Sample quantitative consistency and classification workflow

Abstract

The present technology relates to a method and apparatus for classifying a sample. The method collects raw data from an analytical instrument (e.g., LC-MS); quantifying consistency parameters from raw chromatographic data (e.g., peak detection) by specifying criteria and determining probability ratios; and constructing a learning model quantification step by weighting the presence, absence, or modulation of one or more factors of the one or more samples against a positive criterion. The method may then apply the learning model to an unknown sample to detect the presence, absence, or modulation of one or more factors in the unknown sample, such as the presence, absence, or modulation of a disease state or states.

Description

[0001] Related Applications

[0002] This application claims priority to and the benefit of U.S. Provisional Application Serial No. 63 / 408,987, filed on September 22, 2022, and entitled “Sample Classification Workflow,” the contents of which are incorporated herein by reference in their entirety. Technical Field

[0003] The present disclosure relates to methods, techniques, and processes for sample classification. Background Art

[0004] Classification of biological, chemical, environmental and other samples is often based on the presence / absence of a signal using appropriate detection measurement techniques, or the relative abundance or readout of a single sample or (a subset of) samples from a complete experiment using more advanced analytical techniques such as liquid chromatography-mass spectrometry (LC-MS).

[0005] For most applications, such as clinical and forensic toxicology, environmental applications, and others, reporting the value or amount of a given compound based on the reading of a single feature or at most two features is sufficient as an outcome parameter for product purity / quality in clinical diagnostics, environmental pollution, or industrial manufacturing applications (e.g., pharmaceuticals, food, plastics, etc.). Examples would include the detection and analysis of methylmalonic acid in plasma / blood to monitor vitamin B12 status / deficiency [1] or pesticides in water matrices (ground, surface, etc.) in agricultural applications [2,3].

[0006] However, there is a need for a multiple reaction monitoring (MRM) based LC-MS system that does not require manual analysis of each sample. There is also a need for MRM analysis or targeted analysis derivatives such as parallel reaction monitoring (PRM) for disease detection using informatics. Summary of the invention

[0007] These unmet needs are addressed by the instruments and methods for sample classification disclosed herein. Analysis of multiple features from one or more analytes associated with a given sample state provides quantitative consistency and classification results with associated probabilities. The method can be easily extended to multiple sample state quantitative consistency and classification results. For example, the method disclosed herein can be used to classify SARS-CoV-2 infection states (positive / negative). In addition, the method disclosed herein can be easily extended to other viral infectious diseases, such as influenza A / influenza B and respiratory syncytial virus (RSV). In some examples, multiple sample states can be analyzed together. Joint analysis can include individual classifications and probabilities. For example, the method can classify or determine the probability of the disease state of each sample individually.

[0008] In general, the present technology relates to quantitative agreement and classification of samples based on various properties obtained from the interpretation of instrumental (e.g., MS or LC-MS) data for one or more sample states (e.g., disease, phenotype, etc.). As used herein, quantitative agreement can include actual amounts or concentrations or relative amounts or concentrations.

[0009] In an embodiment, the present technology relates to a mass spectrometer ("MS") / liquid chromatography-mass spectrometer ("LC-MS") instrument for classifying samples. The instrument uses a processing device for executing computer-readable instructions, which are used to execute a method for classifying samples. The method for classifying samples includes collecting raw chromatographic data from one or more analytes of one or more samples, and quantifying consistency parameters from the raw chromatographic data by specifying standards and determining probability ratios, thereby extracting features. In the feature extraction step, the consistency and / or relative abundance of the analyte is quantitatively measured ("quantitative consistency"). Then, the method weights the presence, absence or regulation of one or more factors of one or more samples by comparing the positive judgment criteria, and builds a learning model based on the quantitative consistency parameters (i.e., "model learning quantification"). The method then applies the learning model to an unknown sample to detect the presence, absence or regulation of one or more factors in the unknown sample.

