Method for reconstructing a mass spectrum
The method reconstructs mass spectra from a series of single mass spectra using algorithms to identify precursor ions and fragments, addressing the limitations of tandem mass spectrometry by simplifying and cost-reducing the analysis process.
Patent Information
- Application Number
- FR2023005634
- Authority / Receiving Office
- FR · FR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-06-05
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-06-05
AI Technical Summary
Existing mass spectrometry techniques, particularly tandem mass spectrometry, face limitations such as increased complexity and cost due to the need for sophisticated instruments, resolution constraints that fail to separate ions with close m/z ratios, and the requirement for prior knowledge of the sample composition, leading to complex and costly analyses.
A method for reconstructing a mass spectrum by analyzing a series of mass spectra to identify precursor ions and their fragments without the need for tandem mass spectrometry, using a single mass spectrometer and in-source collision-induced dissociation, employing algorithms to group and reconstruct individual mass spectra based on correspondence scores and unsupervised learning.
Enables precise identification of precursor ions and their fragments, overcoming resolution and cost constraints, allowing analysis of mixed samples without prior knowledge, and reducing the complexity and cost of mass spectrometry.
Smart Images

Figure 00000046_0000 
Figure 00000046_0001 
Figure 00000047_0000
Abstract
Description
Title of the invention: Method for reconstructing a spectrum of mass Technical field of the invention
[0001] The present invention relates generally to a method for reconstructing a mass spectrum.
[0002] It relates in particular to a method for reconstructing an individual mass spectrum of at least one precursor ion from the exploitation of a series of mass spectra.
[0003] It also relates to a device implementing such a process. State of the art
[0004] Mass spectrometry (MS) is an analytical technique used to detect ions from a sample and to analyze these ions based on their mass-z ratio (m / z), where m represents the mass of an ion and z its electric charge. Mass spectrometry is used in numerous applications to analyze, identify, and characterize the chemical structure of ionized molecules.
[0005] A mass spectrometer generally comprises an ionization source to form ions from a sample to be analyzed, an analyzer that separates the ions according to their m / z ratio, and a detector. A mass spectrum is obtained by recording the abundance of ions as a function of their mass-to-charge ratio (m / z). However, simple mass spectrometry only allows the measurement of the m / z ratio and does not allow the differentiation of ions with identical m / z ratios, particularly due to the resolution limitations of the spectrometer used. Furthermore, measuring the m / z ratio does not provide information on the composition or structure of the ions.
[0006] Tandem mass spectrometry is known to be used to study ions of interest in a sample more precisely. Such a technique makes it possible to provide an individual mass spectrum for a given precursor ion after activation. To this end, tandem mass spectrometry includes, after ionization of the sample molecules, a step of selecting an ion of interest using a mass filter. This selected ion is then subjected to an activation step, which consists of increasing its internal energy, potentially leading to its fragmentation. The ions produced from the fragmentation of the selected ion (also called the precursor ion or parent ion) are then analyzed in one or more other mass spectrometry steps on the fragment ions thus generated, in which the mass analysis steps can be separated spatially or temporally depending on the instrument used.
[0007] For example, tandem mass spectrometry can be implemented by isolating an ion of interest in an ion trap by providing it with sufficient internal energy through collision with a gas to cause fragmentation. Detecting the products of this fragmentation can provide information on the nature and / or structure of the precursor ion. Tandem mass spectrometry is fundamental to the applications of mass spectrometry in structural analysis, particularly the sequencing of proteins and other biopolymers (such as sugars or nucleic acids).
[0008] Tandem mass spectrometry is functional but it requires more elaborate instruments generally comprising two mass analyzers coupled in series, which adds complexity to the spectrometers and also increases their costs and maintenance.
[0009] Tandem mass spectrometry is also limited by instrumental constraints, particularly the selection precision, which in some cases is insufficient to separate two ions of interest with very similar masses (very close m / z ratio). In this case, two different ions with similar m / z ratios will be selected and activated simultaneously. The resulting tandem mass spectrum will therefore be the sum of the tandem mass spectra of the two ions selected simultaneously, which complicates the interpretation of the measurements.
[0010] Tandem mass spectrometry can also be slower in terms of analysis speed compared to other methods due to the sequential acquisition of data. Indeed, the ions of interest are selected and fragmented one after the other.
[0011] Furthermore, in tandem mass spectrometry, a criterion for selecting the ions to be analyzed in a sample is necessary. In this case, it is often necessary to have a list of ions of interest that one wishes to fragment, which amounts to having prior knowledge of the sample. In an alternative approach, it is also possible to choose to fragment all ions whose intensity exceeds a certain threshold, again determined according to the sample. Prior knowledge of the sample is therefore necessary.
[0012] The invention aims to remedy at least one of the aforementioned drawbacks. Presentation of the invention
[0013] In order to remedy the aforementioned drawback of the prior art, the present invention proposes a method for reconstructing a mass spectrum, said method comprising the following steps: - acquisition of a plurality of mass spectra measured on an ionized sample comprising at least one precursor ion and fragment ions, the plurality of spectra mass including at least N mass spectra, - determination of a plurality of abundance value series of the ionized sample from the plurality of mass spectra, each abundance value series of the ionized sample being determined from the plurality of mass spectra, each abundance value series including N abundance values where N is an integer greater than or equal to two; - arranging the plurality of abundance value series into pairs of abundance value series, each pair of abundance value series comprising a first abundance value series and a second abundance value series, - for each pair of abundance value series, calculating a correspondence score between the first abundance value series and the second abundance value series, so as to obtain correspondence scores for all pairs of abundance value series, the correspondence scores providing a correspondence score map between all pairs of abundance value series, - determination, based on a mass-to-charge ratio, of at least one precursor ion and the fragment ions associated with each determined precursor ion, by applying a grouping method to the calculated correspondence scorecard, - reconstruction of an individual mass spectrum of each determined precursor ion and each of the associated fragment ions from the correspondence scorecard.
[0014] Thus, thanks to the invention, and in particular to the various stages of the process according to the invention, it is possible to determine the mass spectrum of an individual ion present in a mixture (i.e., a sample that may contain several ions and fragments) without necessarily using a tandem mass spectrometry device. Therefore, the present invention can perform tandem mass spectrometry analyses from a series of simple mass spectra acquired by a mass spectrometry device that is not equipped with a second mass filter (here, an ion trap) but whose desorption / ionization step either causes the fragmentation of the species, or implements an in-source activation step by collision-induced dissociation (CID), enabling the activation of all the ionized species.Therefore, the process is easier to implement and less expensive compared to a process requiring the use of tandem mass spectrometers.
[0015] Furthermore, although this method works from a series of mass spectra, the calculation step and the use of a classification algorithm make it possible to accurately group at least one precursor ion and the fragments associated with that precursor ion. Such a method is therefore functional on mass spectra acquired with Devices whose selection width is sometimes too wide to precisely isolate a single ion. The invention can also be used to identify the presence of isobaric species in a tandem mass spectrometry experiment. Therefore, the method according to the invention also allows working with all types of acquired mass spectra. In other words, the method overcomes the technical limitations of mass spectrometry devices, which may be related to measurement resolution or constraints associated with the ionization sources used.
[0016] Other advantageous and non-limiting features of the process according to the invention, taken individually or according to all technically possible combinations.
[0017] In one embodiment, the step of determining a series of abundance values includes determining a series of mass-to-charge ratio ranges for the plurality of measured mass spectra of size NxM with M an integer greater than 1, and for each mass spectrum of the plurality of mass spectra, a step of integrating abundance values belonging to the same mass-to-charge ratio range of the mass spectrum under consideration, to provide, for each mass-to-charge ratio range of the spectrum under consideration, an integrated abundance value associated with a mass-to-charge ratio determined in each mass-to-charge ratio range, the integrated abundance values over the series of mass-to-charge ratio ranges forming a series of abundance values of the ionized sample.
[0018] In one embodiment, the matching score is based on at least one of the following: - mutual information between the first set of abundance values and the second set of abundance values, said mutual information being based on an entropy of the first set of abundance values and on an entropy of the second set of abundance values; and / or - a correlation between the first series of abundance values and the second series of abundance values.
[0019] In one embodiment, for mutual information, the entropy of the first series of abundance values is an entropy normalized by a first invariance parameter as a function of a first median value determined on this first series of abundance values and the entropy of the second series of abundance values is an entropy normalized by a second invariance parameter as a function of a second median value determined on this second series of abundance values.
[0020] In one embodiment, the matching score being based on mutual information, the method includes a step of estimating the entropy of the first set of abundance values and a step of estimating the entropy of the second set of abundance values using a k nearest neighbors algorithm.
[0021] In one embodiment, when at least two matching scores based on two different elements are calculated in the calculation step, several scorecards are provided as a result of the calculation step, said method further comprising, for each scorecard, a step of filtering said scorecard according to a contrast of the scorecard in question.
[0022] In one embodiment, the method further comprises a step of processing the scorecard before applying the grouping method, said processing step using at least one of the following algorithms: - standardization algorithm, - algorithm for processing diagonal elements of the scorecard, - application of a square function.
[0023] In one embodiment, the scorecard is a two-dimensional matrix, the grouping method determining the at least one precursor ion and its associated fragment ions from the matching scores located outside a diagonal of the scorecard.
[0024] In one embodiment, the grouping method uses at least one unsupervised learning algorithm.
[0025] In one embodiment, the grouping method is based on at least one of the following algorithms: - a hierarchical ascending classification algorithm, - a k-nearest neighbors algorithm, - a k-means algorithm, - a stochastically distributed neighbor incorporation algorithm.
[0026] In one embodiment, the process further comprises, after the reconstruction step, a step of analyzing the individual mass spectrum of the determined precursor ion and the associated fragment ions to characterize the chemical composition of the determined precursor ion and the chemical composition of the fragment ions associated with the considered precursor ion.
[0027] In one embodiment, the method includes a step of processing the acquired mass spectra comprising at least one of the following calculations: - filtering of the mass spectrum considered by a Gaussian or normal nucleus, - a normalization of the mass spectrum under consideration, - a smoothing of the mass spectrum under consideration, - a return to the baseline.
[0028] In one embodiment, the method includes a step for processing the series of abundance values comprising at least one of the following calculations: - a filtering of the series of given abundance values by a Kalman filter; - a filtering of the series of given abundance values by a Fourier filter, - a filtering by moving average of the series of given abundance values.
