Automatic correction method for large-scale multi-batch LC-MS original data

Abstract

The invention relates to an automatic correction method for large-scale multi-batch LC-MS original data. According to the method, automatic retention time and mass-to-charge ratio correction is carried out on the large-scale multi-batch LC-MS original data through multiple steps. The method comprises the following steps: firstly, selecting a quality control sample as a reference file, constructing a high-response region in the reference file as an anchor point region, then obtaining an extracted ion current diagram from a target file, performing cross-correlation operation on the anchor point region and the extracted ion current diagram to realize matching of the position, and obtaining a large number of matching results from a large number of anchor point regions. And outlier matching is removed through distribution of matching results, so that reliable matching is obtained. And finally, performing nonlinear fitting on the matching result of the retention time and the mass-to-charge ratio to obtain a correction curve, thereby realizing direct correction of the original data of the sample.

Description

Technical Field

[0001] The invention relates to the field of LC-MS detection and signal processing, and is an LC-MS raw data correction method based on signal cross-correlation. Background Art

[0002] Non-targeted metabolomics based on LC-MS is an important means to study diseases, drug metabolism, environmental exposure and health. It aims to conduct comprehensive qualitative and quantitative analysis of small molecules in biological samples. In recent years, with the development of precision medicine and cohort studies, the demand for large-scale metabolomics has gradually increased. Large-scale population metabolomics helps to discover epidemiological factors and metabolic associations and promote disease prevention and diagnosis research. However, in large-scale LC-MS analysis, instrument fluctuations are difficult to avoid. Fluctuations in chromatographic retention time and mass spectrometry response will have a significant impact on subsequent data processing and qualitative analysis (Analytical Chemistry 2017 89(3)1399-1404DOI:10.1021 / acs.analchem.6b04337). In addition, large-scale LC-MS analysis is often divided into multiple batches for analysis, and the correction of large deviations in different batches is crucial.

[0003] Traditional LC-MS data correction is mainly performed by the Obiwarp algorithm in XCMS (Analytical Chemistry 2006 78 (17) 6140-52 DOI: 10.1021 / ac0605344), which is based on a dynamic programming algorithm to time-match two samples. However, this method has large errors in cases with large deviations and cannot calibrate the instrument m / z, so it is not suitable for multiple batches of data correction. In recent years, many batch correction algorithms have been developed for multiple batches of metabolomics data, but most of them are based on peak table matching, so the processing parameters between different data sets will also have a significant impact on the final statistical results. Summary of the invention

[0004] The present invention provides a correction method for large-scale multi-batch LC-MS raw data based on local signal cross-correlation. In order to achieve the purpose of the present invention, a quality control sample is selected from the samples to be processed as a reference sample, and other files are used as target files to be corrected. An anchor point region is constructed in the reference file, and an extracted ion current map is obtained in the target file according to the m / z of the anchor point region. The matching of the anchor point and the extracted ion current map is obtained by signal cross-correlation, and the erroneous matching points are removed by the residual distribution of robust linear fitting, and finally a correction function is obtained by nonlinear fitting and a corrected new file is obtained.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a large-scale multi-batch LC-MS raw data automatic correction method, comprising the following steps:

[0006] The first step is to obtain the LC-MS raw data of the sample to be calibrated by mass spectrometry export, select a sample file as the reference file, and other sample files as the target files;

[0007] Step 2: construct the anchor area ROA in the reference file;

[0008] Step 3: Obtain the extracted ion chromatogram XIC corresponding to the m / z of ROA in the target file;

[0009] Step 4: Perform cross-correlation operation on the ROA intensity vector of the reference file and the XIC intensity vector of the target file to obtain the matching points of retention time and m / z;

[0010] Step 5: Perform nonlinear fitting on the matching points to obtain a calibration curve of retention time and m / z;

[0011] Step 6: For the original data of the target file to be corrected, the retention time and m / z in the target file are mapped by the calibration curve function to obtain a corrected new file.

[0012] The anchor region ROA is a mass spectrometry signal of a segment of m / z in the LC-MS raw data with a response higher than a threshold within a set continuous time range, and the mass spectrometry signal includes three vectors of the same length, namely a time series, an m / z series, and an intensity series.

