RU2803128C2 - Method of processing chromato-mass-spectrometric data to increase the efficiency of search for diagnostic markers in clinical studies - Google Patents

Abstract

FIELD: medicine, analytical chemistry and medical diagnostics.

SUBSTANCE: invention can be used to determine potential diagnostic marker compounds in clinical trials. The samples of biological material obtained or isolated from the patient are taken. A uniform distribution of samples of each clinical group between different analyzed batches are carried out. Chromato-mass spectrometric analysis of the received batches is performed. The received data is preprocessed and at least one compound is identified from the received data. A set of compounds, potential markers using a trained classification model are obtained. Based on the Akaike information criterion, compounds, potential markers are selected from the set selected at the previous stage. Compounds are stepwise removed from the generated set of compounds, potential markers, in which the value of the probability of the coefficient differing from zero is less than the value of the probability limit of the coefficient differing from zero.

EFFECT: method improves the accuracy of determining diagnostic markers in clinical trials by using the Akaike information criterion.

1 cl, 5 dwg, 3 ex

Description

TECHNICAL FIELD

This technical solution relates to the field of analytical chemistry, namely to methods for processing data obtained during mass spectrometric analysis of a large number of samples to search for diagnostic markers.

BACKGROUND OF THE ART

The source of information RU 2 743 418 C1, published on February 18, 2021, is known from the prior art, revealing a method for analyzing data on the content of lipid classes of interest in a sample based on mass spectrometric analysis with liquid chromatography, including obtaining liquid chromatography data with mass spectrometry of the analyzed sample , processing the spectra to obtain tables with the intensities of lipid features and their mass-per-charge and retention time values, determining the lattice model, searching for the optimal lattice by selecting the optimal set of parameters, generating an annotation using the optimal lattice, where all the features that fall within the predicted time within a predetermined error, are considered annotated, output of the annotation result in the form of a table where the lipid characteristics are associated with the name of the lipid.

The source of information RU 2 744 021 C1, published on March 02, 2021, is known from the prior art, revealing a method for diagnosing steatosis and non-alcoholic steatohepatitis in women based on venous blood, and the analysis is carried out by the chromatographic method. The data obtained during chromatographic analysis is processed using MetAlign, AIoutput, resulting in a data matrix. Then the data is loaded into the BioClassificator.py software package, which allows them to be processed using the principal component method (PCA), as well as classifiers - SVM (support vector machine), PLS-DA (Partial least squares Discriminant Analysis), Naive Bayes. As a result, an ROC curve is constructed for each method, and the average accuracy and its variance are also output. Using the SVM (support vector machine) classifier, compounds are selected that separate groups of healthy people and patients with non-alcoholic fatty liver disease, which are diagnostic markers.

SUMMARY OF THE INVENTION

The technical problem that the claimed solution is aimed at is the development of a method for processing data obtained using liquid chromatography-mass spectrometry to create a panel of markers for a model that classifies data “pathology/absence of pathology” during clinical trials, based on logistic regression described in the independent claim of the formula. Additional embodiments of the present invention are presented in the dependent claims.

The technical result is to increase the accuracy of determining diagnostic markers during clinical studies. Additionally, the technical result will be to increase the speed of processing data obtained using liquid chromatography-mass spectrometry and identifying diseases in patients. An additional technical result is an increase in the performance of the computing system when solving a given problem (i.e., it allows processing to obtain a result (product) in less time), thereby reducing the load on the central processor of the computing device by reducing the number of processed requests

The claimed result is achieved by implementing a method for determining diagnostic marker compounds during clinical studies by processing chromatography-mass spectrometric data, performed on a computing device that contains a processor and memory storing instructions executed by the processor and including the following steps:

receive biological samples obtained or isolated from the patient;

carry out the distribution of biological samples from different clinical groups evenly between different analyzed batches;

carry out gas chromatography-mass spectrometric analysis of the received batches;

the results of chromatography-mass spectrometric analysis are sent to a computing device, where the received data is preprocessed and at least one compound is identified from the obtained data;

autoscaling the peak areas in at least one compound obtained in the previous step;

the obtained autoscaling results are fed to the input of the trained classification model, the output is a set of connections - potential markers, while the classification model is built on the basis of orthogonal projections on hidden structures;

select compounds - potential markers from the set selected at the previous stage, based on the Akaike information criterion;

carry out step-by-step removal of connections from the generated set of connections - potential markers whose zero coefficient value is greater than a predetermined value.

