, complementing or highlighting the differences and discriminative features that are potential but not fully expressed by the superficial encoding in the fingerprint spectrum. Compared with direct linear transformation or ordinary interpolation, the data adaptive feature enhancement brought by deconvolution encoding significantly expands the comparison basis between the sample and the reference feature vectors in the database, stretching each vector into the same high-level feature space. In addition, the deconvolution process ensures the complete consistency of the enhanced vector with all the visual pattern feature encoding vectors of the protein fingerprint spectrum in the reference database in terms of feature scale, dimension and internal structure distribution. This standardization is not only a technical guarantee for subsequent fair matching, but also can significantly improve the generalization and robustness of the pattern recognition system.

[0042] S612, perform monomer semantic query encoding on each protein fingerprint reference visual pattern feature encoding vector in the set of the enhanced visual pattern feature encoding vectors of the protein fingerprint spectrum to be measured and the protein fingerprint reference visual pattern feature encoding vectors respectively to obtain the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be measured. Specifically, this process can be expressed by the formula as follows: ; ; where, is the set of protein fingerprint reference visual pattern feature encoding vectors, are respectively the 1st, 2nd, th and th protein fingerprint reference visual pattern feature encoding vectors in the set of protein fingerprint reference visual pattern feature encoding vectors, and are respectively the trainable weight matrix and the trainable bias vector, is the vector concatenation operation, is the hyperbolic tangent function, is the th monomer semantic query score encoding vector of the visual pattern of the protein fingerprint spectrum to be measured.

[0043] Specifically, each protein fingerprint reference visual pattern feature encoding vector in the set of the enhanced protein fingerprint spectrum to be measured visual pattern feature encoding vectors and the protein fingerprint spectrum reference visual pattern feature encoding vectors is separately subjected to monomer semantic query encoding to obtain the set of the to-be-measured protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors. Here, the actual effect of the monomer semantic query encoding module is to automatically learn and quantify the deep semantic association and precise matching degree for the corresponding feature pairs composed of the enhanced protein fingerprint spectrum to be measured visual pattern feature encoding vector and each protein fingerprint spectrum reference visual pattern feature encoding vector through deep neural feature interaction. For example, for different Helicobacter pylori strains or related species under the same disease background, although their protein fingerprint visual patterns are overall similar, rich differences will be reflected in the deep distribution structure, peak domain configuration, or sub-block pattern, etc. Through monomer semantic query, a series of such semantic information can be encoded in a high-dimensional manner in the matching vector. The to-be-measured protein fingerprint spectrum visual pattern monomer semantic query score encoding vector is not a single score, but reflects all the association patterns of the corresponding feature pairs in the multi-semantic subspace, making the pairing between the to-be-measured spectrum and all database reference spectra present a rich and detailed dynamic feature relationship distribution.

[0044] In one embodiment, as Figure 5 shown, the monomer semantic weights of each to-be-measured protein fingerprint spectrum visual pattern monomer semantic query score encoding vector in the set of the to-be-measured protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors are respectively calculated to obtain the set of the to-be-measured protein fingerprint spectrum visual pattern monomer query semantic self-attention weights, including: S621, based on the self-distribution characteristics of the feature set of the set of the to-be-measured protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors, determining the monomer semantic matching degree of each to-be-measured protein fingerprint spectrum visual pattern monomer semantic query score encoding vector in the set of the to-be-measured protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to obtain the set of the to-be-measured protein fingerprint spectrum visual pattern monomer semantic matching degrees; S622, inputting the set of the to-be-measured protein fingerprint spectrum visual pattern monomer semantic matching degrees into the relationship gating proxy module to obtain the set of the to-be-measured protein fingerprint spectrum visual pattern monomer query semantic self-attention weights.