[0013] The construction of the anchor area ROA in the reference file comprises the following steps:

[0014] Traverse all primary mass spectrometry scan spectra in the reference file;

[0015] For each spectrum, if there is a point whose intensity exceeds the ROA threshold, the time, m / z, and intensity of the point are recorded;

[0016] Taking the time of this point as the center, the set retention time range is extended to both sides; the time, m / z and intensity of all points within the retention time range and the m / z deviation range of this point are recorded to form the ROA anchor point.

[0017] The extracted ion current XIC is a mass spectrum intensity signal of a specific m / z at all scanning times, and the mass spectrum intensity signal includes three vectors of the same length, namely a time series, an m / z series, and an intensity series.

[0018] The step of obtaining an extracted ion current XIC corresponding to the m / z of ROA in the target file comprises the following steps:

[0019] Traverse all primary mass spectrometry scan spectra in the target file;

[0020] For each spectrum, if there is a point in the set m / z range, the time, m / z and intensity of the point are recorded; if there is no point in the set m / z range, the time, m / z and intensity of the point are recorded;

[0021] The time series, m / z sequence and intensity obtained after traversing all the primary mass spectrometry scan spectra constitute the ion current graph XIC;

[0022] For all ROAs obtained from the reference file, the XIC was obtained by averaging the m / z sequence of each ROA, thus constructing the XIC corresponding to the ROA one by one.

[0023] The cross-correlation operation of the ROA intensity vector of the reference file and the XIC intensity vector of the target file comprises the following steps:

[0024] The similarity measurement is performed on the ROA intensity vector and the XIC intensity vector of two different lengths to obtain the region in the XIC that is most similar to the ROA, thereby obtaining the matching points of retention time and m / z in the reference file and the target file.

[0025] The nonlinear fitting of the matching points comprises the following steps:

[0026] First, the linear trend of the retention time offset of the target file is obtained through robust linear fitting, and the false matching probability of each matching point is calculated according to the residual distribution of the linear fitting. The points with error probability lower than the threshold are screened, and then nonlinear LOESS fitting is performed to obtain the correction curve of retention time and m / z. The correction curve is used to calibrate the target file by applying the correction curve function to the retention time and m / z of the target file.

[0027] An automatic correction system for large-scale multi-batch LC-MS raw data, comprising:

[0028] A file screening module is used to obtain the LC-MS raw data of the sample to be calibrated by mass spectrometry export, select a certain sample file as a reference file, and other sample files as target files;

[0029] Anchor region construction module, used to construct anchor region ROA in reference file;

[0030] An extracted ion chromatogram (XIC) building module is used to obtain an extracted ion chromatogram (XIC) corresponding to the m / z of ROA in a target file;

[0031] A matching module is used to perform a cross-correlation operation on the ROA intensity vector of the reference file and the XIC intensity vector of the target file to obtain a matching point of retention time and m / z;

[0032] A matching point fitting module is used to perform nonlinear fitting on the matching points to obtain a calibration curve of retention time and m / z;

[0033] The correction module is used to map the retention time and m / z of the target file to be corrected through a correction curve function to obtain a corrected new file.

[0034] A large-scale multi-batch LC-MS raw data automatic correction device comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement the large-scale multi-batch LC-MS raw data automatic correction method when executing the computer program.

[0035] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the automatic correction method for large-scale multi-batch LC-MS raw data is implemented.

[0036] The present invention has the following beneficial effects and advantages:

[0037] 1. The present invention proposes a retention time and m / z correction method for LC-MS raw data, which can realize data correction of large-scale and multi-batch samples. The corrected data can be used for subsequent feature extraction and matching, thereby improving data quality.

[0038] 2. The present invention does not need to establish a reference compound database, has no specific requirements for sample information and composition, and does not require manual peak shape inspection. By automatically detecting matching anchor points, building a large number of matching anchor points, and automatically removing false matches, automatic matching of large batches of unknown samples can be achieved.