[0010] In an alternative embodiment, the present technology relates to a method for classifying a sample. The method workflow analyzes specific features obtained from a sample to indicate the presence or absence of a sample state. In some embodiments, the method for classifying sample data uses a mass spectrometer ("MS") / liquid chromatography-mass spectrometer ("LC-MS") instrument. The method includes collecting raw chromatographic data from one or more analytes of one or more samples; quantifying consistency parameters from the raw chromatographic data by specifying criteria and determining probability ratios; constructing a learning model quantitative step based on the quantitative consistency parameters, including weighting the presence, absence or regulation of one or more factors of one or more samples against a positive judgment criterion; and applying the learning model to an unknown sample to detect the presence, absence or regulation of one or more factors in an unknown sample.

[0011] In some embodiments, the present technology can be used to clinically classify a sample using mass spectrometry, i.e., indicate the presence or absence of a disease state, by analyzing which analytes from the sample are present or determining the abundance or relative abundance of a signal of one or more analytes. For example, the present technology includes quantitative identity and analysis of peptides from a protein sample.

[0012] In some embodiments, the present technology can be used for model learning quantification by specifying criteria and determining specific features of analytes that can be used for future diagnosis and detection. Probability ratios can then be calculated by analyzing the specified criteria to generate model learning quantitative results.

[0013] In some embodiments, model learning quantitatively involves nested sampling and / or Markov Chain Monte Carlo ("MCMC") methods.

[0014] In some embodiments, the model learning quantitative results can then be used to weight the presence, absence, or modulation of one or more factors in one or more samples against positive decision criteria to construct a learning model based on the model learning quantitative parameters.

[0015] In some embodiments, the workflow analyzes two or more features (e.g., 2, 3, 4, 5, 6, 7 or more features) that, when detected together in a sample, indicate the presence of a disease / infection state or sample condition or more generally describe the sample or sample state.

[0016] In some embodiments, quantifying the consistency parameter includes quantifying peak detection results determined from the raw chromatographic data.

[0017] In some embodiments, the peak detection results are determined from a ranking scheme for the chromatographic peaks. The ranking scheme may include distance components reflecting the degree of mismatch of various aspects of the chromatographic peaks, selected from the group consisting of consistency of retention time positioning, peak width, and peak area. If the ranking scheme that places the chromatographic peaks in order from worst to best is reliable, the measurements that need adjustment or rejection should almost always appear above those that are immediately acceptable (i.e., semi-automatic inspection).

[0018] In an embodiment, the present technology relates to an instrument (e.g., MS or LC-MS) for classifying or quantifying samples based on abundance (amount or concentration) and for model learning quantification by specifying standards and determining specific features of analytes. The instrument includes a processing device for executing the method by executing computer-readable instructions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present technology will be more fully understood through the following detailed description made in conjunction with the accompanying drawings, in which:

[0020] Figure 1A and Figure 1B The training according to the present technique ( Figure 1A ) and Applications ( Figure 1B ) Schematic diagram of an implementation plan of the workflow.

[0021] Figure 2A-2FAn example of feature detection and extraction consistency is shown. Figure 2A , Figure 2B and Figure 2C Peak consistency for the analytes is shown. Figure 2D , Figure 2E and Figure 2F Additional agreement is shown for another test analyte (internal standard).

[0022] Figure 3A-Figure 3F An example of inconsistency in feature detection and extraction is shown. Figure 3A , Figure 3B and Figure 3C Peak inconsistency of the analytes is shown. Figure 3D , Figure 3E and Figure 3F Additional inconsistency is shown for another test analyte (internal standard).

[0023] Figure 4 A flowchart showing the model learning workflow.

[0024] Figure 5 A flowchart showing the model application workflow. DETAILED DESCRIPTION

[0025] In general, the present technology relates to the classification of samples detected by mass spectrometry, which are indicative of disease states or physiological conditions, such as viral infection (influenza, coronavirus, rhinovirus, etc.), phenotypes, etc., or sample status, such as overall contamination versus non-contamination. The present technology provides processes for peak detection, feature extraction, model building or application, and sample classification.