[0045] In one embodiment, in step S621, based on the self-distribution characteristics of the feature set of the set of encoded vectors of the visual pattern monomer semantic query scores of the protein fingerprint spectrum to be measured, determining the monomer semantic matching degree of each encoded vector of the visual pattern monomer semantic query scores of the protein fingerprint spectrum to be measured in the set of encoded vectors of the visual pattern monomer semantic query scores of the protein fingerprint spectrum to be measured to obtain a set of monomer semantic matching degrees of the visual pattern monomer semantic query scores of the protein fingerprint spectrum to be measured, including: performing inherent alignment calibration on each encoded vector of the visual pattern monomer semantic query scores of the protein fingerprint spectrum to be measured in the set of encoded vectors of the visual pattern monomer semantic query scores of the protein fingerprint spectrum to be measured to obtain a set of calibrated encoded vectors of the visual pattern monomer semantic query scores of the protein fingerprint spectrum to be measured. Specifically, this process can be expressed by the formula: ; ; ; ; ;

[0046] wherein, is the query potential vector of the visual pattern of the protein fingerprint spectrum to be measured, is the projection weight matrix, is the query interaction encoded vector of the visual pattern of the protein fingerprint spectrum to be measured, is the coupling constant, for example, calculated in the same way as in the case of deconvolution enhancement to maintain symmetry, is the query covariance matrix of the visual pattern of the protein fingerprint spectrum to be measured, is the th calibrated encoded vector of the visual pattern monomer semantic query score of the protein fingerprint spectrum to be measured.

[0047] It should be understood that the monomer semantic query encoding adopts the method of directly feature splicing the enhanced encoded vector of the visual pattern feature of the protein fingerprint spectrum to be measured and the encoded vector of the reference visual pattern feature of the corresponding protein fingerprint spectrum in the database to realize the generation of the encoded vector of the visual pattern monomer semantic query score of the protein fingerprint spectrum to be measured. In this process, it is expected to optimize the dynamic characteristic representation from the original feature space to the semantic query encoding space, and focus on calibrating the inherent alignment between the feature space and the semantic query encoding space, significantly improving the representation accuracy of the encoded vector of the visual pattern monomer semantic query score of the protein fingerprint spectrum to be measured.

[0048] Specifically, the enhanced visual pattern feature encoding vector of the protein to be measured and the reference visual pattern feature encoding vectors of each protein fingerprint spectrum are synthesized into a dynamic eigenvector representation through a concatenation operation, that is, the visual pattern query interaction encoding vector of the protein fingerprint spectrum to be measured. This visual pattern query interaction encoding vector of the protein fingerprint spectrum to be measured undergoes further spatial mapping to obtain its projection representation to the eigenstate of the semantic query encoding, that is, the visual pattern query potential vector of the protein fingerprint spectrum to be measured. On this basis, an inherent alignment is performed on the visual pattern query potential vector of the protein fingerprint spectrum to be measured under the mapping representation. This inherent alignment uses the same coupling constant setting as in the deconvolution feature enhancement step, effectively ensuring the symmetry and mathematical consistency of the entire mapping process. At the same time, by reasonably constructing and solving the visual pattern query covariance matrix of the protein fingerprint spectrum to be measured, the covariant path is constrained under the condition of gauge symmetry, thereby fixing the mapping gauge. This mechanism improves the representation accuracy of the monomer semantic query score encoding vector of the visual pattern of the protein fingerprint spectrum to be measured in the high-dimensional semantic space.

[0049] Calculate the monomer semantic matching degree of each calibrated monomer semantic query score encoding vector of the visual pattern of the protein fingerprint spectrum to be measured based on the distribution characteristics of the set of the calibrated monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be measured to obtain the set of the monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be measured. Specifically, this process can be expressed by the formula: ;

[0050] where, is the number of, is the th calibrated monomer semantic query score encoding vector of the visual pattern of the protein fingerprint spectrum to be measured, represents the transpose vector, is the natural exponential function with base e, is function, is the monomer semantic matching degree of the visual pattern of the protein fingerprint spectrum to be measured.