[0039] 3. The present invention can effectively correct the sample data in large-scale metabolomics. For different nonlinear retention times and m / z offsets in large-scale samples, after correction with reference samples, the large-scale sample data can be processed in the same downstream software, avoiding the tedious batch processing of large-scale samples and data loss.

[0040] 4. The present invention can correct different batches of data with large deviations. The integration of different batches of metabolomics data can effectively improve the scale of metabolomics research. The correction of data from different batches and different instruments can realize the integration of a large amount of metabolomics data, which is also of great significance for the batch collection and processing of large-scale samples.

[0041] 5. The calibration steps of the present invention are simple, and the parameters that need to be controlled include reference file selection, ROA intensity threshold, and m / z deviation of ROA and XIC. It is easy to operate and realize automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 (a) Matching anchor point distribution and LOESS fitting regression curve between reference file and target file; (b) Retention time deviation of target file; (c) m / z deviation of target file;

[0043] Figure 2 Total ion currents of the reference file and the target file before (a) and after (b) calibration;

[0044] Figure 3 Characteristic peak distribution of compounds tryptophan-d5 (a), carnitine C6:0-d3 (b), LPE 16:0 (c), PC O 36:5 (d), and PC 38:4 (e) before and after correction;

[0045] Figure 4 Correction deviation of retention time of FI200 data set with reference file;

[0046] Figure 5 False negative (identified after correction) features of the FI200 dataset in XCMS processing before correction (a, b, c, and d are 4 typical examples respectively);

[0047] Figure 6 Flowchart of LC-MS raw data correction. DETAILED DESCRIPTION

[0048] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0049] The present invention discloses a correction method for large-scale multi-batch LC-MS raw data based on local signal cross-correlation. In the study of non-targeted metabolomics and exposure omics, in order to obtain reliable results, large-scale clinical samples are often included and completed through multiple batches of LC-MS experiments. In this case, controlling data quality is a prerequisite for obtaining accurate and reliable findings. However, in large-scale, multi-batch experiments, chromatographic retention time fluctuations, mass spectrometer detector fluctuations, etc. are difficult to avoid, which brings difficulties and challenges to the subsequent data processing and analysis process. The method realizes automatic retention time and mass-to-charge ratio correction of large-scale, multi-batch LC-MS raw data through multiple steps. First, a quality control sample is selected as a reference file, a high response area is constructed as an anchor area in the reference file, and then an extracted ion flow map is obtained for the target file. The anchor area and the extracted ion flow map are cross-correlated to achieve the matching of the retention time and mass-to-charge ratio at the position, a large number of matching results can be obtained through a large number of anchor areas, and the distribution of matching results removes the wrong matching to obtain a reliable match. Finally, the matching results of the retention time and mass-to-charge ratio are nonlinearly fitted to obtain a calibration curve, thereby realizing direct correction of the sample raw data.

[0050] Among them, large-scale refers to research with a sample size of more than 200, and multi-batch refers to the samples in a study, which are batched at different times using the same instrument or different instruments of the same type, with no limit on the number of batches.

[0051] like Figure 6 As shown, a correction method for large-scale multi-batch LC-MS raw data based on local signal cross-correlation includes the following steps:

[0052] The first step is to select the reference file and the target file. Generally, the quality control sample (QC) file is selected as the reference file. If there is no quality control sample, a general sample file can also be selected as the reference file.

[0053] The second step is to construct regions of anchor (ROA) in the reference file. ROA refers to a mass spectrometry signal with a specific m / z (usually with a deviation of less than 0.005) in a certain continuous retention time range (usually 20 to 50 seconds) with a high response (maximum intensity greater than the ROA threshold, such as the maximum intensity greater than 1e5) in the LC-MS raw data, which is composed of three vectors of the same length (time series, m / z sequence, and intensity sequence). For the primary mass spectrometry scan spectrum in the reference file, from the first scan to the last scan, if there is a point in a spectrum with an intensity exceeding the ROA threshold, the time, m / z, and intensity of the point are recorded. With the time of the point as the center, a certain retention time range (usually 20 to 50 seconds) is extended to both sides, and the time, m / z, and intensity of all scan points in this time region with m / z within the deviation range of the point (usually within 0.005) are recorded (the intensity of the scan time without signal in the m / z range is 0), which is the ROA anchor point. By repeating this process in the reference file, all ROAs that meet the intensity conditions can be found, and the ROAs whose highest intensity is not at the center time are removed. All remaining ROAs can be used for subsequent correction.

