Method for converting liquid mass spectrometry data into picture data capable of carrying out AI picture identification analysis

By converting liquid mass spectrometry data into AI-recognizable picture data, the problem of unused signals in liquid mass spectrometry detection is solved, and more comprehensive sample feature display and disease interpretation is achieved, improving the accuracy and efficiency of mass spectrometry analysis.

CN118918205APending Publication Date: 2024-11-08PROTEINT (TIANJIN) BIOTECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411055562.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-08

Smart Images

  • Figure CN118918205A_ABST
    Figure CN118918205A_ABST
Patent Text Reader

Abstract

The invention discloses a method for converting liquid mass spectrometry data into picture data capable of carrying out AI picture recognition, which comprises the following steps: taking detection time as a picture x axis, taking detection mass-to-charge ratios as a picture y axis, and converting the detection intensity of each detected mass-to-charge ratio into points with different colors according to a certain mapping rule to be distributed on a picture. The picture has basic features of the picture, can reflect features of a detection sample, can be used for AI feature recognition, and is used for clustering, grouping, feature finding and the like of the sample. According to the method, all data detected by the liquid phase mass spectrometry can be fully utilized, data filtering missing and omission caused by insufficient qualitative evidence in the data analysis process are avoided, sample features can be more fully displayed, and the method is used for different algorithms based on AI feature recognition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of protein spectrum and bioinformatics, and specifically, relates to a method for converting liquid phase mass spectrometry data into image data that can be used for AI image recognition analysis. Background Art

[0002] Although with the gradual improvement of detection technology and data analysis technology, the detection and identification depth and accuracy of various omics technologies based on liquid chromatography-mass spectrometry, including proteomics, peptides, metabolomics, etc., have been significantly improved, there are still a large number of unidentified signals in the detection data that are not fully utilized. Usually due to the incompleteness of the reference database or the limitations of the analysis method, these unidentified signals will be directly filtered out, which may include special modifications, unknown substances, non-targeted biological substances and other different situations. Usually these unidentified signals will be redundant or even far more than the identified signals, which shows that the identified signals cannot fully reflect the full picture of the sample. When it is necessary to compare the similarities and differences between samples, only considering the identified signals will be one-sided, and the data may be distorted due to data correction. At the same time, the joint manifestation of multiple indicators at the global level is often easily overlooked.

[0003] AI image recognition technology is currently developing rapidly and has made remarkable achievements in different fields. From face recognition, which can identify personal features in different situations and angles, to identifying CT images in the clinical medical field to determine disease characteristics, it has been well applied.

[0004] If there is a method that can enable AI image recognition to read each mass spectrometry data like recognizing facial organs, fully utilize all the mass spectrometry information of each sample, and then identify whether there are specific compound group characteristics in each sample to group and classify the samples for disease diagnosis, it will bring new application directions in the field of mass spectrometry analysis. Summary of the invention

[0005] Purpose of the invention: In order to solve the above-mentioned technical problems, the present invention provides a method for converting liquid mass spectrometry data into image data that can be used for AI image recognition and analysis.

[0006] Technical solution: In order to achieve the above-mentioned invention object, the present invention adopts the following technical solution:

[0007] A method for converting liquid mass spectrometry data into image data that can be used for AI image recognition, comprising the following steps:

[0008] (1) Converting liquid mass spectrometry data into universal readable files using platform software;

[0009] (2) extracting time or scan spectrum number, mass-to-charge ratio and response intensity information from a universally readable file;

[0010] (3) using the information of step (2) above as the x-axis according to the time or scan spectrum number and the mass-to-charge ratio as the y-axis;

[0011] (4) converting the detection signal of each mass-to-charge ratio into a z value between 0 and 255, and filling it into the coordinate matrix of step (3);

[0012] (5) Based on the matrix obtained in step (4), the z value of each point is used as the hue and converted into a color code, and the matrix is ​​converted into an image format of x*y points for AI image recognition.

[0013] As a specific implementation scheme, in step (1), the platform software for converting data is selected from MSconvert, DataAnalysis or ThermoRawFileParser, etc.

[0014] As a specific implementation scheme, in step (1), the universal readable files include mgf, mzML, mzXML, text, ms1, ms2, cms1, cms2, preferably mgf, mzXML, text, ms1, ms2. If the mapping data is limited to the primary spectrum, ms1 is preferred; if the mapping data is limited to the secondary spectrum, ms2 is preferred; the rest are preferably mgf, mzXML, and text.

[0015] As a specific implementation scheme, in step (2), the time or scan spectrum number is the retention time or scan spectrum number of each single mass spectrum; the mass-to-charge ratio is the mass to charge ratio of the ion detected by the mass spectrometer, that is, m / z; and the response intensity information is the signal intensity of different ions with different mass-to-charge ratios based on the current mass spectrum.