In a particular embodiment of the proposed solution, the biological sample is a lipidomic, metabolomic or peptidomic extract of blood, plasma, blood serum, epithelial scraping, or biopsy material.

In another particular embodiment of the proposed solution, chromatographic analysis is performed on a reverse phase column.

In another particular embodiment of the proposed solution, chromatographic analysis is performed on a normal-phase column.

In another particular embodiment of the proposed solution, chromatographic analysis is performed on a hydrophilic column.

In another particular embodiment of the proposed solution, mass spectrometric analysis is performed using dependent scanning.

In another particular embodiment of the proposed solution, mass spectrometric analysis is performed using independent scanning.

In another particular embodiment of the proposed solution, mass spectrometric analysis is performed without the use of scanning.

In another particular embodiment of the proposed solution, the marginal probability of a zero coefficient value is greater or less than 0.05.

In another particular embodiment of the proposed solution, the boundary value of the difference between the vector of coefficients is greater or less than 0.0001.

DESCRIPTION OF DRAWINGS

The implementation of the invention will be described further in accordance with the accompanying drawings, which are presented to explain the essence of the invention and in no way limit the scope of the invention. The following drawings are attached to the application:

Figure 1 illustrates the distribution of biological samples that were not normalized in the space of three principal components. Group 1 – normal breast tissue from patients without metastasis, group 2 – normal breast tissue from patients with metastasis, group 3 – tumor breast tissue from patients without metastasis, group 4 – tumor breast tissue from patients with metastasis, group qc – group of quality control samples.

Figure 2 illustrates the distribution of biological samples after autoscaling in the space of three principal components. Group 1 – normal breast tissue from patients without metastasis, group 2 – normal breast tissue from patients with metastasis, group 3 – tumor breast tissue from patients without metastasis, group 4 – tumor breast tissue from patients with metastasis, group qc – group of quality control samples.

Figure 3 illustrates lipids selected as markers in tumor tissue by the proposed method.

Figure 4 illustrates lipids selected as markers in normal tissue by the proposed method.

Figure 5 illustrates an example of a circuit diagram for the operation of a computing device.

DETAILED DESCRIPTION OF THE INVENTION

The following detailed description of the invention sets forth numerous implementation details designed to provide a clear understanding of the present invention. However, it will be apparent to one skilled in the art how the present invention can be used with or without these implementation details. In other cases, well-known methods, procedures and components have not been described in detail so as not to unduly obscure the features of the present invention.

In addition, from the above discussion it will be clear that the invention is not limited to the above implementation. Numerous possible modifications, alterations, variations and substitutions, while retaining the spirit and form of the present invention, will be apparent to those skilled in the art.

The following terms and definitions apply to this application unless otherwise expressly stated. References to techniques used in the description of this invention refer to well known techniques, including modifications of these techniques and replacement thereof with equivalent techniques known to those skilled in the art.

As used herein, the terms “includes” and “including” are interpreted to mean “including, but not limited to.” These terms are not intended to be construed as “consisting only of.”

The term “biological sample” or “specimen” includes blood, plasma, serum, epithelial scraping, tissue obtained by biopsy from a patient, or other solid or liquid biological material.

The term "patient" refers to a person.

The term “batch” refers to a group of samples that are analyzed continuously.

The term “clinical group” refers to a group of samples united by a common diagnosis of patients from whom the biological sample was collected.

Unless otherwise defined, technical and scientific terms in this application have their standard meanings commonly accepted in the scientific and technical literature.