[0029] In one embodiment, the method includes a compressed acquisition step to reduce the number of acquisitions.
[0030] In one embodiment, the plurality of mass spectra comprises at least 1000 mass spectra, preferably between 2000 and 8000 mass spectra.
[0031] Of course, the various features, variants, and embodiments of the invention can be combined with one another in various ways, provided they are not incompatible or mutually exclusive. Detailed description of the invention
[0032] The following description with regard to the attached drawings, given by way of non-limiting examples, will make it clear what the invention consists of and how it can be carried out.
[0033] On the attached drawings:
[0034] [Fig. 1] is a schematic representation of a device for reconstructing an individual mass spectrum of at least one precursor ion according to the present disclosure;
[0035] [Fig.2] is a schematic representation of a mass spectrometry device used in the device according to this disclosure;
[0036] [Fig.3] is a schematic representation of an example of a process according to this disclosure;
[0037] [Fig.4] is a schematic representation of an embodiment of an acquisition step of a process according to this disclosure.
[0038] [Fig.5] is an example of an average mass spectrum determined in an acquisition step of a process according to this disclosure;
[0039] [Fig.6] is a schematic representation of ranges of m / z ratio values determined from an average mass spectrum obtained in an acquisition step of a process according to this disclosure;
[0040] [Fig.7] is a representation of an example of a first set of abundance values and a second set of abundance values obtained in an arrangement step of a process according to this disclosure;
[0041] [Fig.8] is a representation of the first set of abundance values and the second set of abundance values illustrated in [Fig.7] for which the abundance values of each of the sets have been filtered by a Kalman filter in a process according to the present disclosure;
[0042] [Fig.9] is a schematic representation of a determination step comprising in a step of calculating a matching score in a process according to this disclosure, said score being based on a calculation of mutual information;
[0043] [Fig. 10] is a succession of curves each illustrating a cloud of points of a plurality of mutual information determined from non-invariant entropies;
[0044] [Fig. 11] is a succession of curves each illustrating a cloud of points of a plurality of mutual information determined from invariant entropies according to the present disclosure;
[0045] [Fig. 12] is a scorecard obtained at a step in calculating a matching score based on mutual information by applying a process according to this disclosure;
[0046] [Fig. 13] is a scorecard obtained at a calculation step of a correspondence score based on a partial correlation by applying a process according to this disclosure;
[0047] [Fig. 14] is a scorecard obtained at a step in calculating a correspondence score based on a partial correlation by applying a process according to this disclosure, the correlation scores being determined from pairs of abundance value series filtered by a Kalman filter;
[0048] [Fig. 15] is a schematic representation of a dendrogram obtained by a grouping method used in a process according to this disclosure;
[0049] [Fig. 16] is an example representation of an individual mass spectrum of a precursor ion of an ionized sample, said individual mass spectrum being obtained according to a process according to this disclosure;
[0050] [Fig. 17] is an example representation of another individual mass spectrum of another ionized sample precursor ion, said individual mass spectrum being obtained by a process according to this disclosure;
[0051] [Fig. 18] is a schematic representation of a two-dimensional map obtained by a grouping method used in a process according to this disclosure;
[0052] [Fig. 19] is a schematic representation of the two-dimensional map obtained from [Fig. 18] on which another grouping algorithm has been applied.
[0053] A device 100 for reconstructing a mass spectrum of an individual ion present in a sample according to the present disclosure will be described using [Fig.1] and [Fig.2].
[0054] The device illustrated in [Fig.1] includes a measurement module 110 configured to determine at least one mass spectrum 2, preferably a plurality of mass spectra.
[0055] To this end, the measuring module 110 includes a mass spectrometry device 10 as illustrated in [Fig. 2]. The mass spectrometry device 10 can understand a single-stage mass spectrometer device (as opposed to a tandem mass spectrometer).
[0056] In the example illustrated in [Fig. 2], the mass spectrometry device 10 is a single-stage device. It includes an ionization source 11 for forming ions from the sample 1 to be analyzed and an activation means 12 configured to fragment the ions. In the present disclosure, the chemical composition of the sample is unknown. Therefore, it comprises a mixture including one or more ions suspended in a solution. In other words, the sample may comprise different molecular species.
[0057] In the present disclosure, sample 1 is, for example, a solution containing molecules. This sample 1 is then placed in the gas phase and is subsequently ionized by the ionization source 11.
[0058] The ionization source 11 shall comprise at least one of the following ionization sources, namely: an electron impact ionization source, an electro-nebulization source, a chemical ionization source, a photo-ionization source, a laser ionization source, a matrix-assisted laser-induced desorption ionization source (known as a MALDI source), a surface-assisted desorption ionization (SELDI) source, an atmospheric pressure MALDI source, a matrix-assisted rapid evaporation ionization source, any other process involving a matrix or a surface, any process involving a laser, a plasma ionization desorption source, an atmospheric pressure chemical ionization source or an atmospheric pressure photoionization source, an atmospheric pressure ionization source.
[0059] In a preferred embodiment of the invention, the ionization source 11 comprises an electron-bulking source. These ions are then activated by the activation means 12 configured to fragment the gas-phase ionized ions in an in-source CID (carbon-in-source ionization) step. All ions are activated simultaneously and can produce electrically charged fragments, or fragment ions, also referred to as fragments in the remainder of this disclosure.
[0060] In these different variants, the activation means 12 includes at least one of the following means: activation by collision, activation by absorption of electromagnetic radiation, thermal activation, etc.
[0061] In another embodiment, the mass spectrometry device 10 lacks an activation means 12. In this case, the ionization source can also be configured to fragment the ions of the sample. In other words, this type of ionization source 11 provides sufficient energy during ionization to produce fragments. In practice, such an ionization source 11 includes, for example, an electron impact ionization source or an ionization source 11 using laser desorption technology (LDI).
[0062] In another embodiment (not illustrated), the mass spectrometry device 10 comprises a first mass filter for selecting a range of m / z ratios of interest, an activation means 12, and a second mass filter for analyzing fragmentation products resulting from the activation. In this embodiment, one or more precursors are selected and activated simultaneously. Each of these ions will produce one or more fragment ions. The recorded tandem mass spectrum therefore includes all the generated fragments and, optionally, the precursor ions. Such a mass spectrometry device is also called a two-stage mass spectrometry device.
[0063] The ions obtained at the outlet of the activation means 12 are called ionized samples. They include, in particular, at least one ion, called a precursor ion.
[0064] In this disclosure, a precursor ion or parent ion is defined as an ion that can be broken down into different fragments. Therefore, a fragment ion (or fragment) is a part or chemical structure of the parent ion.
[0065] The mass spectrometry device 10 also includes a mass filter 13 (also called an analyzer) which separates at least one precursor ion as well as the different fragments of the ionized sample according to their m / z ratio.
[0066] In this disclosure, the abundance of each ion can be used to obtain quantitative information on the concentration of the molecular species in the sample. The mass spectrometry device 10 also includes a detector 14 configured to detect ionized species (ionized ions). The mass spectrum 2 is obtained by recording the abundance of the ions as a function of their mass-to-charge ratio (m / z).
[0067] A mass spectrum 2 thus provides information on the chemical composition of the sample, in particular on the number of species present, typically ions and fragments, and on their respective mass.
[0068] As illustrated in [Fig.2], the mass spectrometry device lacks a second mass filter required to perform a tandem mass spectrometry measurement.
[0069] In this disclosure, the mass spectrometry device 10 is also configured to acquire a time series of mass spectra 2. By time series, it is understood that the mass spectrometry device 10 successively (over time) acquires mass spectra. In practice, the sample 1 used to acquire these mass spectra 2 is identical for each mass spectrum 2 acquisition by the mass spectrometry device 10. However, the ionized sample for each acquisition (of mass spectrum) may vary during the acquisitions.
[0070] Optionally, the mass spectrometry device 10 may also include a computer (not shown) including a memory for storing the mass spectra acquired at the output of the detector.
[0071] The device 100 illustrated in [Fig. 1] also includes a processing unit 120.
[0072] By processing unit 120 is meant a computer, a processor, a central processing unit (CPU), or any electronic device capable of executing a series of commands and / or calculations. Typically, the processing unit 120 includes a processor, memory, and various input and output interfaces. For this purpose, the processing unit 120 is configured to control the operations and functions performed by the device 100. The input and output interface can be the connection point enabling the device 100 to receive data and send commands or data. The processing unit 120 can be connected to other processing units to exchange information with them.
[0073] The processing unit 120 comprises for this purpose one or more (computing) modules configured to perform a series of commands and / or calculations which will be explained below with [Fig. 3]. Typically, the processing unit is configured to implement the process explained below.
[0074] The device 100 may also include a communication module 130 configured to exchange data with elements connected to the device 100 and / or elements internal to the device 100. By exchanging data, we mean sending and / or receiving data, the received data corresponding in this example to input data which may include a multitude of mass spectra, for example a part of the series of mass spectra or the entire series of mass spectra acquired by the measurement module 110.
[0075] In this example, the communication module 130 receives the mass spectra determined by the measurement module 110 and stores them in the storage module 140.
[0076] The communication module 130 is optionally connected in this example to a database 4, external to the device 100. The database 4 typically includes previously acquired mass spectra.
[0077] Database 4 is used in particular when device 100 does not include a measuring device 110. Therefore, in this case, device 100 does not directly measure mass spectra 2 but extracts previously recorded mass spectra 2 from database 4. Consequently, it is possible to work with previous mass spectra, as no measurements are then necessary. In this case, these received mass spectra (from database 4) were preferably all acquired by the same mass spectrometry device 10.
[0078] In the example considered, the communication module 130 receives mass spectra 2 from the measuring device 110 (for example, a series of mass spectra) and from the database 4. Similarly, these received mass spectra were acquired by the same mass spectrometry device 10. In other words, the mass spectra in the database were previously acquired by the measuring device 110 and stored in the database, which is external to the device 100. In a variant, this database is contained within the measuring device 110 (memory of the device 10) but remains separate from the memory of the measuring device 110.
[0079] To receive data from database 4, the communication module 130 is also configured to be connected to at least one communication network 5, which may be a wireless or wired communication network. In practice, the wireless communication network uses at least one of the following technologies: Wi-Fi, Bluetooth, 2G / 3G / 4G / 5G, etc. The wired communication network uses ADSL, etc. Thus, in device 100, the mass spectra measured by the measuring device 110 or extracted from database 4 are received by the communication module 130, which transfers this data to the storage module 140 of device 100.