[0016] As a specific implementation scheme, in step (2), the x-axis is labeled with time or scan spectrum number, or the x-axis is labeled with time, such as scan1, scan2, scan3, etc. For the x-axis, each time point or a scan spectrum is used as an x-axis point, or they are merged according to the mass spectrometry scanning cycle, such as in DDA (data independent analysis mode), a primary spectrum and 40 secondary spectra of the strongest signal ions are one cycle, or in DIA (data dependent analysis mode), a primary spectrum and 40 secondary spectra of a preset mass-to-charge ratio interval window are one cycle. In this case, the x-coordinate information of the 41 spectra in the above case can be merged into one x-coordinate point, or the time period occupied by this cycle can be used as one x-coordinate point.

[0017] As a specific implementation scheme, in step (3), the mass-to-charge ratio is used as the y-axis, and the mass-to-charge ratio needs to be divided into intervals, and the interval size is set to 0.001 to 1 Da, preferably 0.1 to 1 Da. If the computer has superior computing power, the interval size can be reduced.

[0018] As a specific implementation scheme, in step (4), each mass-to-charge ratio detection signal is converted into a z value between 0 and 255. The conversion method specifically includes the following method: predetermining the effective signal intensity range [1*10 x0 , 1*10 x1 ), the number of orders of magnitude m = x1-x0, if the signal strength of E4~E9 is effective, that is, 10,000~9,999,999,999, then x0=4, x1=10, then m=10-4 is 6, each order of magnitude is numbered from 1 to m in sequence, and the range of each order of magnitude is [1,10)*10 x (x is the current order of magnitude index); divide each order of magnitude into n equal parts, such as n is 36, then the E4 order of magnitude is divided into 1E4, 1.25E4, 1.5E4, 1.75E4 ... 9.75E4, the effective number of divisions is 36, each dividing point is an equal division interval, each equal division interval is numbered from 1 to n in sequence, where m*n≤255; all detection signals data1=a*10 x , a∈[1,10), confirm the interval segment according to the above rules, marked as (m1,n1), where m1=rounded (log10(data1))-x0+1, is the order of magnitude number of a detection signal, n1=rounded (((data1 / 10 X )-1) / (9 / n))+1, where x is the magnitude of data1, such as signal data 4000000=4×10 6 , where x is 6, n1 is the number of the equal-division interval where the detection signal is located, and finally (m1, n1) is calculated according to z = (m1-1)*n+n1-1 to obtain the z value. If the signal strength of E4~E9 is the effective strength, that is, 10,000~9,999,999,999, then x0 = 4, x1 = 10, then m = 10-4 is 6, and each order of magnitude is divided into 36 equal parts, n = 36. For the signal data 2,550,000, calculated by the above formula, m1 = rounded (log10(data1))-x0+1=3, n1 = rounded (((data1 / 10 X )-1) / (9 / n))+1=7, z=(m1-1)*n+n1-1=78.

[0019] The “rounding” mentioned in the above formula refers to the integer before the decimal point of the obtained value, rather than the integer after rounding. The “[” in the range symbol means inclusive, and “)” means exclusive.

[0020] Furthermore, in the conversion method, a value above the effective signal strength range is adjusted to the upper limit of the effective signal strength, and a value below the effective signal strength range is adjusted to the lower limit of the effective signal strength or reset to zero, preferably reset to zero.

[0021] As a specific implementation scheme, in step (5), the z value of each point is used as the hue and converted into a color code by using the HSL color mode, locking the brightness (L) and saturation (S) as fixed values, and the value in the matrix is ​​used as the hue (H). The hexadecimal color code is obtained based on the HSL value; the saturation range is 64 to 255, and the brightness range is 64 to 223. For example, when HSL is (89, 128, 128), the color code is #40C04C.

[0022] As a specific implementation scheme, in step (5), the matrix is ​​converted into a picture format of x*y points, and the color of each (x, y) point is drawn based on the color code corresponding to the (x, y) point; the drawing software is selected from the ggplot2 package based on R language, Matplotlib, Seaborn, Pillow or GIMP, Photoshop, ImageJ, etc. based on Python.

[0023] Beneficial effects: Compared with the prior art, the method of the present invention uses the detection time as the x-axis of the image, and the detection mass-to-charge ratio as the y-axis of the image. The detection intensity of each detected mass-to-charge ratio is converted into points of different colors distributed on the image according to a certain mapping rule. The image has the basic characteristics of the image and can reflect the characteristics of the detected sample. It can be used for AI feature recognition and for clustering, grouping, and finding features of samples. This method can make full use of all data from liquid mass spectrometry detection, avoid omissions in data filtering caused by insufficient qualitative evidence during data analysis, and can more fully display sample characteristics for different algorithms based on AI feature recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Set up for MSconvert software.

