Machine learning system for analyte identification

Abstract

An example method performed via a computing device for providing support for a mass spectrometry (MS) system includes obtaining from a mass spectrometry library (i) a ranked hit list of reference spectra corresponding to a set of fragmented spectra of an analyte acquired with the MS system; and (ii) a first set of metadata corresponding to the list of sorted hit items for the reference spectrum. The method also includes obtaining, from the MS system, a second set of metadata corresponding to the set of fragmented spectra. The method further includes evaluating an order of entries in the ranked hit item list with a machine learning model based on the first set of metadata and the second set of metadata.

Description

TECHNICAL FIELD

[0001] Various examples generally relate to, without limitation, support systems for scientific instruments, such as mass spectrometry systems. BACKGROUND

[0002] Determining the identity of a compound is one of the main tasks of chemical analysis. For compounds in complex mixtures, this can be achieved using the fragmentation pattern of their ions. The corresponding mass spectrum can provide a direct readout of the elemental composition and cleavable bonds of a compound. When combined with gas chromatography or liquid chromatography (GC or LC), mass spectrometry (MS) can distinguish between hundreds of components in a complex mixture.

[0003] Steady progress in sensitivity and resolution of mass spectrometers continues to impart new capabilities to detect an increasing number of components in chemical mixtures. The processing of an ever-increasing number of identifiable compounds and associated digitized data poses significant challenges for the efficient use of such advanced instruments. Example tools for analyzing high-resolution mass spectrometry (HR-MS) data include the use of spectral libraries: collections of chemical structures and their mass spectra, which can support the rapid and reliable identification of compounds whose fragmentation patterns are measured with MS instruments. SUMMARY

[0004] Various examples, aspects, features, and implementations of machine learning (ML) systems for analyte identification are disclosed herein, among other things. In one example, an ML classification algorithm for small molecule identification proposes a modified scoring and ranking system that takes into account certain metadata of a query spectrum and its corresponding mass spectral library matches. Such metadata can include, but are not limited to, normalized collision energies, ion activation method parameters, and / or relevant features pre-selected for candidate compounds. In at least some examples, the modified scoring and ranking system advantageously causes the corresponding ML system to generate a modified hit list with a higher proportion of correct candidates at its top, due to the joint use of similarity scores and data-driven model stacking.

[0005] One example provides a method for providing support for a mass spectrometry (MS) system, performed via a computing device, the method comprising: obtaining, from a mass spectral library, (i) a ranked hit list of reference spectra corresponding to a set of fragmentation spectra of an analyte acquired with the MS system; and (ii) a first set of metadata corresponding to the ranked hit list of reference spectra; obtaining, from the MS system, a second set of metadata corresponding to the set of fragmentation spectra; and evaluating the order of items in the ranked hit list with a machine learning (ML) model based on the first set of metadata and the second set of metadata.

[0006] Another example provides an apparatus for providing support to a MS system, the apparatus comprising: an interface device; a processing device; and a storage device comprising program code, wherein the storage device and program code are configured to enable the apparatus, with the interface device and the processing device, to at least: obtain, from a mass spectral library, (i) a ranked hit list of reference spectra corresponding to a set of fragmentation spectra of an analyte collected with the MS system; and (ii) a first set of metadata corresponding to the ranked hit list of reference spectra; obtain, from the MS system, a second set of metadata corresponding to the set of fragmentation spectra; and evaluate an order of items in the ranked hit list with a ML model based on the first set of metadata and the second set of metadata. BRIEF DESCRIPTION OF DRAWINGS

[0007] The foregoing aspects and many of the attendant advantages of this disclosure will become more readily appreciated as the same become better understood by reference to the following detailed description, when taken in conjunction with the accompanying drawings, wherein:

[0008] Figure 1 is a block diagram illustrating a machine learning (ML) system for analyte identification according to some examples.

[0009] Figure 2 is a block diagram illustrating a model module in a ML system according to one example. Figure 1

[0010] Figure 3 is a block diagram illustrating an encoder in a model module according to one example. Figure 2

[0011] Figure 4 is a block diagram illustrating a workflow for preparing input to an encoder according to one example. Figure 3

[0012] Figure 5 is a block diagram illustrating a training process applied to a ML model in a model module according to one example. Figure 2

[0013] Figure 6 is a flowchart illustrating a method of analyte identification implemented using a ML system according to some examples. Figure 1

[0014] certain operations of the method according to one example are illustrated graphically. Figures 7A-7C Figure 6

[0015] Figure 8 is a block diagram illustrating a computing device according to some examples, one or more instances of which can be used in a ML system. Figure 1 ​​​​​​​

[0016] Figure 9 is a block diagram showing a MS instrument support system in which some or all of the mass spectrometer support methods and / or functions disclosed herein can be performed according to some examples.

[0017] Figures 10A-10B shows a method of utilizing a mass spectral library according to one example. Figure 6 The improvements achievable by the method relative to conventional analyte identification methods. DETAILED DESCRIPTION

[0018] Mass spectral libraries are an important resource for analytical chemists in a variety of applications. For example, the National Institute of Standards and Technology (NIST) provides several curated mass spectral reference databases. In addition, NIST produces and distributes search software that interacts with these libraries. Other mass spectral libraries and corresponding search software that interact with these libraries are also available from various other providers, including but not limited to mzCloud, Mass Frontier, and myLibrary.

[0019] A typical mass spectral library search algorithm computes a match score (sometimes also called a match factor) between a query spectrum and a set of reference spectra. In some examples, the match score is expressed as an integer between 0 and 999, used to quantitatively describe the “similarity” between a pair of spectra. Different search algorithms often differ in the way the match score is computed. In different search configurations, the set of reference spectra can include the entire library of spectra (i.e., no pre-search is performed) or a selected subset of library spectra determined during preselection or pre-search. The search algorithm typically returns a list of reference spectra sorted in descending order of match score. The returned sorted list is often referred to as the “hit list” for the query.

[0020] Three commonly used algorithms implemented in the NIST MS Search software are normal identity search, simple similarity search, and hybrid similarity search. These algorithms each use some or all of the following basic operations: pre-search, peak matching, dot product computation, match score computation, hit list sorting and display. In use, pre-search selects a subset of library spectra that are likely to yield high match scores. The goal of the normal identity search algorithm is to return a hit list that contains a correct identification of the query spectrum, preferably at the top of the hit list. The goal of both the “simple” and “hybrid” similarity searches is to return a hit list that aids the analyst in speculating a structure for the compound (analyte) of the query.