The proposed method for determining diagnostic marker compounds during clinical studies by processing chromatography-mass spectrometric data includes the following steps:

• obtain biological samples from the patient;

• create batches for analysis, with such a distribution of biological samples between them that the relative number of biological samples from a particular clinical group is the same in each batch. For example, if 20 Group 1 samples (normal breast tissue from patients without metastasis) and 28 Group 2 samples (normal breast tissue from patients with metastasis) are required, then it is preferable to create 2 batches containing 10 samples from each group 1 (normal breast tissue from patients without metastasis) and 14 samples of group 2 (normal breast tissue from patients with metastasis), or 4 lots containing 5 samples of group 1 (normal breast tissue from patients without metastasis) and 7 samples each from group 2 (normal breast tissue from patients with metastasis);

• performing chromatography-mass spectrometric analysis of created batches using liquid chromatography and high-resolution mass analyzers and soft ionization methods (for example, chemical ionization, electrospray). Chromatographic analysis can be performed on a reverse phase column, a normal phase column, or a hydrophilic column. Mass spectrometric analysis can be performed using dependent scanning or using independent scanning. The analysis results are written to a file and transferred to a computing device using network communication tools. Chromatography-mass spectrometric analysis data can be obtained from the outside;

carry out the configuration of a computing device, where the core of the operating system, as well as at least two threads, are allocated for processing the data obtained as a result of chromatography-mass spectrometric analysis, as well as at least two threads, while doing the following:

• preprocessing of data obtained as a result of chromatography-mass spectrometric analysis and identification of compounds;

• autoscaling of peak areas in preprocessed data;

• construction of a classification model and selection of a set of compounds as a result of the classification model;

• selection of compounds from the set created in the previous paragraph, based on the Akaike information criterion;

• step-by-step removal of connections from the generated set of connections for which the probability of a zero coefficient value is greater than a threshold value.

Data preprocessing can be carried out using freely distributed software packages MzMine or XCMS. The purpose of data preprocessing is to obtain a table containing information about the mass-to-charge ratio of registered ions obtained as a result of gas chromatography-mass spectrometric analysis, their release times and peak areas to which the corresponding ions belong, for each sample.

Identification of compounds can be based on precise ion masses and/or release times, as well as information on fragmentation spectra obtained from mass spectrometric analysis if tandem mass spectrometry was performed in a dependent or independent scan format. For identification, the HMDB, Lipid MAPS, SMDB, Metlin libraries, as well as databases created in the laboratory based on the above, can be used.

Autoscaling is carried out to eliminate differences between batches.

Autoscaling of peak areas in preprocessed data was carried out using the formula , where p is the peak number, s is the sample number, b is the batch number, <I _p,b > and sd(I _p,b ) is the average value and standard deviation of peak area p for batch b, <I _p > and sd (I _p ) – average value and standard deviation of peak area p for all samples. I _p,s,b and I* _p,s,b – values of peak area p from sample s of batch b before and after normalization. As a result of autoscaling, a data set is created about the belonging of each sample to a specific clinical group and the peak area values of each identified compound.

A data set containing information about the membership of each sample in a specific clinical group and the peak area values of each identified compound after autoscaling was used to build a “pathology/no pathology” classification model based on orthogonal projections onto hidden structures:

1. Information about the relative intensity of peaks in samples is presented as a matrix of independent variables n*mX, where n is the number of samples, m is the number of compounds. Information about the clinical state of the sample is presented as a column of response variables y of height p, where 0 denotes the “control” state, 1 – the “disease” state;

2. Pareto scaling of the matrix of independent variables is performed (1), where – i-th column of matrix X , is the average value of the i-th column and – standard deviation of variables in the i-th column;

3. Pareto scaling of the dependent variable column is performed (2), where – average value of response variables, – standard deviation of response variables;

4. Weights for independent variables are calculated (3);

5. The calculated vector is normalized (4);

6. Predictive scores are calculated (5);

7. The predictive load of independent variables is calculated (6);

8. The vector of orthogonal loads is calculated (7);

9. The calculated vector of orthogonal loads is normalized (8);

10. Orthogonal accounts are calculated (9);

11. The orthogonal loading of the independent variables is calculated (10);

12. Data that does not contain an orthogonal component is calculated (eleven)

13. Predictive scores are calculated from independent variables that do not contain an orthogonal component (12);

14. The predictive load from independent variables that do not contain an orthogonal component is calculated (13);

15. A vector containing a predictive load from independent variables without an orthogonal component is normalized (14);

16. A vector containing the values of the projections of the variable is calculated (15).

The obtained variable projection (PP) values were used as a criterion for selecting potential marker compounds. A potential marker compound is a compound whose level makes it possible to characterize the clinical identity of the analyzed biomaterial.