[0025] Figure 2 Open the figure in the readable format control-1.ms1.

[0026] Figure 3 Image converted to a color-labeled matrix.

[0027] Figure 4 This is the image converted in Application Example 1.

[0028] Figure 5 This is the image converted in Application Example 2.

[0029] Figure 6 This is a schematic comparison diagram of the data after qualitative and quantitative analysis of the proteome. DETAILED DESCRIPTION

[0030] The present invention can be better understood according to the following examples. However, the contents described in the examples are only used to illustrate the present invention, and should not and will not limit the invention described in detail in the claims.

[0031] Example

[0032] 1. Select a proteomics data control-1.raw as the test data. This data is the proteomics data detected on the Thermo Fisher Scientific U3000-Q Exactive HF liquid chromatography mass spectrometer;

[0033] 2. Use MSconvert software according to Figure 1 Set up, extract the primary mass spectrometry data from the data for display, convert control-1.raw to a readable format control-1.ms1, open as Figure 2 As shown, this example uses R language for subsequent processing;

[0034] 3. According to the format of ms1, the first column is the cycle of S, I, I, I, I, m / z345-1515, and each cycle is a scan spectrum. Therefore, this ms1 can be divided into x=814 scan files based on the cycle;

[0035] 4. Based on the mass spectrometer model and related parameters, the minimum detection limit is set to E4, so all rows with the second column value less than 10,000 in the scan file are filtered out;

[0036] 5. Determine that the maximum detection signal is 6E9, so the maximum order of magnitude is set to E9, the order of magnitude is E4 to E9, the number m is 6, and the order of magnitude fraction n is set to 36;

[0037] 6. In each scan file, although the m / z interval is displayed as 345-1515, subsequent analysis can be performed according to the preset parameters of 350-1500; 1Da is a partition interval, and each partition interval takes the maximum quantitative value, such as 350-350.9999 as a partition area, and a maximum value is taken, that is, each scan can be cut into 1500-350=1150 partitions, and each partition area will have a maximum quantitative value. If the partition area is empty, it will be marked as 0;

[0038] 7. Summarize the above 814 files and 1150 partitions to obtain a quantitative matrix of x*y=814*1150;

[0039] 8. Convert the values ​​in the quantitative matrix in step 7: 1. If it is 0, mark it as '#FFFFFF'; 2. Use the formula to calculate the m1, n1 and z value of the remainder data, m1 = rounded (log10(data1))-x0+1, n1 = rounded (((data1 / 10 X )-1) / (9 / n))+1, and use the formula to get z=(m1-1)*n+n1-1; 3. Use (z, 128, 128) as the HSL value (brightness L and saturation S are fixed values ​​128, and z value is used as hue H), then convert the HSL value to RGB value and then to hexadecimal value, and fill it back into the original matrix position;

[0040] 9. Use the ggplot2 package to convert the color-coded matrix into a picture, such as Figure 3 It is available for subsequent use.

[0041] Application Example 1

[0042] During routine sample testing, quality control is inserted at fixed times or sample intervals to evaluate the test quality. However, it is not possible to timely check whether there are problems with each sample test. Usually, after running some samples, each raw data needs to be manually opened to confirm the test situation and whether there are any signal abnormalities. Among them, in protein spectra, the formation of droplets at the needle tip of the ion source causing broken spray is a relatively common problem. The image generated by the method of the present invention is as follows: Figure 4 , there are a lot of blank entries in the middle, that is, the signal vacuum area.

[0043] Figure 4 The example in the figure is an obvious spray interruption. The problem can be easily confirmed by manually checking the raw data, but there is a problem of long prompt lag. In most cases, the spray interruption is short-term, which is not easy to find when checking the raw data due to the dense spectrum.

[0044] At this time, the graph of the previous spray break is given to TensorFlow for learning, and the obtained model is integrated into an automated reading and analysis program, which can achieve the following: After a data collection is completed, the patent is used to automatically generate a picture, and the model is used to determine whether it belongs to the spray break group. If it is a spray break, an error report is generated and a prompt message is sent. Based on this, the status of the protein spectrum instrument will be confirmed in real time, and the validity of the data will be confirmed to reduce data loss.