[0021] Whenever expert analysts review hit lists, they can often identify certain weaknesses in the putative identifications of analytes. One example weakness can occur when the reference spectra do not sufficiently match the corresponding experimental conditions of the query items. Another example weakness can occur when there are too few peaks for a sufficiently credible identification. Some embodiments disclosed herein aim to improve the credibility of the identification workflow by replacing the human-centric manual “expert analysis” hit list processing approach with a machine learning (ML) model. In some examples, the ML system for analyte identification of the present disclosure can advantageously enable higher specificity (e.g., by reducing the number of false positives) than algorithms that rely on spectral similarity metrics derived primarily from the aforementioned dot product metrics, etc.

[0022] According to one example, the ML classification algorithm for small molecule identification proposes a modified scoring and ranking system that takes into account the annotations and metadata of the query spectra matches to library spectra. In addition to the similarity metric between the unknown spectrum and the candidate spectra, the ML classification algorithm with a suitably chosen classifier (e.g., random forest, logistic regression, Bayesian network, XGBOOST, LightGBM, etc.) also takes into account certain metadata such as the normalized collision energy, ion activation method parameters, and relevant features pre-selected for the candidate compounds (e.g., the number of candidate compounds, average spectral match score, etc.). In some examples, the ancillary information related to the fragment spectra input into the ML classification algorithm can include one or more of the following: precursor ion m / z, precursor ion formula, charge state, peak count, peak sparsity, peak accuracy, peak intensity, peak distance, neutral loss, peak fragmentation curve, peak formula, peak structure, chemical class, compound class, and one or more mean, median, average, standard deviation, standard error, relative standard deviation, relative error, or variance values. In at least some examples, the modified scoring and ranking system advantageously causes the corresponding ML system to generate hit lists with a higher proportion of correct candidates at the top.

[0023] Figure 1 is a block diagram illustrating an ML system 100 for analyte identification according to some examples. The system 100 can be implemented by circuitry (e.g., including electrical components and / or optical components), such as one or more programmed computing devices. The following description refers to Figure 8 Examples of computing devices that can implement the ML system 100, alone or in combination, are described in more detail below. In addition, the following description refers to Figure 9 Examples of systems of interconnected computing devices in connection with a corresponding MS instrument are described in more detail below. In the illustrated example, the ML system 100 includes an input module 110, a service module 120, a model module 130, and an output module 140.

[0024] The input module 110 can be provided as part of a user interface through which a user can submit a search query for analyte identification. In one example, a search query submitted via the input module 110 includes one or more fragmentation mass spectra acquired with a corresponding MS instrument. The submitted mass spectra typically correspond to the same unknown compound. Each mass spectrum is part of a spectral dataset that typically includes, in addition to the mass spectrum itself, a set of associated metadata. Herein, the term “mass spectrum” refers to a list of (m / z, I) data points, where m, z, and I are mass, charge, and intensity, respectively. The metadata can include one or more of a set of spectral tags, a set of sample characteristics, and / or a set of acquisition parameters used by the MS instrument to measure the spectrum. Various embodiments can accept one or more of: electron ionization (El) spectra, small molecule tandem spectra, and peptide tandem spectra. El searches are typically performed at unit mass resolution, while tandem searches typically accept high resolution spectra using either relative (ppm) or absolute (m / z) tolerances.

[0025] The service module 120 provides connectivity between the input module 110 and the model module 130. In some examples, the trained ML model provided by the model module 130 resides on the same server as the relevant mass spectral library. In some other examples, the trained ML model provided by the model module 130 resides on a different server connected by a network or on a local (to the user) computing device. In the latter examples, the service module 120 also provides operational interconnectivity between the model module 130 and the mass spectral library (not explicitly shown, see, e.g., Figure 4 ) In some examples, the service module 120 is configured to provide flexibility in data, library, and ML model access modes, e.g., by means of FaaS functionality, various programming languages, and “big data” solutions, etc. Here, the acronym “FaaS” stands for Function as a Service, which is a cloud computing service that allows customers to develop, run, and manage application functionalities without the infrastructure build and maintenance complexities typically associated with developing and launching applications.

[0026] In some examples, the trained ML model provided by the model module 130 is a random forest classifier. In some other examples, suitable alternatives to random forest classifiers can be similarly used. Example alternatives include gradient boosting models, k-nearest neighbor (kNN) algorithms, variational autoencoders, etc.

[0027] In some examples, the output module 140 is provided as part of the same user interface as the input module 110. In such examples, the service module 120 additionally provides operational connectivity between the model module 130 and the output module 140, e.g., as shown in Figure 1The example output generated by the trained ML model of model module 130 and provided to the user via output module 140 includes one or more of the following: (i) an estimated probability that the compound characterized by the queried spectrum belongs to a given class of compounds; (ii) an estimated probability that the compound characterized by the queried spectrum belongs to a given chemical class; (iii) an estimated probability that the queried spectrum is similar to some or all of the library spectra from the hit list; and (iv) a comparison plot of the queried spectrum to the library spectra from the hit list.

[0028] Figure 2 is a block diagram illustrating model module 130 according to one example. In the illustrated example, model module 130 includes encoder 210 and trained ML model 220. Encoder 210 operates to convert inputs 202, 204, 206, and 208 to corresponding feature vectors 212. Feature vectors 212 are then processed with trained ML model 220 to generate output 222 for output module 140.

[0029] In some examples, inputs 202, 204, 206, and 208 are as follows. Input 202 has a set of mass spectra of an analyte acquired with a corresponding MS instrument. In some examples, such a set includes a single spectrum. In some other examples, such a set includes multiple spectra. Input 204 has a set of metadata corresponding to the spectra of input 202. Input 206 has a hit list of spectra obtained from a library of mass spectra by a conventional search for closest matches to the spectra of input 202. Input 208 has a set of metadata corresponding to the hit list spectra of input 206, retrieved from the library of mass spectra along with the spectra.

[0030] In one example, to accommodate the widely varying inputs 202, 204, 206, 208 for different analyte samples, encoder 210 first performs tokenization configured to map inputs 202, 204, 206, 208 to corresponding token vectors of fixed length. The token vectors are then further encoded by encoder 210 to obtain corresponding feature vectors 212. Feature vectors 212 can be qualitatively understood as containing, in a form that can be exploited, information of various features conveyed by inputs 202, 204, 206, 208, useful to guide generation of output 222 in trained ML model 220.

[0031] Figure 3 is a block diagram illustrating encoder 210 according to another example. In the illustrated example, encoder 210 includes as Figure 3The feature extraction block 310, the feature engineering block 320 and the feature encoding block 330 are connected in series as shown. The inputs 202, 204, 206, 208 are applied to the feature extraction block 310. The output 212 is generated by the feature encoding block 330.