[0051] Specifically, the calculation of the monomer semantic matching degree is not only based on the numerical characteristics of the calibrated monomer semantic query score encoding vector of the visual pattern of the protein fingerprint spectrum to be measured itself, but also comprehensively considers the distribution characteristics of the set of the calibrated monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be measured. In this way, those feature pairs that can significantly distinguish the spectrum to be measured from the reference database and have high biological relevance and discriminative efficacy will be given a higher matching degree, while the influence of noise-type, redundant-type or weakly discriminative feature pairs will be automatically weakened. Finally, all monomer semantic matching degrees jointly form a set of monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be measured.

[0052] Specifically, the calculation process of inputting the set of monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be measured into the relational gating proxy module to obtain the set of self-attention weights of the monomer query semantics of the visual pattern of the protein fingerprint spectrum to be measured can be expressed by the formula as follows: ;

[0053] where, is a masking function, is a preset threshold, for example, 0.5. Of course, this is only an example here and can be adjusted according to the actual situation specifically. is the self-attention weight of the monomer query semantics of the visual pattern of the protein fingerprint spectrum to be measured.

[0054] Specifically, by means of a trainable gating mechanism, the relational gating proxy module can effectively retain those monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be measured that are most relevant to the discrimination of target strains such as Helicobacter pylori, and at the same time automatically suppress the adverse effects of accidentally high scores, weakly relevant or redundant pairings on the overall decision result. Based on the context understanding of the global feature space, the relational gating proxy module realizes the dynamic weight allocation of the monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be measured in the subsequent self-attention aggregation, so that the region focused by the final self-attention mechanism highly coincides with the actual discrimination requirements.

[0055] Specifically, the calculation process of aggregating the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be measured based on the set of self-attention weights of the monomer query semantics of the visual pattern of the protein fingerprint spectrum to be measured to obtain the feature query response encoding vector of the fingerprint spectrum visual pattern can be expressed by the formula as follows: ;

[0056] where, is the feature query response encoding vector of the fingerprint spectrum visual pattern.

[0057] First, through in-depth modeling in the feature domain, a monomer semantic query score encoding vector between each protein fingerprint spectrum visual pattern to be measured and the database reference pattern is obtained, and by utilizing the context distribution characteristics of the feature set and the relationship gating proxy mechanism, an adaptive monomer query semantic self-attention weight is assigned to each encoding vector. These weights reflect the actual contributions and discriminative powers of the encoding vectors in the overall feature expression, effectively highlighting the highly discriminative matching information while weakening or suppressing noise and redundancy. On this basis, in the aggregation step, using the self-attention weight as the weighting coefficient, a dynamic linear combination of all monomer semantic query score encoding vectors is performed. During the aggregation process, not only the local matching information at the micro level is weighted and integrated according to its importance, but also the global semantic abstraction and consistent expression of the overall visual structure diagram of the protein fingerprint spectrum are realized.

[0058] Specifically, in step S7, based on the fingerprint spectrum visual pattern feature query response feature, the bacterial species detection result is determined. In one embodiment, determining the bacterial species detection result based on the fingerprint spectrum visual pattern feature query response feature includes: passing the fingerprint spectrum visual pattern feature query response encoding vector through a bacterial species detector based on a classifier to obtain the bacterial species detection result, and the bacterial species detection result is used to represent the bacterial species label. Specifically, the Softmax classifier is widely used in the final multi-class discrimination module after deep learning features because of its rigorous structure, strong differentiability, and suitability for multi-class probability modeling. In a specific embodiment of the present application, the fingerprint spectrum visual pattern feature query response encoding vector is input into the Softmax classifier. If the fingerprint spectrum visual pattern feature query response encoding vector is a fixed-length high-dimensional vector and there are m bacterial species in the system reference database, then for each category i to be discriminated, the Softmax classifier first multiplies by the learnable weight vector of this category, then adds the corresponding bias to obtain the corresponding score, further obtains the probability distribution through exponential operation, and finally outputs the category corresponding to the largest pi as the bacterial species detection result, and this probability distribution can be directly regarded as an indication of the model's credibility.