In a particular embodiment of the invention, the lower limit of the PP for classification as potential markers was 1. Other values of the minimum PP for classifying compounds as potential markers can be determined.

From the set of compounds selected by the PP value in the previous paragraph, the compounds leading to the maximum increase in the Akaike information criterion (AIC) were selected one at a time at each iteration of variable selection according to an algorithm containing the stages of calculating the AKA (Stage 1), selecting the maximum AKA ( Stage 2) and comparison of the old and new ICA (Stage 3):

A variable is randomly selected from a set of variables formed on the basis of the PP values. Next comes the ICA calculation stage (1) – 6))

The log-likelihood function is constructed (16), where – response variable taking values 0 or 1, is the combined vector of unity and independent variables, is the combined vector of the free term and coefficients of the variables.

The function is differentiated by , obtaining the equations:

(17).

1) The second derivative is calculated:

(18).

2) Based on the Newton-Raphson method, the vector is calculated

where k is the iteration number, X is the matrix of the unit vector and independent variables, W is the diagonal matrix with elements , y is the response variable vector and p is the probability vector . Calculation happens while . In a particular embodiment of the invention, the relative difference in vector modules between iterations from zero = 0.0001. Other values of the significance limit for the difference between coefficients and zero can be determined.

3) Substitute the calculated values into the log-likelihood function from step 1).

4) Calculate the information criterion using the formula (19), where N is the number of independent variables involved in the regressions.

Repeat Stage 1 for all m variables.

Select the variable for which the calculated AIC value will be maximum (Step 2). This value is designated as AIC'.

Perform Stage 1 and Stage 2 for the combination of “previously selected variable + each of the remaining variables.”

Compare AIC with AIC’ value (Step 3).

If AIC from step 5 is greater than AIC’, repeat Steps 1-3, having as constant variables the variables selected earlier and denoting as AIC’ the value from step 5. If AIC from step 5 is less than AIC’, step 7.

We use the variables for which AIC’ was obtained and the coefficients calculated for them further.

Next, they check for statistically significant inequality of coefficients with variables equal to zero with the removal of variables that do not satisfy this condition:

1. Using the previously selected variables, a log-likelihood function was constructed (20), where – response variable taking values 0 or 1, is the combined vector of unity and independent variables, is the combined vector of the free term and coefficients of the variables.

2. The function is differentiated by , obtaining the equations

(21)

3. The second derivative is calculated

(22)

4. Based on the Newton-Raphson method, the vector is calculated

where k is the iteration number, X is the matrix of the unit vector and independent variables, W is the diagonal matrix with elements , y is the response variable vector and p is the probability vector . Calculation happens while . In a particular embodiment of the invention, the relative difference in vector modules between iterations from zero = 0.0001. Other values of the significance limit for the difference between coefficients and zero can be determined.

5. Substituting the calculated values into the matrix (23), calculate the standard error values for the coefficient (24), where j is the serial number of the coefficient in the vector .

6. Calculate the probability of the coefficient differing from zero (25), where – distribution of the square of the independent standard normal random variable θ.

7. If exists , Where is a certain critical value and i > 0, then the variable corresponding , is excluded from the involved set of variables and steps 1-6 are repeated.

In a particular embodiment of the invention, the significance limit of the difference from zero is = 0.05. Other values of the significance limit for the difference between coefficients and zero can be determined.

The following examples of implementation of the method are given in order to disclose the characteristics of the present invention.

Example 1. Normalization of data obtained during gas chromatography-mass spectrometric analysis of breast biopsy material.