[0054] The third step is to obtain the extracted ion chromatograms (XIC) in the target file. XIC refers to the mass spectrum intensity signal of a specific m / z (wider than the m / z range of ROA, usually with a deviation of less than 0.02) at all scanning times in an LC-MS file, and is composed of three vectors of the same length (time series, m / z sequence, intensity sequence). For the primary mass spectrometry scan spectrum in the target file, from the first scan to the last scan, if there is a scan point in the m / z range in a certain scan, the time, m / z and intensity of the point are recorded; if there is no scan point in the m / z range in a certain scan, the time, m / z and intensity of the point are recorded (0 at this time). The time series, m / z sequence and intensity sequence obtained after all scans are completed constitute the XIC. For all ROA obtained from the reference file, the XIC is obtained by taking the mean of their m / z sequences, thereby constructing an XIC corresponding to each ROA.

[0055] The fourth step is to perform a cross-correlation operation on the ROA intensity vector of the reference file and the XIC intensity vector of the target file. Cross-correlation is a method of measuring the similarity of two time series vectors of different lengths (here referring to the ROA intensity vector and the XIC intensity vector) in signal processing. By cross-correlating ROA and XIC, a cross-correlation result with the same length as XIC is obtained. The moment of highest intensity in the result (target file) and the moment of highest intensity in ROA (reference file) are considered to be retention time matching points, and the m / z of the moment of highest intensity in the result (target file) and the m / z of the moment of highest intensity in ROA (reference file) are m / z matching points. Repeat this process for all ROAs to obtain all retention time matching points and m / z matching points.

[0056] The fifth step is to perform nonlinear regression. For the retention time matching points obtained in the third step, a robust linear regression is performed on the retention time of the target file and the retention time of the reference file according to the retention time matching points (the loss function uses the smoothing expression of L1 loss rho(z)=2*((1+z)**0.5-1), z is the residual), so that the fitting residuals of all retention time matching points can be calculated. All residuals are divided by the 68.27% quantile of the residual to obtain the t value of each residual, and the probability (p value) under the two-tailed t distribution with N-2 degrees of freedom (N is the number of ROA) is obtained for each t value. If the p value is less than the probability threshold (usually 0.01), the matching point is considered to be a wrong match. Remove the wrong matching points, and perform nonlinear LOESS (Locally Weighted Scatterplot Smoothing) fitting on the remaining retention time matching points to obtain the retention time nonlinear mapping function of the target file and the reference file. Similarly, LOESS fitting is performed on the remaining m / z matching points to obtain the m / z nonlinear mapping function of the target file and the reference file. The mapping function is applied to the retention time and m / z data of the target file, so that the file is corrected.

[0057] This method is targeted at raw data directly exported from mass spectrometers in large-scale (sample volume exceeds 200) or multi-batch LC-MS studies. For data in different formats exported from different instruments, they can be converted to mzML format in centroid mode using MSConvert (ProteoWizard) software for processing.

[0058] This calibration method can also be used to calibrate data collected from different batches or different liquid chromatography or mass spectrometry, but it requires the use of chromatographic columns with consistent chromatographic retention rules (such as reverse phase columns or normal phase columns) and mobile phases with similar compositions.

[0059] The mapping function obtained in the fourth step can be applied to the m / z and retention time of all scans of the target file to correct the m / z and retention time of the target file. The correction result can be written as a new file in the mzML format through the python module pyOpenMS. The new file can be subsequently processed or analyzed using any software that supports mzML file processing.

[0060] The calibration effect can be qualitatively checked by observing the alignment of the total ion current graphs of the samples. If the calibration effect is not ideal, multiple calibrations can be performed to achieve a more stable and accurate calibration effect in large-scale or multi-batch LC-MS data. That is, after selecting a reference sample to perform the first to fifth steps of calibration for all target samples, select a reference sample again to perform the first to fifth steps of calibration for all samples, thereby performing multiple calibrations.