[0032] In the feature extraction block 310, the inputs 202, 204, 206 and 208 are used to identify and extract relevant features from the raw data. The corresponding operations of the block 310 include applying data preparation techniques to the raw data set, such as scaling, missing value imputation and separating mixed variables into individual features for further processing. The first intermediate output 312 generated with the feature extraction block 310 will be used in the feature engineering block 320 to create new features or to transform existing features. The corresponding operations of the block 320 can include: (i) for pairs of features, computing the first order discrete difference of the elements; (ii) for individual features, computing the description of the feature set with respect to the group, including but not limited to rank, size, mean, standard deviation and mean to raw numerical features absolute deviation, and features based on relative abundance and matching peaks. The second intermediate output 322 generated by the feature engineering block 320 will be further transformed in the feature encoding block 330. For example, certain ML algorithms are specialized in handling numerical values, and it is therefore recommended to convert the categorical values of relevant features into numerical features, for example in the form of vector values. Thus, in the block 330, certain categorical features (e.g. type of analyser, ion activation method and ionization method) can be converted from their respective categorical values into relevant numerical features that can be handled by the ML model employed in the block 130. In some examples, the feature vector 212 is also subjected to additional processing to ensure that it contains only relevant features (e.g. target labels) and that there are no outliers and missing values in the data set used to train and validate the ML model 220.

[0033] Figure 4 is a block diagram showing a workflow 400 for preparing the inputs 202, 204, 206, 208 according to one example. The workflow 400 comprises operating a corresponding MS instrument to acquire one or more MS spectra 412 of the analyte of interest. A corresponding set of acquisition parameters for each acquired MS spectrum 412 will be output from the MS instrument to form at least part of the input 204 to the encoder 210.

[0034] Workflow 400 also includes submitting a corresponding search query to a mass spectral library 420 using the collected MS spectrum 412 to perform a corresponding search 416 therein. Mass spectral library 420 has a set of searchable reference spectra that are annotated with compound identifiers and further associated with corresponding experimental metadata and compound-related metadata. Search 416 returns a hit list 430 that includes a subset of reference spectra from mass spectral library 420 and further includes a corresponding similarity measure between the experimentally collected MS spectrum 412. In some examples, the similarity measure used with hit list 430 includes the match score mentioned above. The set of experimental metadata and compound-related metadata corresponding to the reference spectra of hit list 430 are also retrieved from mass spectral library 420. In some examples, mass spectral library 420 is a NIST library or a mzCloud library.

[0035] Workflow 400 optionally includes cropping or filtering the set of metadata retrieved from mass spectral library 420 to remove portions of metadata that do not correspond to features used by model module 130. Depending on the system implementation and the specifics of mass spectral library 420 used, the original set of metadata retrieved from mass spectral library 420 or a cropped / fi ltered subset thereof is used to form input 208 of encoder 210.

[0036] Workflow 400 also includes applying one or more pre-processing operations 432 to hit list 430 and related MS spectrum 412. The output generated with pre-processing operations 432 is used to form inputs 202 and 206 of encoder 210. Examples of pre-processing operations 432 include: removing compound labels, removing at least some annotations, removing some library compounds from the hit list based on structural similarity or dissimilarity, removing specific peaks from the hit list, etc.

[0037] Figure 5 is a block diagram illustrating a training process 500 for training ML model 220( Figure 2 ) according to one example. In general, ML model training processes depend on the type of model. In the illustrated example, training process 500 corresponds to an implementation in which ML model 220 is a random forest classifier. Based on the provided description, a person of ordinary skill in the relevant art will be able to implement other training processes suitable for other model types without any undue experimentation.

[0038] A random forest classifier includes multiple decision trees, each of which outputs a respective prediction. When performing a classification task, each decision tree in the random forest votes for one of the classes to which the input might belong. After all the decision trees have voted, the random forest classifier counts which class has the most votes, and this class is the final prediction output by the random forest classifier. An individual decision tree splits data into groups based on the features represented by the data. The decision tree continues to split the data into groups until a small set of data under a label (class representation) is retained. The decision tree determines where to split the represented features based on a purity indication that measures information gain. For a classification task, the decision tree makes decisions based on Gini index or entropy, and for a regression task, based on the sum of squared residuals.

[0039] The random forest logic can be briefly described as follows:

[0040] i. Assume there are N observations. Take a random sample with replacement of these N observations.

[0041] ii. Assume there are M features in the observations. At each node, randomly select m features from the total number of features M, where m < M. Choose the best split among these m variables for splitting the node, and this value is typically kept constant throughout the forest growth.

[0042] iii. Each decision tree in the forest is grown to the maximum depth.

[0043] iv. The forest outputs a prediction based on the aggregated predictions of all the decision trees in the forest. The aggregation can be majority vote or average in different examples.

[0044] The training process 500' applied to the random forest classifier 220 uses a reference MS data volume 502, which is accompanied by corresponding experimental conditions and compound-related metadata. The data volume 502 is encoded by the encoder 210 into a set of feature vectors 504 of corresponding feature vectors 212, the true classification of which is known. The set of feature vectors 504 is then split into a validation data set 510 and a training data set 520. The training data set 520 is used to train the random forest classifier 220', while the validation data set 510 is used to evaluate the performance of the classifier. In one example, 75% of the set of feature vectors 504 is used for training, and the remaining 25% is used for evaluation. In other examples, other split ratios can also be used. Here, the notation 220' indicates that the random forest classifier is not yet fully trained. As mentioned above, the corresponding fully trained random forest classifier is denoted by the reference numeral 220 (see, e.g., Fig. 2). Figure 2

[0045] ​When properly configured and trained, the random forest classifier 220 should perform roughly the same on the data sets 510 and 520. The training process 500 includes an evaluation module 530 configured to evaluate the relative performance of the random forest classifier 220' on the data sets 510 and 520. Based on the evaluation 530, adjustments 532 are applied to the random forest classifier 220'. The evaluation 530 and adjustments 532 are iteratively repeated until applicable training stopping criteria are met. At this point, the random forest classifier 220' is locked and considered to have completed training and is suitable for use in the model module 130 Figure 2 The example operations performed during the training process 500 can include some or all of the following: data exploration, exploratory data analysis (EDA), setting thresholds, splitting training data into training and validation data sets, scaling data representing different features to the same scale, instantiating a forest, making predictions, evaluating classifier performance using scoring methods, evaluating classifier performance using confusion matrices, ordering features by importance, selecting a number of trees for the forest, selecting a metric for splitting data sets from a feature vector, adjusting forest parameters using random grid search, creating a dictionary of values to choose from, and adjusting forest parameters using grid search within a defined parameter space.