[0080] The memory module 140 includes, for example, a memory and / or a hard disk.
[0081] The memory module 140 stores, in particular, computer program instructions designed so that the processing unit 120 implements at least some of the steps of the process described below with reference to [Fig.3] when these instructions are executed by the processing unit 120.
[0082] All mass spectra received by the communication module 130 are concatenated to form a single series of mass spectra. This single series of mass spectra comprises at least 1000 mass spectra, preferably between 2000 and 8000 mass spectra.
[0083] Thus, the mass spectra measured by the measuring device and / or the mass spectra extracted from database 4 are concatenated in the storage module 140 to form a plurality of mass spectra.
[0084] The storage module 140 thus stores this plurality of mass spectra.
[0085] Preferably, the plurality of mass spectra is a time series of mass spectra. In other words, all the mass spectra in the series of mass spectra are measured by the same mass spectrometry device 10 for the same sample 1. Such a configuration has the advantage of obtaining a more precise series of mass spectra because they are directly acquired with the same mass spectrometry device 10. In an alternative, the series of mass spectra can be acquired by Several similar mass spectrometers were used under similar measurement conditions. This has the advantage of making it faster to obtain the mass spectra series.
[0086] Fig. 3 represents an example of a process implemented on device 100.
[0087] This method 1000 is a method for reconstructing a mass spectrum of at least one precursor ion from a plurality of mass spectra measured on the ionized sample comprising at least one precursor ion and fragment ions.
[0088] In practice, the process 1000 is implemented by computer, for example by being implemented by the device 100 illustrated in [Fig.1].
[0089] The 1000 method can also be implemented in microcontrollers or programmable gate networks (e.g. FPGA).
[0090] This method 1000 includes an acquisition step 1010 of a series of mass spectra on the ionized sample comprising at least one precursor ion and fragment ions. In this disclosure, acquisition means the direct measurement of mass spectra and / or the reception of mass spectra.
[0091] To this end, the acquisition step 1010 includes a preliminary (optional) step 1001 to retrieve the plurality of mass spectra. In practice, the preliminary step 1001 begins with a measurement step 1002 of mass spectra by the measuring unit 110 of the device 100 and / or a reception step 1004 of mass spectra from the database 4. An example of a preliminary step is illustrated in [Fig. 4]. In this example, the preliminary step includes the mass spectrum measurement step 1002 by the measurement unit 110 and the mass spectra reception step 1004 from database 4. In this example, all of these mass spectra (from measurement module 110 and database 4) are concatenated in a concatenation step 1006 in memory module 4 to form the plurality of mass spectra of at least one precursor ion and fragment ions.
[0092] This plurality of mass spectra (or series of mass spectra) is then detected and acquired 1008, by the processing unit 120 of the device 100, to reconstruct a mass spectrum of at least one precursor ion from this series of mass spectra.
[0093] The process 1000 illustrated in [Fig. 3] then optionally includes a processing step 1015 for the plurality of acquired mass spectra. Typically, the processing unit 120 processes all the mass spectra of the plurality of mass spectra.
[0094] Various processing steps can be performed on the mass spectra of the mass spectrum series. The processing steps include at least one of the following: - smoothing, for example, using a Savitzky-Golay algorithm with an al- baseline correction algorithm, etc. - normalization, for example by dividing each acquired mass spectrum by a constant a that is a function of the acquired mass spectrum. In practice, the constant a corresponds to the abundance (intensity) of a peak in the mass spectrum, in particular the abundance of the most intense peak (i.e., the one with the highest intensity). Thus, for normalization, each peak in the mass spectrum considered is divided by the abundance of the most intense peak in the mass spectrum; - peak extraction known in English as peak picking algorithm; - elimination or filtering, for example by using a signal-to-noise ratio to remove irrelevant mass spectra; - deisotoping, - grouping of data, for example by techniques known as binning; - filtering, for example by convolving each spectrum into a particular kernel.
[0095] By way of non-limitation, the processing unit 120 can filter each mass spectrum from the plurality of mass spectra in order to attenuate noise on each spectrum related to the measurement of these mass spectra that can impact the abundances of the m / z ratios. Here, typically, the processing unit 120 determines the signal-to-noise ratio of each mass spectrum from the plurality of mass spectra and eliminates, for example, mass spectra with a signal-to-noise ratio less than or equal to 0.5.
[0096] Processing step 1015 may also include compressed sensing. This processing may be performed as an alternative to the processing listed below or in combination with them.
[0097] After filtering, the processing unit 120 normalizes each mass spectrum from the plurality of acquired mass spectra. To do this, for each mass spectrum, the processing unit 120 determines the peak with the maximum abundance (for example, using a peak-picking algorithm) and then divides the abundances of the other peaks in the given mass spectrum by the previously determined maximum value. The normalization constant a is thus made to vary from one mass spectrum to another.
[0098] After normalization, the processing unit 120 can perform peak alignment processing to select ranges of given m / z ratios. This has the advantage of subsequently working on only a portion of the mass spectra in the mass spectrum series.
[0099] Following the treatment step 1015 (if present in the process 1000), the process 1000 then comprises a determination step 1020 of a plurality of series of abundance values of the ionized sample. In practice, in the determination step 1020, a series of abundance values of at least one precursor ion and a series of Abundance values for each fragment ion are determined. In this disclosure, each set of abundance values comprises N abundance values, where N is an integer greater than or equal to two. In practice, each integer assigned from 1 to N corresponds to one mass spectrum out of the plurality of mass spectra.
[0100] In a first embodiment, the determination step 1020 can be implemented by the processing unit 120 as follows. For example, the processing unit 120 iterates through each mass spectrum used as input data for the determination step 1020 and, for each mass spectrum, retrieves and stores an abundance value for each m / z ratio, thus creating a series of abundance values (i.e., equivalent to a list of abundance values) for each m / z ratio of the mass spectra in the plurality of mass spectra. Here, a series of abundance values is understood to be a succession of abundance values obtained from the same given m / z ratio. Thus, typically, for a given m / z ratio, each value in the series of abundance values corresponds to an acquisition (i.e., a measurement), that is, to a given mass spectrum N.Therefore, in a series of abundance values, there are as many abundance values as there are mass spectra used in the determination step 1020. In this example, there are thus as many series of abundance values as there are m / z ratios detected in the mass spectra of the mass spectrum series. Each series of abundance values is typically stored in the memory of the processing unit 120 and / or in the storage module 140.
[0101] In a second variant, the values of the abundance series correspond to abundance values integrated over a range of m / z ratio values.
[0102] To obtain this type of abundance value series, the processing unit 120 can first calculate an average mass spectrum from the plurality of mass spectra by averaging, for each m / z ratio of the mass spectra of the plurality of mass spectra, the abundance values. An average spectrum is thus obtained.
[0103] For example, [Fig. 5] illustrates an average mass spectrum obtained from the plurality of mass spectra acquired from a sample comprising melittin and substance P. The average mass spectrum shown in [Fig. 5] was obtained from a plurality of mass spectra. Each of the plurality of mass spectra was obtained by electron fogging a mixture comprising melittin and substance P in solution in a mixture of water and methanol having a concentration of 105 mol / L. The mixture was, for example, activated by ultraviolet radiation of 20 eV for 500 ms to provide the ionized sample.
[0104] The processing unit 120 can then determine ranges of m / z ratio values from the mean mass spectrum, for example by detecting peaks in the mean mass spectrum and calculating for each peak (22, 23, 24), a range of m / z ratios include at least the peak under consideration.
[0105] For example, [Fig. 6] shows a schematic representation of a mean mass spectrum obtained from the N mass spectra of the plurality of mass spectra. In particular, in this example, each interval W1, W2, W3 covers all the abundance values of the peak under consideration 22, 23, 24 and can also cover m / z ratio values around the peak under consideration. Thus, a range of m / z ratio values corresponds to a subset of the m / z ratios of the mean mass spectrum. In this example, three ranges of m / z ratio values W1, W2, and W3 are detected by the processing unit. In this variant, a range of m / z ratio values W is associated with a peak of the mean mass spectrum. These ranges of m / z ratio values are then recorded.The processing unit 120 can create a series of m / z ratio value ranges of size NxM, where N corresponds to the number of mass spectra in the plurality of mass spectra (here an integer greater than 2) and M corresponds to the number of m / z ratio value ranges (here an integer greater than 2), and which, in the example of [Fig. 6], comprises three elements (the ranges W1, W2, W3). In practice, the M m / z ratio value ranges are similar for each mass spectrum N, which allows all mass spectra to be integrated in a similar way.
[0106] Subsequently, the processing unit 120 can, for each mass spectrum, integrate abundance values over given portions of each mass spectrum, each given portion corresponding to a previously determined range of m / z ratio values. Typically, for a mass spectrum N, the mass spectrum under consideration is integrated over different ranges of values defined by the M ranges of values of the series of m / z ratio values at the coordinate N of the series of m / z ratio values.
[0107] An integrated abundance value is thus obtained for each determined range of m / z ratio values. Integrating the abundance values over the same range of m / z ratio values across the plurality of mass spectra yields an abundance series for an ion or fragment ion. Thus, from all ranges of m / z ratio values, series of abundance values are obtained.
[0108] In this embodiment, each series of abundance values is subsequently associated with a specific m / z ratio within the considered range of m / z ratio values. The specific m / z ratio can correspond to the median value of the m / z ratio within the range of m / z ratio values, or preferably to the m / z ratio associated with a maximum abundance value in the mean mass spectrum over the given range of m / z ratio values. Thus, according to this variant, each abundance series is associated with a peak in the mean spectrum. Fewer series of abundance values are therefore calculated in this variant, which improves the speed of the process. Furthermore, since each series of abundance values is calculated over a defined m / z ratio range based on a peak in the mean spectrum, each series of abundance values can be considered to correspond to an ion in the ionized sample. Furthermore, integrating the abundance values The analysis of the sample allows for the association of ions from the ionized sample belonging to the same isotopic distribution. When the resolution permits, the isotopes are separated, but the goal is to regroup them into a single ion. Of course, this second approach, which could correspond to an isotope regrouping step (de-isotoping), can be performed using other algorithms known to those skilled in the art.
[0109] In practice, the series of abundance values determined in the determination step 1020 can be recorded and grouped in the same file.