Biological samples, namely biopsies of tumor breast tissue and normal breast tissue, were taken from 40 patients with breast cancer without regional metastasis and 48 patients with breast cancer with regional metastasis. Lipids were isolated from tissues using the Folch method [Folch J, Lees M, Sloane Stanley GH. A simple method for the isolation and purification of total lipides from animal tissues. J Biol Chem. 1957;226:497–509].

The obtained biopsy materials of breast tissue were divided into four groups:

Group 1 – normal breast tissue from patients without metastasis;

Group 2 – normal breast tissue from patients with metastasis;

Group 3 – breast tumor tissue from patients without metastasis;

Group 4 – breast tumor tissue from patients with metastasis.

Lipid tissue extracts were divided into three batches:

• 16 samples from group 1, 14 samples from group 2, 16 samples from group 3, 14 samples from group 4;

• 10 samples from group 1, 20 samples from group 2, 10 samples from group 3, 20 samples from group 4;

• 14 samples from group 1, 14 samples from group 2, 14 samples from group 3, 14 samples from group 4.

Based on 10 µl. A quality control sample was generated from each extract.

The separation of lipid extracts was carried out on a Dionex UltiMate 3000 chromatograph (Thermo Scientific, Bremen, Germany) using a Zorbax C18 reverse phase column (length 150 mm, internal diameter 2.1 mm, particle size 5 μm, Agilent, USA) and the following eluents as a mobile phases: eluent A - acetonitrile/water (60/40, v/v) with the addition of 0.1% formic acid and 10 mM ammonium formate; eluent B - acetonitrile/isopropanol/water, (90/8/2, o/o/o), with the addition of 0.1% formic acid and 10 mM ammonium formate. Flow rate 35 µl/min, column temperature 50 ^o C. The proportion of gradient B was changed according to a given algorithm: 0-0.5 min - 30% B, increased to 99% until the 20th minute and maintained the value until the 30th minute and for half a minute returned to 30%. Mass spectrometric analysis was performed using a Maxis Impact instrument with the following settings: range 100-1800 m/z, capillary voltage 4.1 kV in positive ion mode, nebulizer gas pressure 0.7 bar, drying gas flow rate 6 l/min and temperature 200 ^about S.

Tandem mass spectrometry analysis was performed using dependent scanning, in which, after spectrum acquisition, fragmentation spectra were collected at a collision energy of 35 eV of the compounds producing the five most intense peaks in the spectrum, with an isolation window of 5 Da and an exclusion time of 2 minutes.

Quality control samples were analyzed every 10 samples tested.

Data obtained during the analysis in the form of .d files were converted into MzXml format using msConvert software (Proteowizard, 3.0.9987) and preprocessed using the algorithm provided by Koelmel, MzMine software [Pluskal T, Castillo S, Villar-Briones A, Orešič M. MZmine 2: Modular framework for processing, visualizing, and analyzing mass spectrometry-based molecular profile data. BMC Bioinformatics. 2010;11. doi:10.1186/1471-2105-11-395]. Lipid identification was performed using Lipid Match by Koelmel [Koelmel JP, Kroeger NM, Ulmer CZ, Bowden JA, Patterson RE, Cochran JA, et al. LipidMatch: An automated workflow for rule-based lipid identification using untargeted high-resolution tandem mass spectrometry data. BMC Bioinformatics. 2017;18:1–11. doi:10.1186/s12859-017-1744-3]. Ion nomenclature was used according to Lipid Maps terminology in abbreviated form [Sud M, Fahy E, Cotter D, Brown A, Dennis EA, Glass CK, et al. LMSD: LIPID MAPS structure database. Nucleic Acids Res. 2007;35:527–532. doi:10.1093/nar/gkl838].

Data processing was carried out by calculating for each batch: the average peak intensity value of each compound and standard deviation of the peak intensity of each compound , Where – number of samples in batch b, - intensity of peak p in sample i of batch b; for the entire data set, the average peak intensity values for each compound were calculated and standard deviation of the peak intensity of each compound , Where – total number of samples - intensity of peak p in sample i ;

Based on these values, new values for each peak in each sample were calculated using the formula , i.e. carried out autoscaling of the obtained data.