[0046] Figure 6 is a flowchart illustrating an analyte identification method 600 according to some examples. The method 600 can be implemented using, for example, the ML system 100 shown in Figure 1 The method 600 will be described below in connection with Figure 1 , Figure 2 , Figure 4 and Figure 6 .

[0047] The method 600 includes the ML system 100 Figure 1 obtaining a set of fragmentation spectra and a corresponding set of metadata for analytes from a MS system (in block 602). In some examples, the set of fragmentation spectra is obtained through experimental measurements by the MS system. In some other examples, the set of fragmentation spectra is retrieved from a networked storage, the set of fragmentation spectra having been previously transmitted from the MS system to the storage upon completion of experimental acquisition. In various examples, the set of metadata obtained in block 602 includes ancillary information related to experimental conditions, system configurations, and analyte samples.

[0048] The method 600 also includes the ML system 100 obtaining a mass spectral library 420 Figure 4) to obtain a ranked hit list of items and a corresponding set of metadata (in block 604). In various examples, the ranked hit list obtained in block 604 includes reference spectra corresponding to the set of fragmentation spectra of the analyte obtained in block 602. The hit list is typically ordered in descending order of match scores of the items. In a typical example, the match scores are computed at the mass spectral library 420 using a spectral similarity metric based on methods such as dot product or Spearman rank correlation coefficient of corresponding spectral pairs. The operation of block 604 typically includes submitting a search query with the set of fragmentation spectra obtained in block 602 to the mass spectral library 420 via the service module 120 Figure 1

[0049] In various examples, the set of metadata obtained in blocks 602, 604 can include one or more of the following components: normalized or absolute collision energies; one or more parameters of ion activation methods; a number of candidate compounds; one or more match scores; identities of precursor ions; precursor ion molecular formulas, fragment ion molecular formulas, a number of peaks in the spectra; sparsity measures; spectral peak counts, peak intensities, and corresponding accuracies; one or more distances between peaks in the spectra, neutral losses, peak fragmentation curves; one or more spectral labels; mean; median; average; standard deviation; standard error; relative standard deviation; relative error; and / or variance values of selected numerical features. In some examples, the labels and parameters are grouped into subcategories, including but not limited to query and library spectral metadata, constructed feature metadata differences, scoring statistics (e.g., ranks, sizes, number of unique values, mean, standard deviation, z-scores, absolute deviations) for different groupings (e.g., by query spectrum, by query spectrum and library compound, by query spectrum and library compound groupings), and top scores in library compound groups. The subcategories are then used to compute statistical results for each query spectrum. In some examples, the two sets of metadata can have identical component compositions. In some other examples, the two sets of metadata can have different component compositions.

[0050] The method 600 further includes the ML system 100 evaluating the order of the items in the hit list using the ML model 220 (in block 606). In various examples, the evaluation is based on the two sets of metadata obtained in blocks 602, 604, among other factors. In one example, the operation of block 606 includes: (i) generating a feature vector 212 based on the set of fragmentation spectra of the analyte, the reference spectra from the ranked hit list, and the two sets of metadata obtained in blocks 602, 604 by means of the encoder 210 Figure 2 Figure 2 ​​In response to feature vector 212, ML model 220 may change the order of entries in the hit list in at least some cases. In this case, the operation on block 606 may also include determining an adjustment value for the matching score provided by mass spectrometry library 420 by means of the corresponding reference spectrum of the sorted hit list.

[0051] In some examples, operations on block 606 include ML model 220 determining an estimated probability that the analyte belongs to a specified compound class, an estimated probability that the analyte belongs to a specified chemical class, and / or an estimated probability that the analyte belongs to the same compound class as the compound corresponding to a selected reference spectrum in the sorted hit list, wherein the sorted hit list is received from mass spectral library 420 in block 604. In some cases, at least some of these estimated probabilities may differ from the corresponding probabilities determined at mass spectral library 420.

[0052] Method 600 also includes the ML system 100 taking a responsive action (in block 608). The responsive action taken is typically based on the evaluation results of block 606. In one example, the action on block 608 includes displaying a corrected list of hits, in which the order of entries has been changed, to the user on a display device. In another example, the action on block 608 includes informing the user that the order of entries in the hits list obtained in block 604 remains unchanged. In yet another example, the action on block 608 includes suggesting the chemical identity of the analyte and displaying the corresponding molecular and structural information on a display device. Other responsive actions may also be taken in block 608 in other examples. Method 600 terminates when the action on block 608 is completed.

[0053] Figures 7A-7C The following is a graphical representation of some operations based on method 600 in an example. More specifically, Figure 7A The MS spectra of the analytes involved, acquired by operating the corresponding MS instrument, are displayed graphically in 2021. The normalized collision energy (NCE) corresponding to this experimental operation is NCE = 80%. This NCE value is from metadata 204 ( Figure 2 An example of ), and therefore in Figure 7A The figure is marked with reference numeral 2041. Figure 7B and Figure 7C The results are displayed graphically by searching 416 ( Figure 4 The hit list spectra 2061 and 2062, obtained from mass spectrometry library 420 for the closest match of MS spectrum 2021, are shown. The metadata corresponding to the hit list spectra 2061 and 2062 retrieved from mass spectrometry library 420 along with these spectra contains NCE values ​​of NCE = 20% and NCE = 80%, respectively. These NCE values ​​are examples of metadata 208, and therefore... Figures 7B-7CThe figures are labeled using reference numerals 2081 and 2082 respectively.

[0054] In the example shown, mass spectrum library 420 ( Figure 4 The mass spectrometry library 420 is configured to sort the entries in the hit list based on a spectral similarity metric, such as a cosine (dot product) metric, without considering the NCE value. According to this method, spectrum 2061 has the highest match score among the hit list entries. Therefore, the mass spectrometry library 420 places spectrum 2061 at the top of the hit list. Consequently, spectrum 2062 is listed below spectrum 2061 in the hit list output by the mass spectrometry library 420.

[0055] Based on the sorting of the hit list above, it can be predicted that the compound corresponding to spectrum 2061 is the same as or most closely related to the analyte involved. However, this is an incorrect analyte identification and can be corrected using ML system 100 ( Figure 1 This allows for beneficial corrections. More specifically, by simultaneously considering NCE values ​​2041, 2081, and 2082 as part of metadata 204 and 208, model module 130 ( Figure 1 , Figure 2 The system reorders the hit list received from mass spectrometry library 420. After this reordering, spectrum 2062 appears at the top of the revised hit list. Based on the revised hit list order, it can be predicted that the compound corresponding to spectrum 2062 is the same as or closest to the analyte involved. In this particular example, the latter prediction provides the correct analyte identification.