[0110] The process 1000 optionally includes a step 1025 for processing the series of abundance values comprising at least one of the following calculations: - filtering the given abundance series using a Kalman filter; - filtering the given abundance series using a Fourier filter; - filtering the given abundance series using a moving average.
[0111] Such a step 1025 allows noise to be removed from the abundance values of abundance value series. For example, in integration over intervals (second variant of the determination step 1020), the processing unit 120 can apply a time filter to each interval. This has the advantage of attenuating noise throughout the acquisition and highlighting temporal variations. Typically, the Kalman filter may be preferred for removing integration noise. Typically, [Fig. 8] illustrates an example comprising an abundance value series 32 of an ion and a filtered abundance value series 33 of the same ion, here filtered by a Kalman filter. It can be seen that the abundance values of the filtered series 33 fluctuate less. The signal is more stable.
[0112] Subsequently, the process 1000 includes a step of arranging the plurality of abundance value series into pairs of abundance value series, each pair of abundance value series comprising a first abundance value series (denoted X) and a second abundance value series (denoted Y).
[0113] In practice, the processing unit 120 retrieves a series of abundance values from the plurality of series of values and combines it with another series of abundance values from the plurality of series of abundance values. Preferably, there can be as many combinations as there are series of abundance values. Thus, for P determined series of abundance values (with P an integer greater than 2, P2 pairs of series of abundance values are determined by the processing unit 120, considering that a series of abundance values can form a pair of series of abundance values with itself).
[0114] Fig. 7 illustrates a pair of abundance value series consisting of a first series of abundance values 30 associated in this example with an m / z ratio of 1365 and a second series of abundance values 31 associated in this example with an m / z ratio of 1424. Here, N corresponds to 5532.
[0115] The process 1000 also includes, for each pair of abundance value series, a calculation step 1040 of a correspondence score between the first abundance value series and the second abundance value series. This correspondence score is thus calculated for all pairs of abundance value series. The set of correspondence scores obtained for all pairs of abundance value series provides a scorecard, which is typically a symmetric matrix of size PxP.
[0116] In practice, the matching score is based on at least one of the following: - mutual information between the first series of abundance values and the second series of abundance values, said mutual information being based on an entropy of the first series of abundance values and on an entropy of the second series of abundance values; - a correlation between the first series of abundance values and the second series of abundance values.
[0117] The matching score determined in calculation step 1040 is based on one of the cases mentioned above or on a mixture of the cases mentioned above. However, the scorecard is composed of matching scores of the same type, that is to say, calculated in the same way using the same calculation method.
[0118] Of course, the processing unit 120 can calculate different correspondence scores, for example a score based on mutual information, a score based on correlation and a score based on mutual information and on the previously calculated correlation.
[0119] The different correspondence scores that can be determined at calculation step 1040 will be explained below.
[0120] These example matching scores are determined for each pair of previously recorded abundance value series. For the sake of simplicity, the calculations described below describe an application of calculation step 1040 to a single pair of abundance value series.
[0121] Mutual information
[0122] In this variant, the calculation step 1040 includes a mutual information determination step 1041. Different embodiments of the mutual information calculation step will be described below.
[0123] Figures 9, 10, 11, and 12 illustrate an embodiment in which the matching score is based on mutual information (mathematics). In this embodiment, the mutual information is determined by an information theory method, which will be described below, that uses entropy calculations.
[0124] Mutual information between two variables
[0125] In a first embodiment of step 1041, the mutual information between two variables is defined by the following formula:
[0126] [Math.l] / (XY)=h(X)+h(Y)-h(X, Y)
[0127] with X corresponding to a first variable (here, for example, it may correspond to the first series of abundance values defined for an ion of the ionized mixture, the ion being a precursor ion or a parent ion) and Y a second variable (here, for example, it may correspond to the second series of abundance values defined for an ion of the ionized mixture).
[0128] Here, h() corresponds to the entropy (also called intrinsic entropy) defined for a given random variable expressed by the mathematical function:
[0129] [Math.2] h[x^ = J
[0130] where px(x) corresponds to the probability density function of the continuous random variable X (corresponding to the probability that a signal appears in the variable X) (here the abundance values of each acquisition considered).
[0131] The entropy h() can also be written for non-continuous variables by the following formula: p * log (pÿ) with corresponding to the probability that a signal appears in the variable X.
[0132] The entropy h() cannot be directly calculated in the present example. It is therefore necessary to approximate this quantity h() by means of an estimation.
[0133] Thus, to determine the correspondence score of each pair of abundance value series, the processing unit 120 estimates (in an estimation step A1 of the mutual information determination step 1041), for each pair, the entropy of the first abundance value series and the entropy of the second abundance value series using an estimation algorithm. These estimated entropies are denoted H.
[0134] The estimation step Al will be described according to this disclosure.
[0135] In the present disclosure, we seek an equivalence between the entropy h() and the estimated entropy H().
[0136] In practice, the entropy estimated for a variable X (corresponding here, for example, to the first series of abundance values) is expressed according to the formula:
[0137] [Math.3] H(X] = d^average^^ distance (L)) + log(y) + psi(n)~ psi(L)
[0138] with d corresponding to the dimension of the variable X (here equal to 1), Y distanced} corresponding to the set of distance values determined by the estimation algorithm (the calculation of this distance is explained below), n corresponding to the number of abundance values in the abundance series (here n is equal to N), L corresponding to the number of neighbors considered in the estimation algorithm (here equal to 1, for example, when the distance is determined for a single neighbor), psi() corresponding to a Digamma function that can be expressed as psi(x^ = 9ui) can satisfy a recursion psi(x+l) = psi (x) + 1 / x with psi(l) = -C for C = 0.5777, where C is the Euler-Mascheroni constant, and V corresponds to the volume of a unit hypersphere expressed according to the following formula:
[0139] [Math.4] ^[rl - v L 1 r(*+i)
[0140] with r the radius of the hypersphere (equal to 1), a gamma function F can be expressed by r.
[0141] Different methods can be used to estimate the entropy of the first set of abundance values and the second set of abundance values. In particular, the processing unit 120 can use at least one of the following methods: a histogram-based method, a k-nearest neighbors algorithm-based method, a kernel density estimation method, etc.
[0142] In practice, in the present disclosure, the processing unit 120 can use a k-nearest neighbors (kNN) algorithm to estimate the entropy H of a series of given abundance values used in the calculation of mutual information. The kNN algorithm is chosen because this method is robust, compared to an estimation method using a histogram, and works on small sample sizes.
[0143] Typically, the kNN algorithm uses a distance calculation between the abundance values of the same series in order to approximate the probability density of the series.
[0144] In practice, the k nearest neighbors method uses a Euclidean distance to estimate the joint entropy between the data (abundance values) of the first set of abundance values and the data of the second set of abundance.
[0145] In this example, the Euclidean distance used depends on the first set of abundance values and the second set of abundance values considered, in particular on the treatments that may have been carried out in the treatment step 1015.
[0146] In the following, two consecutive abundance values from the first series of abundance values (associated with an ion) are denoted by A = (x, xi+i) and two values Consecutive abundances of the second series of abundance values (associated with an ion) are denoted by B = (yi5 yi+1), with x; an abundance value of the first series of abundance values for an acquisition i, with xi+i an abundance value of the first series of abundance values for an acquisition i+1, with y; an abundance value of the second series of abundance values for an acquisition i and with yi+i an abundance value of the second series of abundance values for an acquisition i+1.
[0147] In practice, when no normalization is calculated at the treatment step 1015 or when the process 1000 does not include a treatment step 1015, the Euclidean distance between the abundance values of a first series of abundance values and the abundance values of the second series of abundance values is expressed by the formula:
[0148] [Math.5]
[0149] with A a variable (i.e. data corresponding here to the first series of abundance values (A=X)), B a variable (i.e. data corresponding here to the second series of abundance values (B=X)), x; a variable corresponding to a signal from A (therefore here to an abundance value included in the first series of abundance values at acquisition i, i.e. an abundance value of an ion of the ionized sample), xi+iau signal following x; (therefore the abundance value of the ion at acquisition i+1), y; a variable corresponding to a signal from B (therefore here to an abundance value included in the second series of abundance values at acquisition i, i.e. an abundance value of another ion of the ionized sample), yi+iau signal following y; (therefore corresponding to an abundance value of the other ion at acquisition i+1).
[0150] In the case of normalization of mass spectra at processing step 1015, the Euclidean distance defined by formula Math. 5 does not apply. Therefore, the entropy h() is not directly equivalent to the entropy estimated H() by the estimation algorithm.
[0151] Indeed, the entropy of a variable X (here the variable X corresponds to a series of abundance values of an ion included in the abundance series pair) is not invariant under multiplication, as described by the following formula:
[0152] [Math.6] hkkX) =h(X) + log(\k\)
[0153] with k is a multiplicative constant, X a variable corresponding here to a series of abundances of an ion (for example a precursor ion), h corresponding to the intrinsic entropy of the variable X.
[0154] In this variant, when estimating the entropy of the first series of abundance values and the second series of abundance values from data that have been normalized by the same normalization constant, denoted k (for example when a = k), the entropy is also invariant under multiplication by a multiplicative constant k (here k = a) as described by the following formula:
[0155] [Math.7] h(kX, kY) — h(X, K) + log(k2)
[0156] with k is a multiplicative constant, X a variable corresponding in this case to the first series of abundance values, Y a variable corresponding in this case to the second series of abundance values.
[0157] Similarly, equation Math. 7 does not apply directly when the multiplicative constant of the first set of abundance values is different from that of the second set of abundance values, for example when the normalization constant that was used to determine the data of the first set of abundance values is different from that which was used to determine the data of the second set of abundance values.
[0158] Thus, in the case of normalization at the processing step 1015, to have an equivalence between the entropy h() and the estimated entropy H, it is necessary to work from invariant entropy and invariant estimated entropy.
[0159] By invariant, we mean that the entropy considered, for example the entropy estimated on a series of given abundance values (H(X), with X the series of given abundance values), is equal to the estimated entropy calculated on the same series of abundance values multiplied by a constant (H(kX) with X the series of given abundance values and k the multiplicative constant, here possibly equal to the normalization constant used in the processing step 1015).