[0026] The present technology addresses challenges associated with manual inspection of a collection of samples describing sample states based on multiple analytes and features because decision metrics associated with acceptance criteria and / or optional quantitative reporting (i.e., quantitative agreement) can be extracted or obtained from multiple features from multiple analytes for one or more disease states or sample conditions.

[0027] The workflow steps of this technique can be found in Figure 1A and Figure 1B The flowchart is summarized in FIG.

[0028] Figure 1A A training workflow is shown starting with raw chromatographic data collected from an instrument such as an MS or LC-MS device. The instrument may contain processing means that perform the method steps of peak detection, feature extraction and training (model creation), resulting in the construction of a learning model.

[0029] Figure 1B shows the model application workflow, which is similar to Figure 1A, but this workflow applies an existing learned model to the results from the feature extraction step to classify the samples.

[0030] Figure 2A-2F Chromatographic peaks of the analytes used for feature detection and extraction consistency are shown.

[0031] Figure 2A , Figure 2B and Figure 2C Peak consistency of the analytes across the three chromatographic peaks is shown. Figure 2D , Figure 2E and Figure 2F Additional consistency of another test analyte (internal standard) across the three chromatographic peaks is shown. Figure 2A-2C and Figure 2D-2F As seen on each of the , the individual peaks align, indicating peak consistency.

[0032] Figure 3A , Figure 3B and Figure 3C Peak inconsistency of the analytes across three chromatographic peaks is shown. Figure 3D , Figure 3E and Figure 3F Additional inconsistency is shown for another test analyte (internal standard) across three chromatographic peaks. Figure 3A-3C and Figure 3D-Figure 3F As can be seen when analyzing the peaks, the peaks do not align, which indicates peak inconsistency.

[0033] Figure 4 and Figure 5 They are examples of model learning workflow and model application workflow, respectively.

[0034] exist Figure 4 In FIG. 4 , the model learning workflow has three stages: peak detection ( 401 ), feature extraction (quantitative consistency) ( 405 ), and learning the model ( 410 ). Figure 5 The model application in Figure 2 shows a similar workflow, but the final stage is to apply the learned model to classify the samples, rather than Figure 4 Build a learning model as in .

[0035] The first step of the peak detection phase (401) is to collect raw chromatographic data. The raw chromatographic data is collected from one or more analytes of one or more samples. The method is not limited in the number of analytes / samples collected. The collection can be in the form of batches of data collected simultaneously or sequentially. In an example, the raw chromatographic data includes retention time and relative abundance.

[0036] In an example, the method collects raw chromatographic data from an analytical instrument. Preferably, the analytical instrument is a mass spectrometer ("MS") or a liquid chromatography-mass spectrometer ("LC-MS").

[0037] In an example, the analytical instrument (e.g., MS or LC-MS) from which the raw data is collected also includes a processor that uses the sample classification method of the present technology. The instrument can collect a batch of samples / analytes simultaneously or sequentially, which is advantageous considering that the method is similar to MRM in that it can perform multiple analyses on multiple analytes.

[0038] like Figure 4 As shown in the peak detection stage (401) in FIG. 4 , once the raw chromatographic data is collected, peak detection is performed using peak detection parameters.

[0039] One example of a peak detection parameter for the peak detection stage (401) is through the use of a ranking scheme. In particular, the present technique can reduce the burden of manual chromatographic review by the user by employing a ranking scheme for chromatographic peaks from the raw chromatographic data. If the ranking scheme that places chromatographic peaks in order from worst to best is reliable, measurements that require adjustment or rejection should almost always appear above those that are immediately acceptable (i.e., semi-automatic review).