[0056] Figure 8 This is a block diagram illustrating a computing device 800 according to some examples, one or more instances of which may be used in an ML system 100. In some examples, one instance of the computing device 800 is configured to implement model module 130. In some examples, one or more instances of the computing device 800 may be used in workflow 400, process 500, and / or method 600.

[0057] Figure 8 The computing device 800 shown includes multiple components, but any one or more of these components may be omitted or duplicated depending on the specific application and setup. In some embodiments, some or all of the components included in the computing device 800 may be mounted to one or more motherboards and enclosed in a housing. In some embodiments, some of these components may be integrated into a single system-on-a-chip (SoC) (e.g., the SoC may include one or more electronic processing devices 802 and one or more storage devices 804). Furthermore, in various embodiments, the computing device 800 may not include... Figure 8The computing device 800 can include one or more components as shown in FIG. 8, but can include interface circuitry to couple with one or more components using any suitable interface (e.g., a universal serial bus (USB) interface, a high-definition multimedia interface (HDMI) interface, a controller area network (CAN) interface, a serial peripheral interface (SPI) interface, an Ethernet interface, a wireless interface, or any other suitable interface). For example, the computing device 800 can not include the display device 810, but can include display device interface circuitry (e.g., a connector and drive circuitry) to couple with an external display device 810.

[0058] The computing device 800 includes a processing device 802 (e.g., one or more processing devices). As used herein, the terms “electronic processor device” and “processing device” are interchangeable and refer to any device or portion of a device that processes electronic data from registers and / or memory to transform that electronic data into other electronic data that can be stored in registers and / or memory. In various embodiments, the processing device 802 can include one or more digital signal processors (DSPs), application-specific integrated circuits (ASICs), central processing units (CPUs), graphics processing units (GPUs), server processors, or any other suitable processing devices.

[0059] The computing device 800 also includes a storage device 804 (e.g., one or more storage devices). In various embodiments, the storage device 804 can include one or more memory devices, such as random access memory (RAM) devices (e.g., static RAM (SRAM) devices, magnetic RAM (MRAM) devices, dynamic RAM (DRAM) devices, resistive RAM (RRAM) devices, or conductive-bridge RAM (CBRAM) devices), hard disk drive-based memory devices, solid-state memory devices, network drives, cloud drives, or any combination of memory devices. In some embodiments, the storage device 804 can include memory that is co-packaged on a die with the processing device 802. In such embodiments, the memory can serve as cache memory and include, for example, embedded dynamic random access memory (eDRAM) or spin-transfer torque magnetic random access memory (STT-MRAM). In some embodiments, the storage device 804 can include a non-transitory computer-readable medium having instructions stored thereon that, when executed by one or more processing devices (e.g., the processing device 802), cause the computing device 800 to perform any suitable method disclosed below or portions of such methods.

[0060] The computing device 800 also includes an interface device 806 (such as one or more interface devices 806). In various embodiments, the interface device(s) 806 can include one or more communication chips, connectors, and / or other hardware and software to manage communications between the computing device 800 and other computing devices. For example, the interface device(s) 806 can include circuitry to manage wireless communications, such as the transmission and / or reception of data between the computing device 800 and other computing devices. The term “wireless” and its derivatives can be used to describe circuits, devices, systems, methods, techniques, communications channels, and / or the like that can communicate data through the use of modulated electromagnetic radiation through a non-solid medium. The term does not imply that the associated devices do not contain any wires, although in some embodiments they might not. The circuitry to manage wireless communications included in the interface device(s) 806 can implement one or more wireless standards or protocols, including but not limited to Wi-Fi (IEEE 802.11 family), Bluetooth (including the Bluetooth Low Energy or BLE version), Zigbee, 3G, 4G, 5G, and / or the like. The circuitry to manage wireless communications included in the interface device(s) 806 can operate in accordance with a Global System for Mobile Communications (GSM), Universal Terrestrial Radio Access (UTRA), Bluetooth, and / or the like. In some embodiments, the circuitry to manage wireless communications included in the interface device(s) 806 can operate in accordance with Enhanced Data for GSM Evolution (EDGE), General Packet Radio Service (GPRS) Evolution for Mobile Communications (UMTS), High Speed Packet Access (HSPA), Evolved HSPA (E-HSPA), or LTE network. In some embodiments, the circuitry to manage wireless communications included in the interface device(s) 806 can operate in accordance with Enhanced Data for GSM Evolution (EDGE), GSM EDGE Radio Access Network (GERAN), Universal Terrestrial Radio Access Network (UTRAN), or Evolved UTRAN (E-UTRAN). In some embodiments, the circuitry to manage wireless communications included in the interface device(s) 806 can operate in accordance with Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Digital Enhanced Cordless Telecommunications (DECT), Evolution-Data Optimized (EV-DO), and derivatives thereof, as well as any other wireless protocol that is designated as 3G, 4G, 5G, and beyond. The interface device(s) 806 can include one or more antennas (such as one or more antenna arrays) configured to receive and / or transmit wireless signals.

[0061] In some embodiments, the interface device 806 can include circuitry for managing wired communications, such as electrical communication protocols, optical communication protocols, or any other suitable communication protocols. For example, the interface device 806 can include circuitry that supports communications in accordance with Ethernet technology. In some embodiments, the interface device 806 can support both wireless and wired communications, and / or can support multiple wired communication protocols and / or multiple wireless communication protocols. For example, a first set of circuitry of the interface device 806 can be dedicated to short-range wireless communications such as Wi-Fi or Bluetooth, while a second set of circuitry of the interface device 806 can be dedicated to long-range wireless communications such as Global Positioning System (GPS), EDGE, GPRS, CDMA, WiMAX, LTE, EV-DO, and so on. In some other embodiments, a first set of circuitry of the interface device 806 can be dedicated to wireless communications, and a second set of circuitry of the interface device 806 can be dedicated to wired communications.

[0062] The computing device 800 also includes a battery / power supply circuit 808. In various embodiments, the battery / power supply circuit 808 can include one or more energy storage devices (e.g., batteries or capacitors) and / or circuitry for coupling components of the computing device 800 to an energy source separate from the computing device 800 (e.g., to an AC line power supply).

[0063] The computing device 800 also includes a display device 810 (e.g., one or more separate display devices). In various embodiments, the display device 810 can include any visual indicator, such as a heads-up display, a computer monitor, a projector, a touchscreen display, a liquid crystal display (LCD), a light-emitting diode display, or a flat-panel display.

[0064] The computing device 800 also includes additional input / output (I / O) devices 812. In various embodiments, the I / O devices 812 can include one or more data / signal transmission interfaces, audio I / O devices (e.g., microphones or microphone arrays, speakers, headphones, earbuds, alarms, and so on), audio codecs, video codecs, printers, sensors (e.g., thermocouples or other temperature sensors, humidity sensors, pressure sensors, vibration sensors, and so on), image capture devices (e.g., one or more cameras), human interface devices (e.g., keyboards, cursor control devices such as mice, stylus, trackballs, or touchpads), and so on.