Changes in the distribution of sample coordinates in the first three coordinates of the principal components are presented in Figures 1 and 2. After normalization, the relative deviation of the total ion current value for quality control samples decreased from 7% to 4%.

Example 2. Selection of markers of regional metastasis from biopsy material of breast tumor tissue

Based on the data obtained in example 1, the contribution of compounds to the separation of tumor tissue samples with and without metastasis in the space of principal components oriented along the variance of the dependent variable and orthogonal to it was determined based on the work of the classifier, which was trained using formulas (1)-( 15). Of the 317 identified compounds, 54 had an DI value greater than 1. These compounds predominantly belong to the classes of triacylglycerides (31), phosphatidylcholines (14), sphingomyelins (6) and diagcylglycerides (3).

Step-by-step selection of variables according to the Akaike information criterion using formulas (16) – (19) led to the selection (connections are given in the order of addition to the model) SM 18:2/16:0 (ICA for the model with SM 18:2/16:0 - 112.11), SM 18:1/16:0 (ICA for model with {SM 18:2/16:0, SM 18:1/16:0} -90.26), SM 18:0/24: 1 (IKA for model with {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1} -83.22), PC 16:0_18:2 (IKA for model with {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2 } -77.69), TG 12:0_14:1_18:2 (IKA for model with {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2, TG 12:0_14:1_18:2} - 73.70), TG 12:0_16:1_18:2 (ICA for model with {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2 , TG 12:0_14:1_18:2, TG 12:0_16:1_18:2} -65.19), TG 10:0_12:0_18:2 (ICA for model with {SM 18:2/16:0, SM 18:1 /16:0, SM 18:0/24:1, PC 16:0_18:2, TG 12:0_14:1_18:2, TG 12:0_16:1_18:2, TG 10:0_12:0_18:2} -45.04 ), PC 16:0_18:1 (ICA for model with {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2, TG 12 :0_14:1_18:2, TG 12:0_16:1_18:2, TG 10:0_12:0_18:2, PC 16:0_18:1} -18.00).

For a model based on connections {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2, TG 12:0_14:1_18:2 , TG 12:0_16:1_18:2, TG 10:0_12:0_18:2, PC 16:0_18:1}, the probability of zero coefficients for variables calculated using formulas (20)-(25) was {0.98534, 0.98534, 0.98537, 0.98535, 0.98537, 0.98538, 0.98537, 0.98537} respectively. The highest probability of the coefficient being equal to zero for the variable was for TG 12:0_16:1_18:2 (p = 0.98538). By excluding this variable from the set of variables involved in the model, we recalculate the probability of zero coefficients for a new set of variables {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2, TG 12:0_14:1_18:2, TG 10:0_12:0_18:2, PC 16:0_18:1} and we get {0.00001, 0.00005, 0.02287, 0.00790, 0.03214, 0.14315, 0.15197} respectively. The coefficient with the variable PC has the highest probability of being equal to zero 16:0_18:1 (p = 0.15197). By excluding this variable from the set of variables involved in the model, we recalculate the probability of zero coefficients for a new set of variables {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2, TG 12:0_14:1_18:2, TG 10:0_12:0_18:2} and we get {0.000005, 0.000017, 0.032705, 0.009234, 0.041054, 0.1672568 } respectively. The coefficient with the TG variable has the highest probability of being equal to zero 10:0_12:0_18:2 (p = 0.1672568). By excluding this variable from the set of variables involved in the model, we recalculate the probability of zero coefficients for a new set of variables {SM 18:2/16:0, SM 18:1/16:0, SM 18:0/24:1, PC 16:0_18:2, TG 12:0_14:1_18:2}. The probability of coefficients being zero for this set of variables is {0.00001, 0.00002, 0.03959, 0.00373, 0.02490}. We obtain a model with statistically significantly different coefficients from zero for the variables (Figure 3).