[0040] As part of the ranking scheme peak detection parameters of the peak detection stage (401), distance components can be provided to aid in the interpretability of the ranking scheme. These components will reflect the degree of mismatch in various aspects of the chromatographic peak measurements, such as consistency in retention time positioning, peak width and peak area (including information about any ion ratios), and peak asymmetry. Given the ranking and separate distance components, the user or system will quickly be able to assess points that do not require further examination.

[0041] Sorting Scheme Peak Detection Parameters The sorting scheme may be optimized using one or more sorting scheme parameters. Examples of one or more sorting scheme parameters include batch center, sample center (for each sample), compound center (for each compound, relative to the sample center), change in transition measurement from compound center, overall variance scale of measurements, precursor abundance (one precursor abundance per compound), transition efficiency (one transition efficiency per transition), and overall variance scale of measurements.

[0042] Using the estimates of the ordering scheme parameters from the peak detection stage (401), the distance of each data point to the ideal value is constructed. The estimates can be calculated at each iteration of the MCMC algorithm, and the squares of the distances averaged over the MCMC runs yield the final distance.

[0043] Once the ranking scheme data is collected for the peak detection stage (401), the data is analyzed to determine specific peaks for the analyte (ie, "peak detections"). The results from the peak detections can then be used in the feature extraction stage (405).

[0044] In the feature extraction stage (405), the identity and / or relative abundance of the analytes is quantified ("quantitative identity"). In some examples, the abundance may be expressed as a concentration derived from peak integration. In some examples, the identity value is a likelihood ratio.

[0045] Feature extraction (quantification of identity) of the feature extraction stage (405) reduces the raw chromatographic data via peak detection to produce likelihood ratios for each analyte in each sample. In some embodiments, the number / amount of identities is reduced using, for example, a Markov Chain Monte Carlo (MCMC) method.

[0046] The consistent results from the feature extraction stage (405) are then used to quantify the learned model ("learned model quantification") (410). As part of the learned model quantification, the method performs learning and cross-validation, using a training set with criteria assigned to each sample, to optimize the parameters of one or more models of the prior probability of the analyte given the presence, absence or adjustment of the factor of the control positive decision criteria. This creates an optimized model.

[0047] For example, learning model quantification uses Markov chain Monte Carlo (MCMC) methods to build models for LC-MS MRM data. The good-bad data model approach is used to assign peaks to ON sets controlled by sub-model types A and B, or OFF data controlled by sub-model type C: Type A is used to measure ON peaks, such as retention time or peak width, which should not have systematic changes between MRM transitions of compounds in a particular sample; Type B is used to measure ON peaks associated with abundance, such as peak height or peak area, which have systematic changes between MRM transitions of compounds in a particular sample (due to different efficiencies of precursor cleavage to products); Type C models are used for all OFF peak attributes. This can be a simple unified model. For peaks in the ON group, distances are defined based on the central estimates of the parameters of the type A and type B models.

[0048] As part of the learning model phase (410), the peak detection results from the peak detection phase (401) and the consistency results from the feature extraction phase (405) can be used to perform model learning quantification by specifying criteria and determining specific features of the analyte that can be used for future diagnosis and detection. Probability ratios can then be calculated by analyzing the specified criteria to produce model learning quantitative results.

[0049] In an example, learning model quantitatively involves nested sampling and / or Markov chain Monte Carlo ("MCMC") methods. The method is not limited in the model learning quantitative method and may include additional machine learning methods. Model learning quantitative may also involve cross-validation, such as leave-one-out cross-validation.

[0050] In the example, the learning model of the learning model stage (410) quantitatively builds a model that will provide the distance of each measurement from the central estimate and the probability of being "good" (i.e., a measurement that belongs to the consensus of good measurements). This consensus may come from the analysis of a single batch, or it may be influenced by historical / training data collected previously.

[0051] The model learning quantitative results may then be used to weight the presence, absence or regulation of one or more factors in one or more samples against positive determination criteria, and construct a learning model based on the model learning quantitative parameters (410).