[0160] To obtain such an equivalence, the entropy estimated from the first series of abundance values used in the calculation of mutual information in two variables is normalized by a first invariance parameter and the entropy estimated from the second series of abundance values used in the calculation of mutual information in two variables is normalized by a second invariance parameter.
[0161] In practice, the processing unit 120 uses, for a series of given abundance values, the invariant parameter m(X) (also called invariant measure) defined according to the following three properties: i) m(X) = rx >0 ; ii) m(kX) = km(X) = krx iii) m(X + b) = m(X) = rx. with X the series of abundance values considered, m(X) the invariant measure of the series of abundance values considered, rx> 0 being a median value of the set of distances between each pair of abundances closest in the series considered, k a chosen multiplicative constant (which may depend on the normalization constant a of the processing step) and b an additive constant whose value may, for example, depend on noise (for example, noise included in the abundance values of the series of abundance values).
[0162] Using the properties defined above, we can obtain an equivalence between the entropy h() and the estimated entropy H(), such that:
[0163] [Math. 8] h(x) = / / (7^)
[0164] with X a series of given abundance values, h the intrinsic entropy, H the estimated entropy determined in the calculation step, m(x) the invariant measure. (It should be noted that the invariance property also allows for the equivalence H(X) = 1
[0165] Formula Math. 8 can be generalized to an entropy between several variables, typically when several series of abundance values are extracted from a plurality of mass spectra. In other words, it is particularly useful when several series of abundance values of ionized ions and / or fragments of ionized ions are to be considered. In this case, the equivalence of formula Math. 8 generalizes to n dimensions (n = N):
[0166] [Math.9] h(xyz) -2- -2-1
[0167] with X, Y.. ..Z corresponding to the series of abundance values of the ion X or Y or Z considered.
[0168] Thus, the equivalence sought for a calculated entropy between two variables is written as follows:
[0169] [Math. 10] htx y1 -2- 1 Ij n\m(xy m{Y} J
[0170] with X and Y two series of abundance values where X is associated hereafter with the first series of abundance values belonging to a pair of given abundance value series (here corresponding to a series of abundance values of an ion of the ionized sample) and Y with the second series of abundance values belonging to the same pair (here corresponding to a series of abundance values of another ion of the ionized sample), m(X) with the first invariance parameter of series X (i.e., the invariance parameter of the first series of abundance values) and m(Y) with the second invariance parameter of series Y (i.e., the invariance parameter of the second series of abundance values), h the entropy and H the estimated entropy.
[0171] Based on the above properties, when the multiplicative constant of the first series of abundance values (here A=X) is identical to the multiplicative constant of the second series of abundance values (here B=Y), the Euclidean distance used in the mutual information determination step 1041 can also be expressed by the following formula:
[0172] [Math. 11] d[kA, kBj = (kXj-kxi+ 3 )" + ( / cy- ky.+]) - kd^
[0173] Indeed, by multiplying the first series of abundance values and the second series of abundance values by the same multiplicative constant (k), the abundances at each acquisition of the first series of abundances are written kA = (kx, kxi+i) and the abundances at each acquisition of the second series of abundances are written kB = (ky i, kyi+i). Typically, k can be a negative or positive number.
[0174] When the multiplicative constant of the first series of abundance values (denoted here aj) is not identical to the multiplicative constant of the second series of abundance values (denoted here a2), the Euclidean distance used in the determination step 1041 can also be expressed by the following formula:
[0175] [Math. 12] _ __ _ _ OjB] = J
[0176] with ai being the multiplicative constant of the first series of abundance values and a2 the multiplicative constant of the second series of abundance values. Similarly, al and a2 are two constants that can be negative or positive.
[0177] To calculate such a distance d(aiA, a2B), the processing unit 120 can perform a change of variable by setting:
[0178] [Math. 13]
[0179] [Math. 14]
[0180] Using formulas Math. 13 and Math. 14, we obtain the following equations:
[0181] [Math. 15] U|A — ~ a (G- , And
[0182] [Math. 16] rtjy aB / 7 W --- ■■■»■■■ --- ....-7..., , And
[0183] [Math. 17] 4=(¾¾1) , And
[0184] [Math. 18] 5 = (¾¾)
[0185] The processing unit 120 thus uses the formulas described above to define the Euclidean distance in its estimation algorithm according to the formula described below:
[0186] [Math. 19] / > / aix \2 / thy «2y d^A, a2B ) + \^77 - tte- J
[0187] , , I y ” 77 y 75 z , rf(aiA,^B) (^-^) =d(A,B )
[0188] It will now be described how the processing unit 120 determines the invariance parameter of the two series of abundance values of each pair of series of abundance values (determination step of the invariance parameter Ail of each series of the pair of series considered in the estimation step Al).
[0189] As explained above, each invariant measure is associated with a series of abundance values. The processing unit 120 therefore determines, for each series of abundance values determined previously, an invariant measure (step Ail). This invariant measure, as explained above, depends on the abundance values of the series of abundance values.
[0190] In practice, the invariant measure can be determined upstream of the determination estimation 1041, for example following the determination step 1020.
[0191] An example of a step for determining an invariance parameter on a first series of abundance values and an invariance parameter on a second series of abundance values (step Ail) will therefore be described.
[0192] For a given series of abundance values (for example, the first series of abundance values), the processing unit 120 calculates (step PI) all the differences between the abundance values included in the series of abundance values. To do this, the processing unit 120 iterates through the series of abundance values and calculates, for each abundance value, a difference with all the other abundance values in the series of abundance values. Thus, for each abundance value in the series of abundance values, Nl differences are determined. A list of difference values is, for example, stored for each value of abundance.
[0193] The processing unit 120 then searches, for each abundance value in the series of abundance values, for the smallest difference between that abundance value and another abundance value included in the series of abundance values (step P2). To do this, the processing unit scans the list of differences associated with the given abundance value and searches for the smallest difference obtained in the list of differences associated with that abundance value. As a result, each abundance value is associated with a minimum distance. Thus, N minimum distances are determined for this series of abundance values. These minimum distances are then sorted in ascending order.
[0194] The processing unit 120 then determines a median from these N minimum distances (step P3) and then multiplies this median value by the number N of abundance values in the series of abundance values considered (step P4). This last calculated value is then assigned to the invariance parameter (here the first invariance parameter) of the series considered.
[0195] These steps are also carried out on the other series of the mass spectrum pair (for example on the second series) to obtain the second invariance parameter.
[0196] Thus, the step of determining the invariance parameter Al 1 provides, for each series of abundance values, an invariance parameter. The invariance parameter respects the three properties i, ii) and ii) explained above.
[0197] The estimation step A1 includes a normalization step (step A12) of each abundance value series by its invariance parameter. Thus, for example, in this step, the processing unit 120 divides each abundance value in the considered abundance value series by the invariance parameter determined for that series. Hereafter, the first series of normalized abundance values is denoted by -Xr for the first abundance value series, and the second series of normalized abundance values is denoted by -Xr for the second abundance value series.
[0198] Therefore, after the normalization step, each series of abundance values is normalized.
[0199] The entropy estimation in the estimation step performed on the pairs of mass spectra uses these series of normalized abundance values.
[0200] Thus, in process 1000, in the estimation step Al (included in the mutual information determination step 1041), the processing unit 120 determines (step A13) an estimated invariant entropy using the estimation algorithm (for example, the kNN algorithm in this example) as explained below.
[0201] In practice, in this example, the estimation algorithm determines for - the invariant entropy / 4%) — x 1 for the ion associated with the first spectrum of \ 7 \ m( X ) / mass, - the invariant entropy ht y ] — H { y 1 for the ion associated with the second spectrum of ■ / \ «413 / mass, and - the two-dimensional invariant entropy y\ — fil xyj as defined \ ' / \ m(X) ' m(Y) / in equation Math. 10 using equation Math. 3 as defined above to determine each estimated entropy H. The distance formula used in the kNN algorithm is the Euclidean distance defined in formula Math. 12 with k preferably equal to 1.
[0202] In the mutual information determination step 1041, the processing unit determines (step A2) the conditional information I(X,Y) between the first abundance series and the second abundance series on the basis of the formula Math. 1, now written, thanks to the previous calculations using the estimated entropies H:
[0203] [Math.20] I(X, Y) = H(X) + H(Y) - H(X, Y)
[0204] The mutual information determination step 1041 is calculated for each pair of series in the mass spectrum series. Thus, for each pair of abundance value series, the calculated mutual information I corresponds to a matching score between the first abundance series and the second abundance series of a given abundance series pair.
[0205] Thus, the set of correspondence scores calculated in calculation step 1040 (in a step 1042) can be represented as a two-dimensional matrix, with, for example, along the rows, all the first series of abundance values for the set of pairs, and along the columns, all the second series of abundance values for the pairs. Each "pixel" of the matrix corresponds to a correspondence score calculated from a given pair of ions, the ions being identifiable by the row and column of the matrix. This data, illustrated in two dimensions, corresponds to the scorecard (which is a two-dimensional matrix).
[0206] Figure 10 illustrates various scatter plots of mutual information calculated from non-invariant estimated entropies H (i.e., from a first set of abundance values and a second set of abundance values not normalized by their invariance parameter, as explained above). These curves were plotted with the abundance values (intensity) of one ion (here, in particular, the ion associated with a first set of abundance values, which has an m / z ratio of 1222) on the x-axis and the abundance values of the other ion in the pair considered (here, the ion associated with the second set of abundance values, which has an m / z ratio of 1712) on the y-axis. In this example, a protein mixture comprising ubiquitin, cytochrome C, and lysozyme is in solution in a mixture of water and methanol. exhibiting a concentration of 105 mol / L. A region centered on m / z 1760 and with a width of 100 m / z was selected for a tandem mass spectrometry step. The selected ions were then subjected to activation by 20 eV ultraviolet radiation for 500 ms. Typically, here, the multiplicative constant ka was applied to the first set of abundance values, and the estimated values for both sets were calculated using formula Math. 3 with a given multiplicative constant k, specifically with k equal to 1 for curve 50, k equal to 5 for curve 51, and k equal to 10 for curve 52.
[0207] Figure 11 illustrates various scatter plots of mutual information calculated from an estimated invariant entropy H (i.e., from a first set of abundance values and a second set of abundance values normalized by their invariance parameter as explained above). Typically, the estimated entropy H for both sets was calculated using formula Math. 3 with a given multiplicative constant k, with k equal to 1 for curve 53, k equal to 5 for curve 54, and k equal to 10 for curve 55.