[0061] The specific implementation scheme adopted by the present invention is as follows:

[0062] The first step is to select the reference file and the target file. From the large-scale or multi-batch LC-MS raw data to be processed, select the quality control (QC) sample as the reference sample and other samples as the target sample. If there are multiple or no QC samples in the samples to be processed, you can manually select samples with higher response and better peak behavior as reference samples by observing the chromatogram of the samples.

[0063] The second step is to construct regions of anchor (ROA) in the reference file. For the primary mass spectrometry scan spectrum in the reference file, from the first scan to the last scan, if there is a point in a spectrum whose intensity exceeds the ROA threshold, the time, m / z and intensity of the point are recorded. With the time of the point as the center, a certain retention time range (usually 20 to 50 seconds) is extended to both sides, and the time, m / z and intensity of all scan points within this time region and the m / z within the deviation range of the point (usually within 0.005) are recorded (the intensity of the scan time without signal in this m / z range is 0), which is the ROA anchor point. Repeating this process in the reference file, all ROAs that meet the intensity conditions can be found, and the ROAs with the highest intensity not at the center time are removed. All remaining ROAs can be used for subsequent correction. For example, 5% of the maximum intensity in this file is set as the intensity threshold, and the region where the local highest point exceeds the threshold is selected as the ROA. The ROA region is often a relatively stable peak substance or signal in the sample or solvent, so it can be used as a reference for correction in other samples of the same biological matrix.

[0064] The third step is to obtain the extracted ion chromatograms (XIC) in the target file. The extraction process of XIC is as follows: for the primary mass spectrometry scan spectrum in the target file, traverse from the first scan to the last scan. If there is a scan point in the m / z range in a certain scan, then record the time, m / z and intensity of the point. If there is no scan point in the m / z range in a certain scan, then record the time, m / z and intensity of the point (0 at this time). The time series, m / z sequence and intensity sequence obtained after all scans are completed constitute the XIC. For the m / z sequence mean of each ROA obtained in the reference file, extract the XIC with a relatively wide m / z range (for example, the deviation is within 0.02) in the target file. The time series range of the XIC is the same as the scan time range of the entire file.

[0065] The fourth step is to perform a cross-correlation operation on the ROA intensity vector of the reference file and the XIC intensity vector of the target file. Cross-correlation is a method of measuring the similarity of two time series vectors of different lengths in signal processing. Here, the ROA intensity vector and the XIC intensity vector are cross-correlated by the "signal.correlate(mode='same')" function in the python module scipy to obtain a cross-correlation result with the same length as XIC.

[0066] Step 5: Remove the wrong matching points. The result of the cross-correlation between ROA and XIC in step 4 is a vector with the same length as XIC (hereinafter referred to as Cor), and its intensity at different positions represents the potential ROA matching area in the target file XIC. If there is only one highest point in Cor or the intensity of the first highest point is 5 times higher than the second, then the moment of the highest point and the moment of the ROA highest point are the first type of retention time matching anchor points. In addition, if there are multiple highest points in Cor, the intensities of the first five highest points are all regarded as the second type of retention time matching anchor points. Since the first type of anchor points are more reliable than the second type of anchor points, a robust linear regression is performed on the first type of anchor points using a soft margin loss function (through the "optimize.least_squares" function in the python module scipy, loss = 'soft_l1', f_scale = 1.0). This regression method can effectively reduce the impact of outliers. In 2006, Harvey et al. proposed a method ROUT for detecting outliers by calculating the false positive rate in nonlinear regression (BMC Bioinformatics 2006 7 DOI:10.1186 / 1471-2105-7-123). This method modified ROUT. That is, the residuals Residual of all Class 1 anchor points are calculated according to the linear regression equation, and the robust standard deviation RSDR of the anchor points is calculated according to formula (1), where P68 is the 68.27% quantile of the residuals and N is the number of Class 1 anchor points. The residuals are sorted in ascending order, and only the Class 1 anchor points with the largest residuals of 30% are considered. For each anchor point, α is calculated according to formula (2), where i is the sorting number and Q is the false positive rate threshold (set to 0.01). The t value is calculated according to formula (3), and the two-tailed probability (p value) of the t value is calculated under the t distribution with N-2 degrees of freedom. If the p value is less than α, the point and the point with a larger residual are considered outliers, and the residual of the point is considered the outlier threshold. The residual of the second type of anchor point is calculated according to the linear equation, and the residual greater than the outlier threshold is considered an outlier.