[0052] Once a learning model is constructed, the model can be applied to unknown analytes to detect the presence, absence, or modulation of one or more factors in an unknown sample. Figure 5 Model application workflow shown.

[0053] Figure 5 The workflow has Figure 4 The peak detection stage (401) and feature extraction stage (405) in FIG. 5 are similar to the peak detection stage (501) and feature extraction stage (505) in FIG. 5 . However, Figure 5 Model application workflow and Figure 4 The difference is that Figure 5 The model application phase (510) replaces Figure 4 The learning model phase (410) of the present invention is described in detail. Specifically, the model application phase applies the optimized model to unknown sample data. The optimized model is applied to the extracted features of the unknown sample to detect the presence, absence, or modulation of factors describing the sample state. In an embodiment, the application of the optimized model detects the presence or absence of a disease state such as a virus, or requests manual inspection of the data.

[0054] In an example, one or more factors describe the presence or absence of a disease state or disease states. Disease states include, for example, viral infections and cancer.

[0055] Viral infections are not limited and include, for example, SARS-CoV-2, HIV, influenza, and other viral strains that affect humans and other animals.

[0056] Cancer is not limited and includes, for example, tumor cells from non-malignant cancers and malignant cancers in humans and other animals.

[0057] The analyte of the sample is not particularly limited in chemical structure. In an example, the analyte may be one or more endogenous peptides. Preferably, the analyte comprises peptides from a cell sample containing a disease state or a cell sample without a disease state. The sample may comprise an internal standard from which characteristic information is inferred. In a preferred example, the sample comprises an isotope-labeled peptide.

[0058] Once a model has been developed for a specific analyte, the method can be applied to other similar analytes. Purely as a quality check, experts can compare the MRM chromatograms with their own manual interpretations to validate the method results.

[0059] Example 1: COVID-19 Analysis

[0060] Samples from COVID-19 patients were collected and peptide analytes from the samples were analyzed using a classification method. Based on this method, a model threshold was formulated. Manual interpretation of each peptide in the MRM data was compared with the results of applying this method to confirm detection consistency.

[0061] Quantitative consistency parameters include retention time (t R ), curves (peak shape), relative abundance, etc. During the feature extraction process, peak detection is performed on the training data including calibration samples and QC, and changes in retention time and relative abundance are analyzed to generate likelihood values ​​for endogenous peptides. Calibration and QC samples and internal standards can be given a higher prior probability to truly reflect the expected retention time and relative abundance to assist in this process. The parameters of the model are optimized using peptide likelihoods and the known conditions of the samples, and the prediction accuracy of the model is calculated by repeating several rounds of optimization and leave-one-out cross-validation.

[0062] The learning model quantitatively learns the optimal weights associated with the peptide combinations.

[0063] Nested sampling is used by randomly sampling within likelihood constraints to produce known statistical results. Likelihood values ​​naturally define contours of equal likelihood. The prior volume is the proportion of prior results contained within the contours of equal likelihood. Nested sampling steadily compresses the prior volume. As the prior volume shrinks geometrically and the likelihood increases, the posterior distribution is traversed. Eventually, the reduction in volume destroys any increase in likelihood and the algorithm terminates. Statistics for a state can be accumulated along the way by providing specific weights.

[0064] Nested sampling stably compresses the prior distribution to produce specific evidence criteria and information, such as the logarithm of the compression ratio of the posterior to the prior and the normalized posterior probability distribution. In examples, nested sampling can be dynamic, which allows the algorithm to adapt to the shape of the posterior distribution in real time, thereby improving both accuracy and efficiency.

[0065] The method assigns a switch to each transition in each sample. If the switch is on, the transition is a member of the group, i.e., in the group of "good" transitions for the peptide in question; if the switch is off, the transition is considered to be generated independently by the background (similar to outlier detection). The "on" state of the switch can also encode which peak is associated with the transition from multiple candidate peaks.