Testing the model using the method of internal cross-validation for a separate object (a method in which the model built on the basis of M-1 samples is tested M times on the remaining sample, where M is the number of samples) gave an area under the operating curve of 0.81, sensitivity and specificities of 94% and 65% at a threshold of 0.39.

Example 3. Selection of markers of regional metastasis from biopsy material of normal breast tissue

Based on the data obtained in example 1, the contribution of compounds to the separation of normal tissue samples with and without metastasis was determined in the space of principal components oriented along the variance of the dependent variable and orthogonal to it using formulas (1)-(15). Of the 317 identified compounds, 60 the PP value turned out to be greater than 1. These compounds belong to the classes of lyso- and phosphotidylcholines (20), triacylglycerides (18), diagcylglycerides (12), sphingomyelins (6), and phosphotidylethanolamines (4).

Step-by-step selection of variables according to the Akaike information criterion using formulas (16) – (19) led to the selection (compounds are given in the order of addition to the model) TG 10:0_18:1_18:3 (ICA for the model with TG 10:0_18:1_18:3 - 103.5), PC O-16:1/18:1 (ICA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1} -96.48), DG 18: 0_18:1 (ICA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1} -92.43), SM d18:1/18:0 (ICA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18:1/18:0} -90.2), LPC 16 :0 (ICA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18:1/18:0, LPC 16:0} - 87.62), TG 12:0_18:1_8:0 (ICA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18:1 /18:0, LPC 16:0, TG 12:0_18:1_8:0} -85.65), TG 10:0_18:2_18:2 (ICA for model with {TG 10:0_18:1_18:3, PC O -16:1/18:1, DG 18:0_18:1, SM d18:1/18:0, LPC 16:0, TG 12:0_18:1_8:0, 10:0_18:2_18:2} -75, 60), OxTG 18:1_18:2_18:3(OH) (IKA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18: 1/18:0, LPC 16:0, TG 12:0_18:1_8:0, 10:0_18:2_18:2, OxTG 18:1_18:2_18:3(OH)} -62.17), PC P-16 :0/20:4 (IKA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18:1/18:0, LPC 16 :0, TG 12:0_18:1_8:0, 10:0_18:2_18:2, OxTG 18:1_18:2_18:3(OH), PC P-16:0/20:4} -55.63), PC 12:0_14:1 (IKA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18:1/18:0, LPC 16: 0, TG 12:0_18:1_8:0, 10:0_18:2_18:2, OxTG 18:1_18:2_18:3(OH), PC P-16:0/20:4, PC 12:0_14:1} - 48.87), DG 18:2_18:2 (IKA for model with {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18:1/18 :0, LPC 16:0, TG 12:0_18:1_8:0, TG 10:0_18:2_18:2, OxTG 18:1_18:2_18:3(OH), PC P-16:0/20:4, PC 12:0_14:1, DG 18:2_18:2} -24).

For a model based on connections {TG 10:0_18:1_18:3, PC O-16:1/18:1, DG 18:0_18:1, SM d18:1/18:0, LPC 16:0, TG 12:0_18:1_8:0, TG 10:0_18:2_18:2, OxTG 18:1_18:2_18:3(OH), PC P-16:0/20:4, PC 12:0_14:1, DG 18: 2_18:2}, the probability of zero coefficients for variables calculated using formulas (20) - (25) was {0.9956, 0.9973, 0.9964, 0.9970, 0.9968, 0.9972, 0, 9959, 0.9970, 0.9967, 0.9967, 0.9966} respectively. The coefficient with PC O-16:1/18:1 had the highest probability of being equal to zero (p = 0.9973). Excluding this variable, we build a model based on connections {TG 10:0_18:1_18:3, DG 18:0_18:1, SM d18:1/18:0, LPC 16:0, TG 12:0_18:1_8:0, TG 10:0_18:2_18:2, OxTG 18:1_18:2_18:3(OH), PC P-16:0/20:4, PC 12:0_14:1, DG 18:2_18:2}, we obtain the following probabilities of equality zero coefficients for variables: {0.003, 0.010, 0.039, 0.003, 0.007, 0.012, 0.004, 0.012, 0.021}. We obtain a model with statistically significantly different coefficients from zero for the variables (Figure 4). Testing of the model through internal cross-validation on a single subject yielded an area under the operational curve of 0.79, sensitivity and specificity of 88% and 58% at a threshold of 0.15.