[0067] RSDR=P68 N / (N-2) (1)

[0068] α_i=(Q(N-(i-1))) / N (2)

[0069] t=Residual / RSDR (3)

[0070] The sixth step is to perform nonlinear fitting (Nonlinear regression). After removing outliers from the first and second category anchor points using the method in the fourth step, the remaining retention time matching points are fitted with nonlinear LOESS (Locally Weighted Scatterplot Smoothing) using the "api.nonparametric.lowess" in the python module statsmodels to obtain the retention time nonlinear mapping function of the target file and the reference file. When LOESS estimates the predicted value, the local 1 / 3 points (the frac parameter in the lowess function is set to 1 / 3) are used for estimation. Similarly, LOESS is fitted on the remaining m / z matching points to obtain the m / z nonlinear mapping function of the target file and the reference file.

[0071] In the seventh step, the retention time and m / z mapping function of the target file and the reference file are obtained according to the fifth step. This function is applied to all the time and m / z values ​​of the target file. Therefore, all the target files can be mapped to the same standard (reference file). The new file can be used for feature extraction and matching using any other LC-MS data processing software. Since the retention time and m / z have been recalibrated, the data quality will be greatly improved.

[0072] The present invention uses the matching of anchor regions to calibrate the original data file based on the similarity of local signals between LC-MS samples. The calibrated file can normally carry out any downstream data processing flow, improving data quality without affecting any downstream operations. Since it does not rely on any known knowledge and sample information, the present invention can calibrate different batches of LC-MS data and is also suitable for continuous analysis of large-scale metabolomics samples.

[0073] Example

[0074] In order to prove the effectiveness and feasibility of the present invention, the present invention is illustrated and verified by large-scale, multi-batch LC-MS data.

[0075] In order to illustrate the calibration process and effect of the method of the present invention in actual LC-MS samples, one sample was selected from each of two metabolomics studies conducted in different years for testing, one human serum sample as a reference sample and the other plasma sample as a target sample. Both samples were analyzed using ultra-high performance liquid chromatography (Thermo Vanquish UHPLC)-high resolution mass spectrometry (Thermo Q Exactive quadrupole-Orbitrap) instruments, and the SOP method (Analytical Chemistry 2018 90 (12), 7635-7643 DOI: 10.1021 / acs.analchem.8b01482) was used to simulate the calibration process of multiple batches of samples with large deviations. First, all primary mass spectrometry scans were traversed in the reference sample, and the region with an intensity exceeding the ROA threshold (0.5% of the maximum intensity of the file) was constructed as the ROA anchor region. The ROA time range was 30s, and the m / z deviation allowed in the ROA was 0.001. A total of 500 ROA anchors were detected. Then, the XIC of all target files was extracted according to the m / z sequence mean of 500 ROA regions with an m / z deviation within 0.02. The potential retention time matching points were obtained by cross-correlating the reference file ROA and the target file XIC. According to the cross-correlation results, a total of 183 first-class anchor points and 199 second-class anchor points were obtained. Robust linear fitting was performed on the retention times of the target file and the reference file of the first-class anchor points, and the fitting residuals of all matching anchor points were calculated. After all the residual points were sorted in descending order, the t value and α were calculated according to formulas 1 to 3 in turn. The p value corresponding to the t value was obtained according to the two-tailed t distribution with 181 degrees of freedom. If the p value of the anchor point is less than the corresponding α, the anchor point is regarded as an incorrect matching point, and the residual of the anchor point is the residual threshold of the incorrect matching point. According to the robust linear fitting equation and the residual threshold, the anchor point closest to the robust linear fitting line and with a residual less than the threshold was selected from the second-class matching anchor points. After removing the incorrectly matched anchor points, the retention time and m / z of the remaining anchor points are nonlinearly fitted to obtain the mapping function of the retention time and m / z of the target file and the reference file. The mapping function is applied to the retention time and m / z of all scan spectra of the target file, and the file is rewritten using the python module pyopenms to obtain the corrected target file. The distribution of the first and second anchor points, the retention time and m / z correction results are shown in Figure 2. Figure 1 After correcting the original data of the target file, the total ion chromatrogram (TIC) of the corrected data is obtained. The TIC diagrams before and after correction are shown in Figure 2As shown in the figure, it can be found that the retention times are clearly aligned after correction. In order to further check the effect of retention time correction, some characteristic peaks of internal standards or endogenous compounds are selected to check the peak alignment effect. The results are shown in the figure. Figure 3 As shown, it is explained that the present invention can effectively correct the large deviation data (within 20s) of different batches.