[0065] According to particular embodiments, the various components of interface device 806 and / or I / O device 812 can be configured to output suitable control signals, receive suitable control / telemetry signals, and receive and transmit data streams. In some examples, interface device 806 and / or I / O device 812 includes one or more analog-to-digital converters (ADCs) for converting received analog signals to a digital form suitable for operations performed by processing device 802 and / or storage device 804. In some additional examples, interface device 806 and / or I / O device 812 includes one or more digital-to-analog converters (DACs) for converting digital signals provided by processing device 802 and / or storage device 804 to an analog form suitable for transmission over a communication channel.

[0066] Figure 9 is a block diagram illustrating an MS instrument support system 900 in which some or all of the scientific instrument support methods disclosed herein can be performed according to some examples. Various MS instrument support modules and methods disclosed herein (e.g., system 100, workflow 400, process 500, and / or method 600) can be implemented by one or more of MS instrument 910, user local computing device 920, service computing device 930, and remote computing device 940 of MS instrument support system 900.

[0067] Any of MS instrument 910, user local computing device 920, service computing device 930, or remote computing device 940 can include a computing device 800 as described with reference to Figure 8 Any of MS instrument 910, user local computing device 920, service computing device 930, or remote computing device 940 can take the form of any appropriate embodiment of computing device 800.

[0068] Scientific instrument 910, user local computing device 920, service computing device 930, and / or remote computing device 940 can each include a respective processing device 802, a respective storage device 804, and a respective interface device 806. Processing device 802 can take any suitable form, including any of the forms of processing device 802 discussed herein with reference to Figure 8 Scientific instrument 910, user local computing device 920, service computing device 930, and / or remote computing device 940 can each include a respective processing device 802, a respective storage device 804, and a respective interface device 806. Processing device 802 can take any suitable form, including any of the forms of processing device 802 discussed herein with reference to Figure 8The form of any of the storage devices 804 discussed, and included in the different devices of the scientific instrument 910, user local computing device 920, service computing device 930, or remote computing device 940, can be the same or different. The interface devices 806 can take any suitable form, including those referenced herein with respect to the computing device 800. Figure 8 The form of any of the interface devices 806 discussed, and included in the different devices of the scientific instrument 910, user local computing device 920, service computing device 930, or remote computing device 940, can be the same or different.

[0069] The MS instrument 910, user local computing device 920, service computing device 930, and remote computing device 940 can communicate with other elements of the MS instrument support system 900 via communication paths 908. The communication paths 908 can communicatively couple the interface devices 806 of different elements of the MS instrument support system 900, as shown, and can be wired or wireless communication paths (e.g., according to any of the communication technologies discussed with respect to the interface devices 806 of the computing device 800 herein). Figure 8 Figure 9 The particular MS instrument support system 900 depicted in FIG. 10 includes communication paths between each pair of devices of the scientific instrument 910, user local computing device 920, service computing device 930, and remote computing device 940, but such a “fully connected” implementation is purely illustrative, and in various embodiments various communication paths 908 can not be present. For example, in some embodiments the service computing device 930 can not have a direct communication path 908 between its interface device 806 and the interface device 806 of the MS instrument 910, but can communicate with the MS instrument 910 via the communication path 908 between the service computing device 930 and the user local computing device 920 and the communication path 908 between the user local computing device 920 and the MS instrument 910. The MS instrument 910 can be incorporated into a more general and / or more functional scientific instrument.

[0070] Figures 10A-10B Improvements achievable with the method 600 relative to conventional analyte identification methods are illustrated. More specifically, Figure 10A A query spectrum 1002 corresponding to an example control compound (in this case rutin) used for comparative evaluation of the relative performance of several analyte identification methods is graphically illustrated. Figure 10B ​A table is shown that contains a ranked list of library hits obtained for the query spectrum 1002. The table shown has six columns labeled 1010-1020, respectively. Column 1010 shows the ranking of the entries. Column 1012 shows the match scores for the three algorithms shown. Column 1014 shows the compound names from the library. Column 1016 shows the compound structures. Column 1018 shows the hit summary. Column 1020 shows the hit item metadata from the library. The entries are ranked according to their HighChem HighRes scores.

[0071] The top five hits are characterized by having relatively high similarity scores. However, none of the top twelve hits is the correct actual compound (rutin), which only appears in the thirteenth position. Although the top three hits have HighChem HighRes scores higher than 80, closer inspection of the hit summaries (column 1018) reveals that the number of matching peaks between the query spectrum 1002 and the corresponding library spectrum is relatively low, only two. In addition, as shown by some of the metadata (column 1020), such as the collision energy level, the peak intensities are not matched accurately enough either. In contrast, according to various embodiments, the ML model 130 is trained to learn the relevant metadata and parameters to correctly re-rank the library hits so that the correct hit is pushed closer to the top of the hit list. In the example shown, a well-trained ML model 130 embodiment would have placed Figure 10B The ranking position of rutin in the table is changed from the thirteenth position to a ranking within the top five hits. In other words, the example embodiments disclosed herein would be able to achieve a significant reduction in the occurrence of false positives in the compound identification process.

[0072] According to one example disclosed above, for example in the abstract section and / or in the section Figure 1-1 0, there is provided a method performed by a computing device for providing support for a mass spectrometry (MS) system, the method comprising: obtaining, from a mass spectral library, (i) a ranked hit list of reference spectra corresponding to a set of fragmentation spectra of an analyte acquired with the MS system; and (ii) a first set of metadata corresponding to the ranked hit list of reference spectra; obtaining, from the MS system, a second set of metadata corresponding to the set of fragmentation spectra; and evaluating, with a machine learning (ML) model based on the first set of metadata and the second set of metadata, an order of entries in the ranked hit list.

[0073] In some examples of the above method, obtaining the ranked hit list comprises submitting a search query with the set of fragmentation spectra to the mass spectral library.

[0074] In some examples of any of the above methods, at least one of the first set of metadata and the second set of metadata includes a respective set of one or more parameters selected from the group of parameters consisting of: normalized or absolute collision energy; one or more parameters of an ion activation method; a number of candidate compounds; one or more match score values; identity of a precursor ion; a number of peaks in a spectrum; a sparsity metric; peak intensities and corresponding accuracies; one or more distances between peaks in a spectrum; one or more spectral signatures; and a mean or average of selected numerical features.

[0075] In some examples of any of the above methods, the match score is computed using a dot product of the corresponding pair of spectra.