[0066] Once the probability ratio is found, the state of the switch can be resampled to any available state subject to the current likelihood constraints.

[0067] Quantitative agreement of peptide abundance and transition efficiency includes the abundance of endogenous peptides in the sample, the abundance of labeled peptides in the sample, the efficiency of peptide transitions, and the Gaussian likelihood of the model area relative to the observed peak area of ​​the “on” transition.

[0068] Model learning is quantitatively resampled using a nested sampling likelihood constraint on the quadratic log-likelihood, which gives a bound on the amount of acceptable prior “opening” model learning.

[0069] Peptide probabilities are accumulated based on the following: At each sampling point of a nested sampling run, if any transition belonging to the peptide is "on", a value of 1 is multiplied by the nested sampling weight and added to the probability value for the peptide. If all transitions for the peptide are "off", then the probability of the peptide being "on" at that sampling point is the prior probability that all transitions are "off" given the presence of the peptide. This is multiplied by the current nested sampling weight and added to the accumulated value.

[0070] The constructed model then relates the peptide likelihood to the probability of the presence / absence of the virus (i.e., SARS-CoV-2). Each alternative (virus presence / absence) has a prior probability for each configuration of the peptide presence / absence. To optimize the model, a cost function can be pre-established. An arbitrary positive threshold can also be pre-set.

[0071] The model uses decision optimization to minimize expected costs. Costs are evaluated on outcomes as the proportion of negatives and positives, population prevalence incorporating costs, specificity, etc. The optimization calibrates the odds of a positive sample against the decision threshold and can also set the test threshold below the positive threshold.

[0072] Table 1 below shows an excerpt of the results after applying the model to SARS-CoV-2, with parameters learned for QC and COVID-19 patient data. The threshold for indicating a positive sample state (infected with SARS-CoV-2) is an odds ratio of 1 (**). When the odds ratio is greater than 0.070577 (***), which is the value learned during training, an expert check is required, i.e., an uncertain result is obtained. An odds ratio less than this check threshold (*) indicates a negative infected sample state result (infected with SARS-CoV-2). In this embodiment, 3 features are used for each analyte, and a total of two analytes are used.

[0073] Table 1

[0074] sample Covid-19 chance Neg_QC 0.0340095* Pos_QC 1.0043** Solvent Blank 0.018458* Patient 1 0.0227256* Patient 2 0.019211* Patient 3 0.0336331* Patient 4 0.0386951* Patient 5 1.0043** Patient 6 0.0352765* Patient 7 0.077185*** Patient 8 0.0218494* Patient 9 0.0250914*

[0075] *Odds ratio is less than the threshold

[0076] **Odds ratio is 1

[0077] ***Odds ratio greater than threshold

[0078] Threshold = 0.070577

[0079] Table 2 shows the results of repeated leave-one-out learning and cross-validation cycles. In this process, one sample is omitted from the dataset during training and its positive / checked / negative status is obtained using the learned model parameters accumulated into the statistics.

[0080] The table shows the statistics resulting from omitting each sample from the training set in turn. The training is relative to the assignments made by an expert examining the LC-MS data (features) of each sample.

[0081] Table 2

[0082]

[0083] Table 3 shows the results of the reference assignment (independent interpretation of the LC-MS data) relative to the expert assignment used to generate Table 2.

[0084] Table 3

[0085]

[0086] The results showed that this method consistently and accurately analyzed and produced assay results compared to manual analysis by experts who examined the LC-MS data (features) of each sample.

[0087] The method of the above embodiment is not limited to the same number of features and analytes. Because the model involves learning, the method can adapt to new analytes and become more accurate as the number of samples increases.