[0208] It can be seen that for curves 50, 51, and 52, the mutual information determined with different values of the multiplicative constant is different. Thus, the multiplicative constant used in the calculation of the entropy of the first series of abundance values affects the calculation of the mutual information. Conversely, for curves 53, 54, and 55, the mutual information determined with different values of the multiplicative constant is similar. Thus, the multiplicative constant used in the calculation of the entropy of the first series of abundance values does not affect the calculation of the mutual information.
[0209] Consequently, thanks to the invention, the calculated mutual information is invariant, that is, it is constant due to the estimated invariant entropy and therefore does not depend on the abundances of the ions in the spectra used. The estimated invariant entropy according to this disclosure thus makes it possible to express any estimated entropy value in a single reference frame.
[0210] In process 1000, specifically in step 1041, mutual information is calculated for all pairs of abundance value series. All of this mutual information can be concatenated into a scorecard (step 1042). Thus, in this example, the set of mutual information determined in step 1041 yields the scorecard.
[0211] Figure 12 illustrates a scorecard obtained from a correspondence score based on a mutual information calculation. The mutual information calculation was obtained from a series of abundance values obtained on an ionized sample, the abundance value series comprising values integrated over 10 ranges of mass-to-charge ratio values (i.e., according to the second variant of the step of decomposition). termination). The ionized sample is the same as that in [Fig. 10] and [Fig. 12], comprising a protein mixture including ubiquitin, cytochrome C, and lysozyme. In this figure, the 10 ranges of m / z ratio values for the ions in the first series of pair abundance values are shown in the columns, and the 10 ranges of m / z ratio values for the ions in the second series of pair abundance values are shown in the rows. As illustrated in this figure, each range of values is associated with an m / z ratio determined as described above.
[0212] The resulting scorecard can be seen to be a symmetric matrix, which has the advantage of requiring subsequent calculations to be performed on only half of the scorecard. Furthermore, since fewer pairs of abundance value series were used, the implementation time of the 1000 process is improved.
[0213] A second embodiment of step 1041 will now be described. In the following, the calculated mutual information is directly written from the estimated mutual information H thanks to the equivalence between the entropy h and the entropy H discussed above.
[0214] Conditional mutual information.
[0215] According to this embodiment, the mutual information used is conditional mutual information which is defined by the following formula:
[0216] [Math.21] I(X, Y\Z ) = H(X, Z) + H(Y, Z) - H(X, Y,Z)-H(Z)
[0217] with X, Y, and Z an additional variable, called a conditional variable, H() the estimated entropy.
[0218] In practice, the conditional variable Z comprises at least one of the following: the total number of ions, known as the "Total Ions Current" (TIC). Typically, the total number of ions is the result of summing all the ions in a mass spectrum (or plurality of mass spectra), a series of abundance values for another ion in the mixture, a series of constant abundance values, and so on. Hereafter, the TIC is considered. In this case of the TIC, each mass spectrum in the plurality of mass spectra has a TIC. All the TICs in the plurality of mass spectra form a series of TICs that are used as a partial or conditional variable.
[0219] Therefore, from the series of mass spectra, the processing unit 120 can calculate a TIC (the conditional variable) for each mass spectrum of the plurality of mass spectra. This can, for example, be done upstream, for example in the determination step 1020. Thus, from the plurality of mass spectra, the processing unit 120 obtains a series of conditional variables composed of p conditional variables with p equal to the total number of mass spectra of the plurality of mass spectra (here p = N).
[0220] Conditional mutual information is determined using a process similar to that explained above for the example of mutual information between two variables.
[0221] Thus, in computation step 1041, the processing unit uses the kNN algorithm to estimate the entropies H(X, Z), H(Y, Z), H(X, X, Z) and H(Z) of equation Math. 21. Similarly, each estimated entropy term is an invariant entropy as explained above.
[0222] In practice, the conditional mutual information estimation step includes steps A1 to A2 as described above included in the mutual information calculation step 1041.
[0223] Thus, for each pair of abundance value series, an invariant measure is determined for each variable X (corresponding to the first abundance value series), Y (corresponding to the second abundance value series), Z (corresponding to the TIC value series) in step Al 1 of the entropy estimation step H of the calculation step 1041.
[0224] In practice, to estimate the entropy H(Z) of the conditional variable, the processing unit 120 directly calculates, from the series of conditional variables Z, the invariant measure m(Z) associated with this series. To obtain the estimated entropy H(Z) of this series, the processing unit 120 normalizes the series of conditional values by dividing each conditional variable of the series of conditional variables Z by the invariant measure m(z) (calculation of ). Similarly, the processing unit 120 determines the invariant entropy H(Z) using the formula Math. 3 with L = 1 (a neighbor) (step Al). In this same step Al, the estimated entropies H(X, Z), H(Y, Z) and H(X, Y, Z) are determined in this step Al.
[0225] In practice, in this example, the estimation algorithm also determines in this step (in addition to the previous example) the following estimated entropies: n\m(Xy) m(Z) J h(yz) = Vet \ ' / \ m(Y)' m(Z) / H(X, Y, Z] — h( ) ■ \ / \ m(X) ' m(Y) ' m(Z) / With X the first series of abundance values, Y the second series of abundance values, and Z the series of conditional variables. For this, the processing unit 120 uses equation Math. 3 as defined above to determine each estimated entropy H. The distance formula used in the kNN algorithm is the Euclidean distance defined in formula Math. 11.
[0226] Subsequently, the processing unit determines the conditional mutual information by applying the formula Math. 20.
[0227] A second embodiment of step 1041 will now be described.
[0228] Mutual information between three variables
[0229] According to this embodiment, mutual information is conditional mutual information between three variables, called an information score, and which is defined by the following formula:
[0230] [Math.22] I(X, Y, Z) = H(X) + H(Y) + H(Z) - H(X, Y) - H(X, Z) - H(Y, Z) + H(X, Y, Z)
[0231] In practice, the conditional variable Z is also a series of abundance values, which can be, as in the second example described above, a series of TIC values (allowing us to eliminate internal variations within the system) or a series of abundance values of a third ion (allowing us to highlight relationships related to this third ion) selected from the plurality of mass spectra (similarly to the ion associated with the first and second series of abundance values).
[0232] The conditional mutual information between three variables is determined using a procedure similar to that explained above for the example of conditional mutual information.
[0233] Thus, the processing unit uses the kNN algorithm to estimate the terms H(X), H(Y), H(Z), H(X, Y), H(X, Z), H(Y, Z), and H(X, X, Z) of equation Math. 22. Similarly, each estimated entropy term is an invariant entropy as explained above.
[0234] In practice, the calculation step 1041 of the mutual information between three variables includes steps A1 to A2 as described above.
[0235] The processing unit 120 estimates the invariant entropy of each term H(X), H(Y), H(Z), H(X, Y) H(X, Z), H(Y, Z), and H(X, X, Z) of equation Math. 22, in a manner similar to the first and second examples described above.
[0236] Subsequently, the processing unit 120 determines the conditional mutual information by applying the formula Math. 22.
[0237] As explained above, the correspondence score can also be based on a correlation between the first set of abundance values and the second set of abundance values. Therefore, a variant of calculation step 1040 will now be described in which the correspondence score is based on a correlation.
[0238] Correlation
[0239] According to a variant of method 1000, the correspondence score can be based on a statistical method. In practice, in this embodiment, method 1000 is based on a calculation of the correlation coefficient but lacks an entropy calculation.
[0240] In this embodiment, the processing unit 120 can calculate the correlation coefficient rxy (in a step of calculating a correlation coefficient 1044) on each pair of abundance series from the covariance according to the following formula:
[0241] [Math.23] Horn -
[0242] with Cov(X, Y) denoting the covariance of the variables X (corresponding for example to the first series of abundance values) and Y (corresponding for example to the second series of abundance values), sx corresponding to the standard deviation of the variable X and sy corresponding to the standard deviation of the variable Y. Typically, the standard deviation of the variable considered corresponds to the standard deviation determined from all the abundance values of the abundance series considered.
[0243] In practice, according to this embodiment, the processing unit 120 calculates, for each pair of abundance value series, the covariance between the first abundance value series and the second abundance value series according to the formula for the covariance of a random variable defined as follows:
[0244] [Math.24] Cov(X, Y) = E(XY) - E(X)*E(Y)
[0245] with E() corresponding to the mathematical expectation of the given variable, here the mathematical expectation of the set of abundance values of the abundance series considered.
[0246] Next, the processing unit 120 determines a standard deviation of the variable X and a standard deviation of the variable Y, the standard deviation considered being defined according to the following formula:
[0247] [Math.25] J = TÏMxy = Je(x 2 )-(ex)) 2
[0248] with X the variable considered (here the first series of abundance values or second series of abundance values), E(X) the expectation of the random variable considered, Var(X) the variance of the random variable considered.
[0249] Thus, in this embodiment, the processing unit 120 calculates, for each pair of abundance value series, the covariance as defined in equation Math. 24, as well as the standard deviation sx associated with the first abundance series and the standard deviation sy associated with the second abundance value series. It then determines the correlation coefficient Corxy for each pair of abundance value series.
[0250] Since the calculation step 1040 is performed iteratively by repeating the calculation step on each pair of abundance series values or in parallel on all pairs of abundance series, a multitude of Corxy correlations are provided at the end of the step 1040.
[0251] Similar to the embodiment using a correspondence score based on the calculation of mutual information, all the correlation coefficients calculated on all the pairs of abundance value series are grouped (step 1042) on a two-dimensional matrix (corresponding to a scorecard) with a shape similar to that obtained in the context of mutual information. Thus, compared to [Fig. 12], the matrix is made up of Corxy correlation coefficients. The scale of such a scorecard is, however, different from the example with mutual information, since the calculated correlation coefficients can be negative or positive (unlike mutual information where the score values are positive).
[0252] Furthermore, although not necessary, it is also possible to determine the correlation coefficients from series of normalized abundance values. For example, the normalization coefficient can correspond to the invariance parameter of the series under consideration and be calculated in a similar manner to the example of mutual information.
[0253] A second embodiment of step 1044 will now be described.