In FIG. 5 will further present a general diagram of a computing device (500) that provides data processing necessary to implement the claimed solution.

In general, the device (500) includes components such as: one or more processors (501), at least one memory (502), data storage means (503), input/output interfaces (504), I/O means ( 505), networking tools (506).

The processor (501) of the device is configured to perform the computational operations necessary to implement the proposed method. The processor (501) executes the necessary machine-readable instructions contained in the main memory (502). The processor (501) includes at least two cores and at least one thread. In addition, to carry out data processing according to this solution, one core and at least two threads are allocated.

Memory (502), as a rule, is made in the form of RAM and contains the necessary program logic that provides the required functionality.

The data storage medium (503) can be in the form of HDD, SSD drives, raid array, network storage, flash memory, optical storage devices (CD, DVD, MD, Blue-Ray disks), etc. The means (503) allows long-term storage of various types of information, for example, the aforementioned files with user data sets, a database containing records of time intervals measured for each user, user IDs, etc.

Interfaces (504) are standard means for connecting and working with the server part, for example, USB, RS232, RJ45, LPT, COM, HDMI, PS/2, Lightning, FireWire, etc.

The choice of interfaces (504) depends on the specific design of the device (500), which can be a personal computer, mainframe, server cluster, thin client, smartphone, laptop, etc.

The data I/O means (505) in any embodiment of a system implementing the described method may use a keyboard. The hardware design of the keyboard can be any known: it can be either a built-in keyboard used on a laptop or netbook, or a separate device connected to a desktop computer, server or other computer device. The connection can be either wired, in which the keyboard connecting cable is connected to the PS/2 or USB port located on the system unit of the desktop computer, or wireless, in which the keyboard exchanges data via a wireless communication channel, for example, a radio channel, with base station, which, in turn, is directly connected to the system unit, for example, to one of the USB ports. In addition to the keyboard, I/O data tools can also include: joystick, display (touch display), projector, touchpad, mouse, trackball, light pen, speakers, microphone, etc.

Network communication means (506) are selected from a device that provides network reception and transmission of data, for example, an Ethernet card, WLAN/Wi-Fi module, Bluetooth module, BLE module, NFC module, IrDa, RFID module, GSM modem, etc. Using the means (505), the organization of data exchange is ensured via a wired or wireless data transmission channel, for example, WAN, PAN, LAN, Intranet, Internet, WLAN, WMAN or GSM.

The device components (500) are interfaced via a common data bus (510).

In these application materials, a preferred disclosure of the implementation of the claimed technical solution was presented, which should not be used as limiting other, private embodiments of its implementation, which do not go beyond the scope of the requested scope of legal protection and are obvious to specialists in the relevant field of technology.

Claims (9)

A method for identifying potential diagnostic marker compounds during clinical trials by processing chromatography-mass spectrometric data, performed on a computing device that contains a processor and memory storing instructions executed by the processor and including the following steps: obtain samples of biological material obtained or isolated from the patient; carry out uniform distribution of samples from each clinical group between different analyzed batches; carry out gas chromatography-mass spectrometric analysis of the received batches; the results of chromatography-mass spectrometric analysis are sent to a computing device, where the received data is preprocessed and at least one compound is identified from the obtained data; aligning peak area values between batches by autoscaling in at least one compound obtained in the previous step; the obtained peak area values are fed to the input of a trained classification model that uses orthogonal projections onto hidden structures, and the output is a set of compounds - potential markers; select compounds - potential markers from the set selected at the previous stage, based on the Akaike information criterion; carry out step-by-step removal of compounds from the generated set of compounds - potential markers, for which the probability value of the coefficient differing from zero is less than the value of the probability limit of the coefficient differing from zero.