[0088] Example 2: Circulating endogenous analytes (disease classification)

[0089] In another embodiment, the present technology quantifies the consistency and / or classification of circulating amounts of multiple endogenous analytes in body fluids. For example, the analytes can be biochemical compounds such as vitamin D metabolites and steroid hormones. Conditions associated with vitamin D deficiency include respiratory infections, osteoporosis and other chronic and metabolic diseases such as obesity, metabolic syndrome, type 2 diabetes, cancer, rheumatoid arthritis and inflammatory bowel disease. Vitamin D also plays an important role in the clinical course of infectious and other acute diseases, particularly respiratory bacterial infections, tuberculosis and viral infections, such as infections caused by human immunodeficiency virus and SARS-CoV-2 virus.

[0090] In this example, both 25-hydroxyvitamin-D3 and 25-hydroxyvitamin-D2 as circulating endogenous analytes are analyzed by using peak detection, quantitative consistency, and building / applying a learning model based on the methods disclosed herein. The results of this analysis can be used to determine the values ​​(e.g., concentrations or amounts) of various analytes in a sample. This information can be used to classify the above conditions.

[0091] Example 3: Therapeutic Drug Monitoring (Absolute Determination)

[0092] In another embodiment, as a method for therapeutic drug monitoring, the present technology quantifies and / or classifies the absolute amount of one or more therapeutic drugs. For example, everolimus is a commonly used immunosuppressant with multiple mechanisms of activity, such as high inter- and intra-individual variability. Therefore, accurate, analytically sensitive quantitative methods using the present technology can play a role in studying the pharmacokinetic and pharmacodynamic effects of drug administration.

[0093] Here, the present technology applies the learning model of the present technology, which is optimized by determining the value of therapeutic drug monitoring (including the bias value that may affect the hematocrit measurement). The method of the present technology uses LC-MS / MS instrumentation to analyze the dried blood spot analysis of everolimus. The everolimus sample is analyzed by using peak detection, quantitative consistency, and applying the learning model based on the method disclosed herein to determine the value (e.g., concentration or amount) of everolimus in the sample. This information can be used for therapeutic drug monitoring. Similarly, the calculated bias value of the medical judgment level shows that the hematocrit has no clinical impact on the results.

[0094] Example 4: Multi-disease Case

[0095] In another embodiment, the present technology uses the methods disclosed herein to quantify consistency and / or classify the profiles of multiple diseases from one or more analytes. One or more factors are detected in unknown samples using peak detection, quantitative consistency, and application of learning models based on the methods disclosed herein. These one or more factors can be used to determine a value (e.g., concentration or amount) that can be used to determine one or more disease states. In this embodiment, unknown samples are analyzed based on this method to determine influenza (influenza A / influenza B) and SARS-CoV-2 disease states. Therefore, this embodiment is able to analyze multiple disease states with a single unknown sample.

[0096] Although the present technology has been described with reference to preferred embodiments, it will be understood by those skilled in the art that various changes in form and details may be made without departing from the scope of the present invention as set forth in the following claims.

[0097] References

[0098] [1]EMMineva,M.Zhang,DJRabinowitz,KWPhinney,CMPfeiffer,An LC-MS / MS method for serummethylmalonic acid suitable for monitoring vitamin B12status in population surveys,Anal.Bioanal.Chem.407(2015)2955–2964.https: / / doi.org / 10.1007 / s00216-014-8148-2.

[0099] [2]C.Campanale,C.Massarelli,D.Losacco,D.Bisaccia,M.Triozzi,VFUricchio,The monitoring of pesticides in water matrices and theanalytical criticalities:A review,TrAC Trends Anal.Chem.144(2021)116423.https: / / doi.org / https: / / doi.org / 10.1016 / j.trac.2021.116423.

[0100] [3] J.Lundqvist, C.von AKRosenmai, Ohlsson,T.Le Godec,O.Jonsson,J.Kreuger,A.Oskarsson,Assessment of pesticides in surface watersamples from Swedish agricultural areas by integrated bioanalysis andchemical analysis,Environ.Sci.Eur.31(2019)53.https: / / doi.org / 10.1186 / s12302-019-0241-x.