[0254] Partial correlation
[0255] According to another variant of method 1000, the processing unit 120 can calculate the partial correlation coefficient Corxyz between two variables X, Y, and a third variable Z. This third variable is similar to the variable Z used in the calculation of mutual information. In other words, the variable Z is a series of TIC values or a series of abundance values of another ion, different from the ion associated with the first and second series of abundance values. It can also be a series of constant values or a series of values relating to the intensity of a laser beam, a gas pressure, etc. Using a third variable Z eliminates the need for a quantity affecting the measurements.
[0256] In this case, the step of calculating the correlation coefficient 1044 can be implemented as follows.
[0257] In practice, the partial correlation coefficient of each pair of abundance value series is defined according to the following equation:
[0258] [Math.26] _ CorXY-Corxz*Cori2 ^orXYZ~ r ^ / 777-7-^1-Cin xz ^l-Coryz
[0259] with Cor (), the correlation coefficient between the variables considered.
[0260] For this purpose, the processing unit 120 calculates for each pair of abundance value series, the correlation coefficients Coryx, Corxz and Coryz as determined previously using formula Math. 23 and then applies formula Math. 26 to determine the partial correlation coefficient between three variables.
[0261] The correspondence score obtained by this embodiment is more precise than the correspondence score of the embodiment based on a calculation of correlation coefficient between two variables as explained above.
[0262] Typically, the numerator of equation 23 is equivalent to a standard deviation normalization. Such a standard deviation normalization in the correlation calculation has the same effect as normalization by the invariant measure in the entropy calculation.
[0263] Figure 13 illustrates a scorecard obtained from a correspondence score based on a correlation coefficient (here, a partial correlation coefficient). The results shown correspond to the ionized sample comprising melittin and substance P. The correspondence scores were determined from series of abundance values obtained from an ionized sample, the series of abundance values including non-integrated values (i.e., according to the first variant of determination step 1020). In this figure, the m / z ratio values of the ions from the first series of pair abundance values are shown in the columns (x-axis), and the m / z ratio values of the ions from the second series of pair abundance values are shown in the rows (y-axis).
[0264] It can be seen that the resulting scorecard is a symmetric matrix. Furthermore, it can be seen that more pairs of abundance value series were used, so the implementation time of process 1000 may be longer. The matching scores according to this method can be positive or negative.
[0265] Figure 14 illustrates a matching scorecard obtained with series of abundance values filtered by a Kalman filter as previously shown (i.e., obtained from the mixture of melittin and substance P). It can be seen that the scorecard exhibits better contrast. Contrast can be assessed by taking the average of the matching scores of all pixels in the matching map. In a preferred variant, contrast can be assessed by calculating a variance over the set of matching scores (i.e., all pixels) in the scorecard. In this case, the higher the variance, the greater the contrast.
[0266] A combination of the statistical and probabilistic method for determining the correspondence score calculated at calculation step 1040 will now be described.
[0267] As explained above, the correspondence score can also be based on a correlation between the first set of abundance values and the second set of abundance values and on mutual information between the first set of abundance values and the second set of abundance values.
[0268] Combination of the statistical method and the probabilistic method described above.
[0269] In this embodiment, the correspondence score is based on the probabilistic method based on an entropy calculation applied to the first series of abundance values and the second series of abundance values and on the statistical method based on a calculation of a correlation between the first series of abundance values and the second series of abundance values.
[0270] Thus, in practice, the processing unit 120 determines, for each pair of abundance value series, two preliminary correspondence scores by calculating for each pair of abundance value series: - at least one of the mutual information (step 1041) described above I(X, Y), I(X, YIZ), I(X, Y, Z), that is, by calculating the mutual information defined by one of the following formulas: Math. 20, Math. 21, Math. 22; and - at least one of the correlation scores (step 1044) described above Corxy, Corxyz, i.e. by calculating the correlation coefficient defined by one of the following formulas: Math. 23, Math. 26.
[0271] Preferably, in this embodiment, the processing unit 120 combines the mutual information between three variables I(X, Y, Z) with the partial correlation coefficient Corxyz.
[0272] In this case, the processing unit 120 can calculate a mixed coefficient Sa(X, Y, Z) (step 1045) on each pair of abundance value series which is defined according to the following formula:
[0273] [Math.27] Y, Z) = I(X, Y, Z) * sign(CorXYX)
[0274] with sign(Corxyz) corresponding to the sign of the partial correlation coefficient and I(X, Y, Z) corresponding to the mutual information between three variables.
[0275] In a variant of process 1000, processing unit 120 can weight at least one of the following terms, namely the mutual information between three variables, the partial correlation coefficient, in order to obtain a more precise correspondence score. In this case, processing unit 120 calculates a weighted mixed coefficient Sap(X, Y, Z) on each pair of abundance series, which can be defined according to the following formula:
[0276] [Math.28] Sa^X, Y, Z) = (alpha*I(X, Y, Z\)\beta* CorXYZ)
[0277] with alpha a first weighting coefficient greater than or equal to 1 and beta a second positive weighting coefficient greater than or equal to 1.
[0278] In practice, the first weighting coefficient depends on the correspondence score based on the mutual information considered and the second weighting coefficient depends on the information score based on the correlation coefficient considered.
[0279] For example, the correspondence score can be calculated according to the formula Math. 28 in which the conditional mutual information (i.e., the correspondence score based on the conditional mutual information) is multiplied by the partial correlation coefficient and in which the first weighting coefficient is equal to 1 and the second term of the equation Math. 28 is written sign(beta*Corxyz-0.1) with beta, second correlation coefficient equal to 1. In this example, the value 0.1 is selected because it allows to reduce a correlation intrinsic to the fragmentation process from which the term TIC cannot be freed.
[0280] In a variant, the equation Math. 28 can also be written: Sa^X, Y, Z) = Y, Z)\ g(CorYrz)) with f() and g() two functions that They apply a transformation to each coefficient (here I(X, Y, Z) and Corxyz). These functions can be, for example, a square function, an affine function, a sign function, etc. The function h(a,b) is a function that takes two variables and applies a transformation such as addition or multiplication.
[0281] The other steps of the process 1000 will now be described after obtaining one or more correlation maps from the set of correspondence scores described above.
[0282] In some cases, the correlation map(s) obtained in step 1042 are unusable because they exhibit insufficient contrast. In this case, process 1000 optionally includes a selection step 1050 of the correlation maps based on the contrast of the scorecard under consideration.
[0283] Such a step makes it possible to eliminate unusable correlation maps in order to limit errors in the use of correlation maps in the continuation of process 1000.
[0284] In practice, this selection step 1050 can be directly calculated in the calculation step 1040 after determining the scorecard associated with a given matching score. In this case, the processing unit 120 calculates, after determining the scorecard, a contrast of the given scorecard and then compares this contrast to a given threshold to retain or eliminate this matching card. For example, the threshold can be a value greater than or equal to 0.30, for example, 0.5.
[0285] Typically, in an example, the scorecard resulting from the two-variable mutual information-based correspondence score may have a contrast of less than 0.3. As a result, following selection step 1050, this scorecard may be deleted from the memory of the processing unit or the storage module 140.
[0286] If only one type of scorecard is calculated in the calculation step, only one correspondence map is obtained as a result of step 1040. In this case, if this map has a contrast below the threshold, the process can repeat the calculation step by determining correspondence scores based on another calculation method (i.e., going, for example, from mutual information between two variables to mutual information between three variables, or going to a score based on a correlation coefficient or using a mixed correlation score, etc.).
[0287] The correlation map(s) passing the requirement of step 1050 are retained in the memory of the processing unit and / or the storage module 140. Thus, in the continuation of process 1000, only the correlation map(s) retained following the selection step 1050 are used in the continuation of process 1000. Of course, if process 1000 does not include a selection step 1050 as described above, all the correlation maps determined in the calculation step 1040 are used in process 1000.
[0288] In the illustrated example, following the selection step 1050, the process 1000 includes a determination step 1060, based on a mass-to-charge ratio, of at least one precursor ion and associated fragment ions. For this purpose, the process 1000 uses a classification method that is applied to the correlation map(s) calculated at the end of the calculation step 1040.
[0289] Thus, in this step 1060, the processing unit 120 applies a grouping method to each scorecard to determine at least one precursor ion and the fragment ions associated with this precursor ion. This precursor ion is determined based on an m / z ratio.
[0290] In the case where several precursor ions are determined in this step, each determined precursor ion is associated with its fragment ions.
[0291] Optionally, before applying the grouping method to the correlation maps, process 1000 may include a processing step 1062 of the correlation maps. In practice, this processing step 1062 is carried out in the determination step 1060 but before the application of the classification method.
[0292] Typically, processing step 1062 uses at least one of the following treatments: - an algorithm for processing the diagonal of the given scorecard. In this case, the processing unit 120 can assign the same score value (for example, an integer) to each element present on the diagonal of the scorecard. For example, all elements on the diagonal can be set to zero so that these elements do not significantly influence the classification algorithm. In the case of a scorecard based on correlation coefficients, the integer can be negative; - an algorithm to standardize the correlation maps. In this case, the standardization algorithm can be applied to each row of the correlation maps. This standardization algorithm uses the mean of a given row and its standard deviation. In practice, the processing unit 120 calculates the mean and standard deviation of each row. Then, for each element in a given row, the processing unit subtracts the mean value of that row from the element and divides that value by the standard deviation of that row. Thus, with this processing, each row of the scorecard has a similar weight in the scorecard processed; - an application of a square function in order to accentuate the difference between the small and large coefficients (correspondence scores) of the scorecard considered.
[0293] In this example, process 1000 then uses a grouping (or classification) method on the processed correlation maps (step 1064).
[0294] Typically, the clustering algorithm uses at least one unsupervised learning algorithm which may include at least one of the following algorithms: - a hierarchical ascending classification algorithm, for example using Ward's method (as an aggregation method) on the variance, and / or the Full Link method, - a k-nearest neighbors algorithm, - a k-means algorithm, - a dimensionality reduction algorithm, also known in English as "algorithm t-distributed stochastic neighbor embedding" (t-SNE).
[0295] In the grouping method, each classification algorithm takes a scorecard as input. Thus, if several correlation cards are calculated in step 1040 or retained following the selection step, the processing unit 120 will implement several grouping algorithms.
[0296] The grouping method can also define in the input data other elements, including at least one of the following parameters: the method used (here for example the Ward method in the case of the unsupervised learning algorithm); the distance used in the (aggregation) method used; etc. These data depend on the type of algorithm used and are therefore known to the person skilled in the art.