[0045] Application Example 2

[0046] After proteomic testing of two groups of samples, the method of the present invention was used to generate images, wherein each group had two samples, one for each group. Figure 5 , the two groups of pictures were given to TensorFlow for learning, and it was found that there were different areas between the two groups, such as Figure 5 In the box area logo, the color blocks of group 2 (right) are denser than those of group 1 (left), and there are special color dots, which can be used as a distinguishing mark between the two groups.

[0047] However, when using traditional proteomics to perform qualitative and quantitative data analysis, most of this information will be filtered. The qualitative and quantitative information will be scattered in the protein or peptide quantitative information table because only the labels are retained. For example, Figure 6 As shown in the figure, during the analysis, the same data was used to obtain protein quantitative data after using DIANN for proteome qualitative and quantitative analysis. Figure 5 The retention time (x-axis) and molecular weight (y-axis) information in the above formula will no longer be used. Only the quantitative information ( Figure 5 The color of the midpoint reflects the quantitative intensity), and its multidimensional information is largely lost. Comparisons such as clustering and differences are also based only on mathematical calculations.

[0048] The above embodiments are the best implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for converting liquid mass spectrometry data into image data capable of AI image recognition, characterized in that: The following steps are involved: (1) Converting liquid mass spectrometry data into universal readable files using platform software; (2) extracting time or scan spectrum number, mass-to-charge ratio and response intensity information from a universally readable file; (3) using the information of step (2) above as the x-axis according to the time or scan spectrum number and the mass-to-charge ratio as the y-axis; (4) converting the detection signal of each mass-to-charge ratio into a z value between 0 and 255, and filling it into the coordinate matrix of step (3); (5) Based on the matrix obtained in step (4), the z value of each point is used as the hue and converted into a color code, and the matrix is ​​converted into an image format of x*y points for AI image recognition.

2. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (1), the platform software for converting data is selected from MSconvert, DataAnalysis or ThermoRawFileParser.

3. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (1), the generally readable files include mgf, mzML, mzXML, text, ms1, ms2, cms1, cms2, preferably mgf, mzXML, text, ms1, ms2.

4. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (2), the time or scan spectrum number is the retention time or scan spectrum number of each single mass spectrum; the mass-to-charge ratio is the mass-to-charge ratio of the ion detected by the mass spectrometer, that is, m / z; and the response intensity information is the signal intensity of different ions with different mass-to-charge ratios based on the current mass spectrum.

5. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (2), the x-axis is labeled with time or the scan spectrum number, or the scan spectrum number is used as the x-axis; for the x-axis, each time point or a scan spectrum is used as an x-axis point, or they are merged according to the mass spectrometry scan cycle.

6. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (3), the mass-to-charge ratio is used as the y-axis, and the mass-to-charge ratio needs to be divided into intervals, and the interval size is set to 0.001 to 1 Da, preferably 0.1 to 1 Da.

7. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (4), each mass-to-charge ratio detection signal is converted into a z value between 0 and 255. The conversion method specifically includes the following method: predetermining an effective signal intensity range [1*10 x0 , 1*10 x1 ), the number of its magnitudes is m = x1-x0, each magnitude is numbered from 1 to m, and the range of each magnitude is [1,10)*10 x (x is the current order of magnitude index); divide each order of magnitude into n equal parts, each dividing point is an equal division interval, and each equal division interval is numbered from 1 to n in sequence, where m*n≤255; all detection signals data1=a*10 x According to the above rules, the interval segment is confirmed and marked as (m1, n1), where m1 = rounded (log10(data1))-x0+1, which is the order of magnitude number of a certain detection signal, and n1 = rounded (((data1 / 10 X )-1) / (9 / n))+1, where x is the magnitude value of data1, n1 is the number of the equally divided interval where the detection signal is located, and finally (m1,n1) is calculated according to z=(m1-1)*n+n1-1 to obtain the z value.

8. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 7, characterized in that: In the conversion method, the value above the effective signal strength range is adjusted to the upper limit of the effective signal strength, and the value below the effective signal strength range is adjusted to the lower limit of the effective signal strength or reset to zero, preferably reset to zero.

9. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (5), the z value of each point is used as the hue and converted into a color code. The HSL color mode is used to lock the brightness (L) and saturation (S) as fixed values. The value in the matrix is ​​used as the hue (H), and the hexadecimal color code is obtained based on the HSL value; the saturation range is 64 to 255, and the brightness range is 64 to 223.

10. The method for converting liquid mass spectrometry data into image data capable of AI image recognition according to claim 1, characterized in that: In step (5), the matrix is ​​converted into a picture format of x*y points, and the color of each (x, y) point is drawn based on the color code corresponding to the (x, y) point; the drawing software is selected from the ggplot2 package based on R language, Matplotlib, Seaborn, Pillow or GIMP, Photoshop, ImageJ based on Python.