[0076] In some examples of any of the above methods, the ordered hit list is arranged in descending order of the match scores of the entries.

[0077] In some examples of any of the above methods, the ML model comprises a model selected from the group consisting of: a random forest classifier; a gradient boosting model; a k- nearest neighbor algorithm; and a variational autoencoder.

[0078] In some examples of any of the above methods, the evaluating comprises: generating, with the encoder, a feature vector based on the fragmentation spectrum set of the analyte, the reference spectrum from the ordered hit list, the first set of metadata, and the second set of metadata; and applying the feature vector to the ML model.

[0079] In some examples of any of the above methods, the ML model is configured to change the order of the entries.

[0080] In some examples of any of the above methods, the method further comprises displaying, on a display device, the revised hit list with the changed order of the entries.

[0081] In some examples of any of the above methods, the ML model is configured to determine one or more of: an estimated probability that the analyte belongs to a specified class of compounds; an estimated probability that the analyte belongs to a specified chemical class; and an estimated probability that the analyte belongs to the same chemical class as a compound corresponding to a selected reference spectrum in the ordered hit list.

[0082] In some examples of any of the above methods, at least one of the estimated probabilities is different from a corresponding probability determined at the mass spectral library.

[0083] In some examples of any of the above methods, the ML model is configured to determine, with the respective reference spectrum of the ordered hit list, an adjustment to a match score value provided by the mass spectral library.

[0084] A non-transitory computer-readable medium storing instructions that, when executed by a computing device, cause the computing device to perform operations comprising any of the above methods.

[0085] According to one example disclosed above, e.g. in the abstract section and / or in any combination of the sections Figure 1-1 0, there is provided an apparatus for providing support for a mass spectrometry (MS) system, the apparatus comprising: an interface device; a processing device; and a storage device comprising program code, wherein the storage device and program code are configured to enable the apparatus, by means of the interface device and the processing device, to at least: obtain, from a mass spectral library, (i) a ranked hit list of reference spectra corresponding to a set of fragmentation spectra of an analyte collected by means of the MS system; and (ii) a first set of metadata corresponding to the ranked hit list of reference spectra; obtain, from the MS system, a second set of metadata corresponding to the set of fragmentation spectra; and evaluate an order of items in the ranked hit list by means of a machine learning (ML) model based on the first set of metadata and the second set of metadata.

[0086] In some examples of the above apparatus, at least one of the first set of metadata and the second set of metadata comprises a respective set of one or more parameters selected from the group of parameters consisting of: normalized or absolute collision energy; one or more parameters of an ion activation method; a plurality of candidate compounds; one or more match score values; identity of a precursor ion; number of peaks in a spectrum; sparsity measure; peak intensities and corresponding accuracies; one or more distances between peaks in a spectrum; one or more spectral labels; and mean or average of selected numerical features.

[0087] In some examples of any of the above apparatus, the ML model comprises a model selected from the group consisting of: a random forest classifier; a gradient boosting model; a k-nearest neighbors algorithm; and a variational autoencoder.

[0088] In some examples of any of the above apparatus, the storage device and program code are further configured to enable the apparatus, by means of the interface device and the processing device, to: generate, with an encoder, a feature vector based on the set of fragmentation spectra of the analyte, the reference spectra from the ranked hit list, the first set of metadata, and the second set of metadata; and apply the feature vector to the ML model.

[0089] In some examples of any of the above apparatus, the ML model is configured to change the order of the items.

[0090] In some examples of any of the above apparatus, the storage device and program code are further configured to enable the apparatus, by means of the interface device and the processing device, to display, on a display device, a revised hit list having the changed order of items.

[0091] In some examples of any of the devices described above, the ML model is configured to determine one or more of: an estimated probability that the analyte belongs to a specified chemical class; an estimated probability that the analyte belongs to a specified chemical class; and an estimated probability that the analyte belongs to the same chemical class as a compound corresponding to a selected reference spectrum in the ranked hit list.

[0092] It is to be understood that the above description is intended to be illustrative, and not restrictive. Many specific embodiments and applications other than the examples provided would be apparent to those skilled in the art from the above description. The scope should be determined, not with reference to the above description, but should instead be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled. It is anticipated and intended that future developments will occur in the technologies discussed herein, and that the disclosed systems and methods will be incorporated into such future examples. In sum, it should be understood that the application is capable of modification and variation.

[0093] All terms used in the claims are to be given their broadest interpretation and ordinary meaning, as interpreted by those skilled in the art to which the technology described herein pertains, unless an otherwise expressly set out definition is provided herein. In particular, the use of a singular article, such as "a", "the", "said" and the like, should be understood to mean one or more, unless the context clearly dictates otherwise.

[0094] The Abstract is provided to allow the reader to quickly ascertain the nature of the technical disclosure. It is not intended to be used to interpret or limit the scope or meaning of the claims. In the foregoing Detailed Description, various features are grouped together in examples for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed application requires more features than are explicitly recited in each claim. Rather, inventive subject matter can lie in fewer than all features of a single disclosed example. Thus, the following claims are hereby incorporated into the Detailed Description, where each claim can stand as a separate application.

[0095] Unless specifically stated otherwise, each numerical value and range should be interpreted as approximately as if the word "about" preceded the respective value and range.

[0096] Although elements in the following method claims, if any, are in a specific order, the order is not an absolute requirement unless specifically stated in the claim core itself.

[0097] The use of the ordinal adjectives such as "first," "second," and "third," etc., to refer to a plurality of objects, for example, is merely to distinguish between objects having a same name or description and is not meant to, nor does it, imply that the objects, referred to form an ordered sequence. Such adjectives are used solely to distinguish between two or more nearly-identical or same objects addressed in this Specification, and are not meant to, nor do they, imply the ordering.

[0098] Unless specifically stated otherwise, the term "if' can be or alternatively read as "when" or "while" or "in response to determining" or "in response to ascertaining," the interpretation of which can depend upon the context and particular implementation being presented. For example, a phrase intended to convey the meaning "if A is determined" or "if it is detected that [the condition]” can be interpreted to mean "upon determining A" or "in response to determining A" or "upon detecting [the condition]” or "in response to detecting [the condition]."

[0099] Also for purposes of this description, the terms "couple," "coupling," "connected," "connecting," or the like, shall mean any means of establishing an electrical connection between two or more elements, and it is contemplated that additional intervening elements can be present in said electrical connection, although not necessarily. Conversely, the terms "directly coupled," "directly connected," or the like, imply the absence of additional intervening elements.