[0076] In order to illustrate the correction effect of the method on large-scale LC-MS analysis samples, data from a large-scale LC-MS metabolomics study were selected to illustrate the present invention (Frontiers In Immunology 2023 14 DOI:10.3389 / fimmu.2023.1168308). In order to more clearly illustrate the practicality and feasibility of this method, 200 samples were randomly selected from the metabolomics positive ion mode analysis data to form the test data set FI200. One sample was randomly selected in the FI200 data set as a reference sample, and the other samples were all target samples. According to the correction process of the previous two samples, the retention time and m / z of all target samples were corrected in turn. Since this study carried out strict quality control and instrument maintenance during the experiment, the data quality was high, and there was no obvious retention time deviation before correction. The retention time deviation of the original data corrected by this method is shown as follows. Figure 4 As shown in the figure, the retention time deviation of most files is within 1s, and only a small number of samples have a deviation of 3 to 4s. This result is more reasonable under large-scale sample analysis.

[0077] In order to illustrate the necessity of this correction method for large-scale LC-MS metabolomics analysis, XCMS (version: 3.22.0, R version: 4.3.1) was used to extract features, calibrate retention times and match features on the FI200 data set before and after correction, and then obtain the peak table of features. The raw data before and after correction were processed using the same parameters, as follows: CentWaveParam: snthresh = 8, noise = 500, ppm = 15, peakwidth = c (3, 20), mzdiff = 0.01, prefilter = c (3, 50); ObiwarpParam: binSize = 0.5; PeakDensityParam: minFraction = 0.8, bw = 5, binSize = 0.02, minSamples = 160, maxFeatures = 10 (other parameters are default). Finally, XCMS obtained 1766 features before correction and 1921 features after correction. We manually reviewed the features with retention time deviation of 5s and m / z deviation of more than 0.005 between the two peaks, and found that 200 features were missing or mismatched before correction. Figure 5Figures a, b, c, and d show typical examples of four missing features, indicating that even high-quality large-scale LC-MS data can still suffer from insufficient correction capability, information loss, and peak matching errors under traditional data processing methods. Therefore, using this method to correct the data before processing large-scale LC-MS data can effectively improve data quality.

[0078] The above results show that the present invention can realize automatic correction of multi-batch, large-scale metabolomics data by correcting the LC-MS raw data through a large number of local high response signals, which is helpful for peak matching and integral quantification, and at the same time corrects the m / z differences between different mass spectra or batches to improve data quality.

Claims

1. A method for automatic correction of large-scale multi-batch LC-MS raw data, characterized in that: The following steps are involved: The first step is to obtain the LC-MS raw data of the sample to be calibrated by mass spectrometry export, select a sample file as the reference file, and other sample files as the target files; Step 2: construct the anchor area ROA in the reference file; Step 3: Obtain the extracted ion chromatogram XIC corresponding to the m / z of ROA in the target file; Step 4: Perform cross-correlation operation on the ROA intensity vector of the reference file and the XIC intensity vector of the target file to obtain the matching points of retention time and m / z; Step 5: Perform nonlinear fitting on the matching points to obtain a calibration curve of retention time and m / z; Step 6: For the original data of the target file to be corrected, the retention time and m / z in the target file are mapped by the calibration curve function to obtain a corrected new file.