Claims

1. A mass spectrometer ("MS") / liquid chromatography-mass spectrometer ("LC-MS") instrument for quantifying and classifying a sample, the instrument comprising: A processing device for executing computer readable instructions for executing a method for classifying a sample, the method comprising: collecting raw chromatographic data from one or more analytes from one or more samples; quantifying consistency parameters from the raw chromatographic data by specifying criteria and determining probability ratios; constructing a learning model quantitative step based on the quantitative consistency parameters, including weighting the presence, absence or regulation of one or more factors of the one or more samples against a positive decision criterion; and The learned model is applied to an unknown sample to detect the presence, absence or modulation of the one or more factors in the unknown sample.

2. The apparatus of claim 1, wherein the one or more factors describe the presence, absence or modulation of a disease state or multiple disease states.

3. The apparatus of claim 1, wherein the learning model quantification step comprises nested sampling and / or Markov Chain Monte Carlo ("MCMC") methods.

4. The instrument of claim 1, wherein the raw chromatographic data is collected from multiple analytes run simultaneously or sequentially.

5. The apparatus according to claim 1, wherein the learning model quantification step further comprises leave-one-out cross-validation.

6. The apparatus of claim 1, wherein the step of quantifying a consistency parameter comprises quantifying peak detection results determined from the raw chromatographic data.

7. The instrument of claim 6, wherein the peak detection results are determined from an ordering scheme of chromatographic peaks.

8. The apparatus of claim 7, wherein the ranking scheme includes a distance component reflecting the degree of mismatch of various aspects of the chromatographic peaks, the distance component being selected from the group consisting of consistency of retention time positioning, peak width, and peak area.

9. The instrument of claim 1, wherein the raw chromatographic data comprises retention time and relative abundance.

10. The apparatus of claim 1, wherein the one or more samples include endogenous analytes or isotopically labeled analytes.

11. The apparatus of claim 1, wherein the method further comprises determining one or more disease states based on the presence or absence of the one or more factors in the unknown sample.

12. A method for classifying sample data from a mass spectrometer ("MS") / liquid chromatography-mass spectrometer ("LC-MS") instrument, the method comprising: collecting raw chromatographic data from one or more analytes from one or more samples; quantifying consistency parameters from the raw chromatographic data by specifying criteria and determining probability ratios; A learning model quantitative step is constructed based on the quantitative consistency parameters, including weighting the presence, absence or regulation of one or more factors of the one or more samples against a positive decision criterion; as well as The learned model is applied to an unknown sample to detect the presence, absence or modulation of the one or more factors in the unknown sample.

13. The method of claim 12, wherein the one or more factors describe the presence, absence or modulation of a disease state or multiple disease states.

14. The method of claim 12, wherein the learning model quantification step comprises nested sampling and / or Markov Chain Monte Carlo ("MCMC") methods.

15. The method of claim 12, wherein the raw chromatographic data is collected from multiple analytes run simultaneously or sequentially.

16. The method according to claim 12, wherein the learning model quantification step further comprises cross validation. The method of claim 16 , wherein the cross-validation comprises leave-one-out cross-validation.

18. The method of claim 12, wherein the step of quantifying a consistency parameter comprises quantifying peak detection results determined from the raw chromatographic data.

19. The method of claim 18, wherein the peak detection result is determined from an ordering scheme of chromatographic peaks.

20. The method of claim 19, wherein the ranking scheme includes a distance component reflecting the degree of mismatch of various aspects of the chromatographic peaks, the distance component being selected from the group consisting of consistency of retention time positioning, peak width, and peak area.

21. The method of claim 12, wherein the raw chromatographic data comprises retention time and relative abundance.

22. The method of claim 12, wherein the one or more samples include endogenous analytes and isotopically labeled analytes.

23. The method of claim 12, wherein the method further comprises determining one or more disease states based on the presence, absence, or modulation of the one or more factors in the unknown sample.

Images