[0059] In summary, the Helicobacter pylori detection method based on mass spectrometry recognition provided by the present application first performs MALDI mass spectrometry acquisition on the sample to be tested to obtain its protein fingerprint spectrum, and then removes noise and standardizes it through preprocessing to further extract the visual pattern features of the features to be tested that have the ability to discriminate, and at the same time extracts the reference visual pattern features of each known strain from the constructed strain reference database, and finally, by introducing the protein fingerprint spectrum pattern feature query response analysis based on the feature domain to learn the complex matching relationship between the features to be tested and the reference features to obtain the fingerprint spectrum visual pattern feature query response encoding representation, and finally determine the strain detection result based on this query response. This strategy based on visual pattern feature extraction and query network matching aims to make full use of the deep pattern information of mass spectrometry data to achieve more reliable Helicobacter pylori mass spectrometry identification.

[0060] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details of the above embodiments are only for the purpose of illustration and ease of understanding, rather than limitation, and the above details do not limit the present application to the use of the above specific details to be implemented.

[0061] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0062] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0063] Finally, it should be noted that the above description has been given for purposes of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present application rather than to limit them. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for detecting Helicobacter pylori based on mass spectrometry, characterized in that: include: The sample to be tested is mixed with the MALDI matrix and then spotted on the MALDI target plate, and the protein fingerprint spectrum is collected using a MALDI-TOF mass spectrometer; Extracting a collection of protein reference fingerprint spectra of known bacterial species from a bacterial species reference database; Preprocessing the protein fingerprint spectrum to obtain a preprocessed protein fingerprint spectrum; Extracting visual pattern features of the fingerprint spectrum from the pre-processed protein fingerprint spectrum to obtain visual pattern features of the fingerprint spectrum of the protein to be tested; Extracting fingerprint spectrum visual pattern features from the set of protein reference fingerprint spectrum of the known bacterial species to obtain a set of protein fingerprint spectrum reference visual pattern features; Performing a fingerprint spectrum pattern feature query on the set of the visual pattern feature of the protein fingerprint spectrum to be tested and the visual pattern feature of the protein fingerprint spectrum reference to obtain a fingerprint spectrum visual pattern feature query response feature; Based on the visual pattern feature query response feature of the fingerprint spectrum, the bacterial species detection result is determined.

2. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 1, characterized in that: The preprocessing includes baseline correction, noise smoothing, peak identification and alignment processing.

3. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 2, characterized in that: Extracting visual pattern features of the fingerprint spectrum from the preprocessed protein fingerprint spectrum to obtain visual pattern features of the protein fingerprint spectrum to be tested, comprising: passing the preprocessed protein fingerprint spectrum through a fingerprint spectrum visual pattern feature extractor based on a hole convolutional neural network model to obtain a visual pattern feature encoding vector of the protein fingerprint spectrum to be tested as the visual pattern feature of the protein fingerprint spectrum to be tested.

4. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 3, characterized in that: Extracting fingerprint spectrum visual pattern features from the set of protein reference fingerprint spectra of the known bacterial strain types to obtain a set of protein fingerprint spectrum reference visual pattern features, including: passing the protein reference fingerprint spectra of each known bacterial strain type in the set of protein reference fingerprint spectra of the known bacterial strain types through the fingerprint spectrum visual pattern feature extractor based on the void convolutional neural network model to obtain a set of protein fingerprint spectrum reference visual pattern feature encoding vectors as the set of protein fingerprint spectrum reference visual pattern features.

5. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 4, characterized in that: Performing a fingerprint spectrum pattern feature query on the set of the visual pattern feature of the protein fingerprint spectrum to be tested and the visual pattern feature of the protein fingerprint spectrum reference to obtain a fingerprint spectrum visual pattern feature query response feature, including: After the visual pattern feature coding vector of the protein fingerprint spectrum to be tested is feature enhanced, it is subjected to monomer semantic query coding together with each protein fingerprint spectrum reference visual pattern feature coding vector in the set of protein fingerprint spectrum reference visual pattern feature coding vectors to obtain a set of monomer semantic query score coding vectors of the visual pattern of the protein fingerprint spectrum to be tested; Calculating the monomer semantic weights of the monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested in the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested respectively to obtain a set of monomer semantic self-attention weights of the visual pattern of the protein fingerprint spectrum to be tested; The set of monomer semantic query score encoding vectors of the protein fingerprint spectrum visual pattern to be tested is aggregated based on the set of monomer semantic self-attention weights of the protein fingerprint spectrum visual pattern to be tested to obtain the fingerprint spectrum visual pattern feature query response encoding vector.

6. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 5, characterized in that: After the visual pattern feature coding vector of the protein fingerprint spectrum to be tested is feature enhanced, it is subjected to monomer semantic query coding together with each protein fingerprint spectrum reference visual pattern feature coding vector in the set of protein fingerprint spectrum reference visual pattern feature coding vectors to obtain a set of monomer semantic query score coding vectors of the visual pattern of the protein fingerprint spectrum to be tested, including: Performing feature enhancement based on deconvolution coding on the visual pattern feature coding vector of the protein fingerprint spectrum to be tested to obtain a reinforced visual pattern feature coding vector of the protein fingerprint spectrum to be tested, wherein the reinforced visual pattern feature coding vector of the protein fingerprint spectrum to be tested has the same feature scale as each protein fingerprint spectrum reference visual pattern feature coding vector in the set of the protein fingerprint spectrum reference visual pattern feature coding vector; The enhanced protein fingerprint spectrum visual pattern feature coding vector to be tested and each protein fingerprint spectrum reference visual pattern feature coding vector in the set of the protein fingerprint spectrum reference visual pattern feature coding vector are respectively subjected to monomer semantic query coding to obtain a set of monomer semantic query score coding vectors of the visual pattern of the protein fingerprint spectrum to be tested.

7. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 6, characterized in that: The monomer semantic weights of the monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested in the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested are calculated respectively to obtain a set of monomer semantic self-attention weights of the visual pattern of the protein fingerprint spectrum to be tested, including: Based on the feature set self-distribution characteristics of the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested, determining the monomer semantic matching degree of each monomer semantic query score encoding vector of the visual pattern of the protein fingerprint spectrum to be tested in the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested to obtain a set of monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be tested; The set of semantic matching degrees of the monomers of the visual pattern of the protein fingerprint spectrum to be tested is input into the relational gating agent module to obtain the set of query semantic self-attention weights of the monomers of the visual pattern of the protein fingerprint spectrum to be tested.

8. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 7, characterized in that: Based on the feature set self-distribution characteristics of the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested, determining the monomer semantic matching degree of each monomer semantic query score encoding vector of the visual pattern of the protein fingerprint spectrum to be tested in the set of monomer semantic query score encoding vectors of the visual pattern of the protein fingerprint spectrum to be tested to obtain a set of monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be tested, comprising: Performing inherent alignment calibration on each of the protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors in the set of protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors to obtain a set of calibrated protein fingerprint spectrum visual pattern monomer semantic query score encoding vectors; Based on the distribution characteristics of the set of monomer semantic query score encoding vectors of the calibrated visual pattern of the protein fingerprint spectrum to be tested, the monomer semantic matching degree of each calibrated visual pattern monomer semantic query score encoding vector of the protein fingerprint spectrum to be tested is calculated to obtain the set of monomer semantic matching degrees of the visual pattern of the protein fingerprint spectrum to be tested.

9. The method for detecting Helicobacter pylori based on mass spectrometry recognition according to claim 8, characterized in that: Based on the fingerprint spectrum visual pattern feature query response feature, determining the strain detection result includes: passing the fingerprint spectrum visual pattern feature query response encoding vector through a classifier-based strain detector to obtain a strain detection result, and the strain detection result is used to represent a strain category label.

Images