2. The automatic correction method for large-scale multi-batch LC-MS raw data according to claim 1, characterized in that: The anchor region ROA is a mass spectrometry signal of a segment of m / z in the LC-MS raw data with a response higher than a threshold within a set continuous time range, and the mass spectrometry signal includes three vectors of the same length, namely a time series, an m / z series, and an intensity series.

3. The automatic correction method for large-scale multi-batch LC-MS raw data according to claim 1, characterized in that: The construction of the anchor area ROA in the reference file comprises the following steps: Traverse all primary mass spectrometry scan spectra in the reference file; For each spectrum, if there is a point whose intensity exceeds the ROA threshold, the time, m / z, and intensity of the point are recorded; Taking the time of this point as the center, the set retention time range is extended to both sides; the time, m / z and intensity of all points within the retention time range and the m / z deviation range of this point are recorded to form the ROA anchor point.

4. The automatic correction method for large-scale multi-batch LC-MS raw data according to claim 1, characterized in that: The extracted ion current XIC is a mass spectrum intensity signal of a specific m / z at all scanning times, and the mass spectrum intensity signal includes three vectors of the same length, namely a time series, an m / z series, and an intensity series.

5. The automatic correction method for large-scale multi-batch LC-MS raw data according to claim 1, characterized in that: The step of obtaining an extracted ion current XIC corresponding to the m / z of ROA in the target file comprises the following steps: Traverse all primary mass spectrometry scan spectra in the target file; For each spectrum, if there is a point in the set m / z range, the time, m / z and intensity of the point are recorded; if there is no point in the set m / z range, the time, m / z and intensity of the point are recorded; The time series, m / z sequence and intensity obtained after traversing all the primary mass spectrometry scan spectra constitute the ion current graph XIC; For all ROAs obtained from the reference file, the XIC was obtained by averaging the m / z sequence of each ROA, thus constructing the XIC corresponding to the ROA one by one.

6. The automatic correction method for large-scale multi-batch LC-MS raw data according to claim 1, characterized in that: The cross-correlation operation of the ROA intensity vector of the reference file and the XIC intensity vector of the target file comprises the following steps: The similarity measurement is performed on the ROA intensity vector and the XIC intensity vector of two different lengths to obtain the region in the XIC that is most similar to the ROA, thereby obtaining the matching points of retention time and m / z in the reference file and the target file.

7. The automatic correction method for large-scale multi-batch LC-MS raw data according to claim 1, characterized in that: The nonlinear fitting of the matching points comprises the following steps: First, the linear trend of the retention time offset of the target file is obtained through robust linear fitting, and the false matching probability of each matching point is calculated according to the residual distribution of the linear fitting. The points with error probability lower than the threshold are screened, and then nonlinear LOESS fitting is performed to obtain the correction curve of retention time and m / z. The correction curve is used to calibrate the target file by applying the correction curve function to the retention time and m / z of the target file.

8. An automatic correction system for large-scale multi-batch LC-MS raw data, characterized in that: include: A file screening module is used to obtain the LC-MS raw data of the sample to be calibrated by mass spectrometry export, select a certain sample file as a reference file, and other sample files as target files; Anchor region construction module, used to construct anchor region ROA in reference file; An extracted ion chromatogram (XIC) building module is used to obtain an extracted ion chromatogram (XIC) corresponding to the m / z of ROA in a target file; A matching module is used to perform a cross-correlation operation on the ROA intensity vector of the reference file and the XIC intensity vector of the target file to obtain a matching point of retention time and m / z; A matching point fitting module is used to perform nonlinear fitting on the matching points to obtain a calibration curve of retention time and m / z; The correction module is used to map the retention time and m / z of the target file to be corrected through a correction curve function to obtain a corrected new file.

9. An automatic correction device for large-scale multi-batch LC-MS raw data, characterized in that: It comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement the automatic correction method for large-scale multi-batch LC-MS raw data as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the automatic correction method for large-scale multi-batch LC-MS raw data as described in any one of claims 1 to 7 is implemented.

Images