[0100] The functionality of the various elements of the drawings shown, including any functional blocks labeled as "processors" and / or "controllers," can be provided through the use of dedicated hardware as well as hardware capable of executing software in association with appropriate software. When provided by a processor, the functions can be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which can be shared. Moreover, explicit use of the term "processor" or "controller" should not be construed to refer exclusively to hardware capable of executing software, and can implicitly include, without limitation, digital signal processor (DSP) hardware, network processor, application specific integrated circuit (ASIC), field programmable gate array (FPGA), read only memory (ROM) for storing software, random access memory (RAM), and non volatile storage. Other hardware, conventional and / or custom, can also be included. Similarly, any switches shown in the drawings are conceptual only. Their function can be carried out through the operation of program logic, through dedicated logic, through the interaction of program control and dedicated logic, or even manually, the particular technique being selectable by the implementer as more specifically understood from the context.

[0101] As used in this application, the term "circuitry" can refer to one or more or all of the following: (a) hardware-only implementations (e.g., in analog and / or digital circuitry); (b) combinations of hardware circuits and software, such as (as applicable): (i) combinations of analog and / or digital hardware circuits with software / firmware, such as (as applicable): (ii) combinations of hardware and software, such as (as applicable): hardware processors and software, including digital signal processors; software, including digital signal processor(s), and memory(ies) that working together are collectively responsible for a device, such as a mobile telephone or server, to perform various functions described herein; (c) hardware, such as a microprocessor(s) or a portion of microprocessor(s), that requires software (e.g., firmware) for operation, either in hardware alone or in hardware and software / firmware

[0102] Those of ordinary skill in the art will appreciate that any block diagrams herein represent illustrative conceptual views of illustrative circuitry embodying the principles of the present disclosure. It will also be appreciated that any flow diagrams, flow charts, state transition diagrams, pseudocode, and the like represent various processes which can be substantially represented in computer readable medium and so executed by a computer or processor, whether or not such computer or processor is explicitly shown.

Claims

1. A method performed via a computing device for providing support for a mass spectrometry (MS) system, the method comprising: obtaining from a mass spectral library: a ranked hit list of reference spectra corresponding to a set of fragmentation spectra of an analyte collected with the MS system; and a first set of metadata corresponding to the ranked hit list of reference spectra; obtaining from the MS system a second set of metadata corresponding to the set of fragmentation spectra; and evaluating an order of items in the ranked hit list with a machine learning (ML) model based on the first set of metadata and the second set of metadata.

2. The method of claim 1, wherein obtaining the ranked hit list comprises submitting a search query with the set of fragmentation spectra to the mass spectral library.

3. The method of claim 1, wherein at least one of the first set of metadata and the second set of metadata comprises a respective set of one or more parameters selected from the group of parameters consisting of: a normalized or absolute collision energy; one or more parameters of an ion activation method; a plurality of candidate compounds; one or more match score values; an identity of a precursor ion; a plurality of peaks in a spectrum; a sparsity measure; a peak intensity and a corresponding accuracy; one or more distances between peaks in a spectrum; one or more spectral labels; and a mean or average of selected numerical features.

4. The method of claim 3, wherein a match score is computed using a dot product of a corresponding pair of spectra.

5. The method of claim 4, wherein the ranked hit list is ordered in descending order of match scores of items.

6. The method of claim 1, wherein the ML model comprises a model selected from the group consisting of: a random forest classifier; a gradient boosting model; a k-nearest neighbor algorithm; and a variational autoencoder.

7. The method of claim 1, wherein evaluating comprises: generating a feature vector based on the set of fragmentation spectra of the analyte, the reference spectra from the ranked hit list, the first set of metadata, and the second set of metadata with an encoder; and applying the feature vector to the ML model.

8. The method of claim 7, wherein the ML model is configured to change the order of items.

9. The method of claim 8, further comprising displaying a revised hit list with the changed order of items on a display device.

10. The method of claim 1, wherein the ML model is configured to determine one or more of: an estimated probability that the analyte belongs to a specified class of compounds; an estimated probability that the analyte belongs to a specified chemical class; and an estimated probability that the analyte belongs to the same chemical class as a compound corresponding to a selected reference spectrum in the ranked hit list.

11. The method of claim 10, wherein at least one of the estimated probabilities is different from a corresponding probability determined at the mass spectral library.

12. The method of claim 11, wherein the ML model is configured to determine, with a respective reference spectrum of the ranked hit list, an adjustment value to a match score value provided by the mass spectral library.

13. A non-transitory computer-readable medium storing instructions that, when executed by a computing device, cause the computing device to perform operations comprising the method of claim 1. ​ ​ ​ 14. An apparatus for providing support to a mass spectrometry (MS) system, the apparatus comprising: an interface device; a processing device; and a storage device comprising program code, wherein the storage device and program code are configured to enable the apparatus, by means of the interface device and the processing device, to at least: obtain, from a mass spectral library: an ordered hit list of reference spectra corresponding to a set of fragmentation spectra of an analyte collected by means of the MS system; and a first set of metadata corresponding to the ordered hit list of reference spectra; obtain, from the MS system, a second set of metadata corresponding to the set of fragmentation spectra; and evaluate, by means of a machine learning (ML) model based on the first set of metadata and the second set of metadata, an order of items in the ordered hit list.

15. The apparatus of claim 14, wherein at least one of the first set of metadata and the second set of metadata comprises a respective set of one or more parameters selected from the following group of parameters: normalization or absolute collision energy; one or more parameters of an ion activation method; a number of candidate compounds; one or more match score values; identity of a precursor ion; a number of peaks in a spectrum; a sparsity measure; peak intensities and corresponding accuracies; one or more distances between peaks in a spectrum; one or more spectral labels; and a mean or average of selected numerical features.

16. The apparatus of claim 14, wherein the ML model comprises a model selected from the following group: a random forest classifier; a gradient boosting model; a k-nearest neighbor algorithm; and a variational autoencoder.

17. The apparatus of claim 14, wherein the storage device and program code are further configured to enable the apparatus, by means of the interface device and the processing device, to: generate, with an encoder, a feature vector based on the set of fragmentation spectra of the analyte, the reference spectra from the ordered hit list, the first set of metadata, and the second set of metadata; and apply the feature vector to the ML model.

18. The apparatus of claim 17, wherein the ML model is configured to change the order of the items.

19. The apparatus of claim 18, wherein the storage device and program code are further configured to enable the apparatus, by means of the interface device and the processing device, to display, on a display device, a revised hit list having the changed order of items.

20. The apparatus of claim 14, wherein the ML model is configured to determine one or more of: an estimated probability that the analyte belongs to a specified class of compounds; an estimated probability that the analyte belongs to a specified chemical class; and an estimated probability that the analyte belongs to the same chemical class as a compound corresponding to a selected reference spectrum in the ordered hit list. ​ ​ ​

Images