[0297] An embodiment of the determination step 1060 used in process 1000 will now be described using [Fig.15], [Fig.16], [Fig.17], [Fig.18] and [Fig.19].
[0298] In the example of process 1000, the determination step 1060 uses a hierarchical clustering algorithm. In practice, in this example, the algorithm used employs Ward's method and defaults to a Euclidean distance. Ward's method is more suitable because it minimizes variance by cluster merging and is recommended for continuous values, unlike the other methods mentioned above, such as the Complete Linkage method, which considers maximum values rather than a global set of the scorecard under consideration.
[0299] Figure 15 illustrates a dendrogram obtained by the hierarchical clustering method used and derived from data calculated on the ionized sample from a mixture of melittin and substance P. Here, the dendrogram includes The x-axis represents the different labels, corresponding in this example to the m / z ratios of precursor and fragment ions, and the y-axis represents the distance between different classes. In this example, each class is characterized by a horizontal line on the dendrogram. Vertical branches / links connect the classes to illustrate the relationships between them.
[0300] As illustrated in [Fig. 15], the dendrogram includes a main class Cl which is divided into two main vertical branches Cil, C12, each main vertical branch being again connected to a new class C2, thus obtaining a tree structure in which the vertical branches illustrate the links / proximities between the different classes connected to these branches.
[0301] Thus, it can be deduced from this dendrogram that the initial mixture comprises two precursor ions (visible through the two links of the main branch). Each main branch is connected to a new class (called a secondary class) which is connected to secondary links which are themselves connected to a new class. Each branch connecting two classes positioned at two different distances illustrates the distance separating these two classes.
[0302] Thus, the two main branches connected to the main class visualize two large clusters of precursor ions, and the secondary branches visualize the fragment ions of this precursor ion. It can be deduced that each secondary branch connected to a secondary class corresponds to a fragment of the precursor ion, and that the tree diagrams (other branches connected to classes positioned closer to the fragment in question) connected to this secondary class define the chemical compounds of this fragment. Consequently, the final vertical branches (the lowest branches, not connected to any lower class) illustrate the m / z ratios of each chemical compound of the fragment ions of the precursor ion in question. Thus, thanks to this dendrogram, the m / z ratios of each chemical compound of the fragment ions of the precursor ion in the class can be directly visualized.
[0303] Thus, after obtaining this dendrogram, the processing unit 120 determines the precursor ions and the fragments associated with each precursor ion by exploiting the different branches and classes of the dendrograms, the distances separating the classes as well as the m / z ratios given at the distance equal to zero (zero ordinate).
[0304] The process 1000 then includes a reconstruction step 1070 of the individual mass spectrum of each determined precursor ion and associated fragments following the determination step 1060. In practice, the processing unit 120 uses the data extracted from the classification method to reconstruct the mass spectrum of each precursor ion, including the m / z ratios grouped into the same class. Thus, for each precursor ion determined in the previous step, the processing unit The 120 retrieves each m / z ratio positioned at the minimum distance (equal to zero) and associates an intensity (i.e., an abundance value) with that ratio. This intensity can be calculated from the series of abundance values corresponding to that ratio (for example, by calculating the average abundance value of that series) or directly from the average spectrum by selecting the relevant m / z ratio on the average spectrum. The processing unit can therefore construct a peak on the mass spectrum of the precursor ion at each determined m / z ratio and assign it a corresponding intensity.
[0305] Such a reconstructed mass spectrum corresponds to a tandem mass spectrum of the determined precursor ion. Thus, with method 1000, it is possible to reconstruct the tandem mass spectrum of each determined precursor ion in the sample used using mass spectra acquired from a simple mass spectrometry device, i.e., without using a tandem mass spectrometry device.
[0306] Optionally, in the example considered, the process 1000 includes a verification of the reconstruction 1080 comprising a comparison between each individual reconstructed mass spectrum and the average mass spectrum obtained from the plurality of mass spectra (as described above). Such an action makes it possible to verify that the fragments associated with this or these precursor ion(s) exist, thereby improving the reliability and accuracy of the reconstruction step 1070.
[0307] Fig. 16 and Fig. 17 each illustrate an individual mass spectrum of one of the two precursor ions determined in the previous step.
[0308] This [Fig.16] includes a first mass spectrum 701 of a first determined precursor ion (here that corresponding to the Melittine ion), and [Fig. 17] a second mass spectrum 702 of a second determined precursor ion (here that corresponding to the ion of substance P).
[0309] Optionally, these individual mass spectra 701, 702 can be compared with the average mass spectrum shown in [Fig. 5] to verify that the combination of the individual mass spectra 701 and 702 includes all the peaks of the previously determined average mass spectrum. This is the case in this example. This can be done following the reconstruction, in reconstruction step 1070 (step 1071).
[0310] Alternatively, the determination step 1060 can use another grouping algorithm (called the first classification algorithm). Similar to the previous reconstruction step, this other classification algorithm is applied individually to each correlation map.
[0311] In this example, the processing unit 120 uses the T-SNE algorithm as its classification algorithm. Typically, the input data used by this algorithm are similar to those used by the hierarchical classification algorithm. rarchic (first classification algorithm) described above.
[0312] Figure 18 illustrates a two-dimensional map provided by the T-SNE algorithm. This map comprises a multitude of points represented on the map (a plane). Each point is associated with an m / z ratio.
[0313] This algorithm is often used when the ionized sample contains two or more precursors and / or when there is a high probability that a fragment ion is common to at least two precursor ions in the ionized mixture. The output data is a list of (x,y) coordinate pairs for each label (ion). These coordinates allow for a spatial visualization of the different highlighted groups.
[0314] Thus, the T-SNE algorithm provides a spatial visualization of the precursor and fragment ions of the ionized mixture. It is also commonly combined with another grouping algorithm, which in this case takes as input the map obtained by the t-SNE algorithm. In one example, the t-SNE algorithm is combined with a k-means algorithm. Figure 19 illustrates the use of these two algorithms successively. It can be seen that on the map obtained by the t-SNR algorithm, there are two groups, G1 and G2 (visible by the dotted lines), each comprising points on the map associated with an m / z ratio. All points in group G1 correspond to fragment ions of one of the precursor ions of the ionized mixture (here, the precursor ion of melittin), and all points in group G2 correspond to fragment ions of another precursor ion of the ionized mixture (here, the precursor ion of substance P).These two groups G1 and G2 can also be separated by a straight line DI determined by the k-means algorithm. It is then possible to associate an abundance value (i.e., intensity) with each point (a given m / z ratio). This abundance value is determined as before (in the case of the hierarchical clustering algorithm) from the mean mass spectrum or the series of abundance values associated with this m / z ratio.
[0315] The 1000 method can also combine several grouping methods to improve reconstruction accuracy. For example, the 1000 method can use a hierarchical clustering algorithm followed by a t-SNE algorithm combined with a k-means algorithm. Combining these methods can be used to verify and / or complete the reconstructed mass spectra, particularly when several precursor ions include fragment ions with a close m / z ratio.
[0316] After reconstructing the individual mass spectrum of each detected precursor ion, the process may optionally include a step 1090 of analyzing the individual mass spectrum of the determined precursor ion and associated fragment ions to characterize the chemical composition of the determined precursor ion and the The chemical composition of the fragment ions associated with the precursor ion under consideration is determined. Indeed, after the reconstruction step (and after the verification step, if present in process 1000), the chemical composition of the precursor ion and its various fragments is unknown (the reconstruction step provides a spectrum with m / z ratios and abundance values associated with each m / z ratio). Therefore, analysis step 1090 allows the chemical composition of the fragments and the precursor ion to be determined for each individual mass spectrum. This can typically be achieved using known techniques, by utilizing a database of (known) reference mass spectra and comparing each mass spectrum determined by process 1000 to the reference mass spectra.
Claims
Claims
1. A method of reconstructing a mass spectrum, said method comprising the following steps: - acquisition (1010) of a plurality of mass spectra measured on an ionized sample comprising at least one precursor ion and fragment ions, - determining (1020) a plurality of series of abundance values of the ionized sample from the plurality of mass spectra, each series of abundance values of the ionized sample being determined from the plurality of mass spectra, each series of abundance values comprising N abundance values where N is an integer greater than or equal to two; - arranging (1030) the plurality of abundance value series into pairs of abundance value series, each pair of abundance value series comprising a first abundance value series and a second abundance value series, - for each pair of abundance value series, calculating (1040) a correspondence score between the first abundance value series and the second abundance value series, so as to obtain correspondence scores for all pairs of abundance value series, the correspondence scores providing a correspondence score map between all pairs of abundance value series, - determination (1060), as a function of a mass to charge ratio, of at least one precursor ion and the fragment ions associated with each determined precursor ion, by applying a grouping method to the calculated correspondence score card, - reconstruction (1070) of an individual mass spectrum of each determined precursor ion and of each of the associated fragment ions from the correspondence score card.
2. The method of claim 2, wherein the step of determining a series of abundance values comprises determining a series of ranges of mass-to-charge ratio values for the plurality of measured mass spectra of size NxM with M an integer greater than 1, and for each mass spectrum of the plurality of mass spectra, a step of integrating abundance values belonging to a same range of mass-to-charge ratio values of the mass spectrum considered, to provide, for each range of mass-to-charge ratio values of the spectrum considered, an integrated abundance value associated with a mass-to-charge ratio, the integrated abundance values over the series of ranges of mass-to-charge ratio values forming a series of abundance values of the ionized sample.
3. A method according to any one of claims 1 to 2, wherein the matching score is based on at least one of the following: - mutual information between the first series of abundance values and the second series of abundance values, said mutual information being based on an entropy of the first series of abundance values and on an entropy of the second series of abundance values; and / or - a correlation between the first series of abundance values and the second series of abundance values.
4. Method according to claim 3, in which, for the mutual information, the entropy of the first series of abundance values is an entropy normalized by a first invariance parameter as a function of a first median value determined on this first series of abundance values and the entropy of the second series of abundance values is an entropy normalized by a second invariance parameter as a function of a second median value determined on this second series of abundance values.
5. A method according to any one of claims 3 to 4, wherein, the matching score being based on mutual information, the method comprises a step of estimating the entropy of the first series of abundance values and a step of estimating the entropy of the second series of abundance values using a k-nearest neighbors algorithm.
6. The method of claim 5, wherein the k-nearest neighbor algorithm is applied to a single neighbor.
7. A method according to any one of claims 3 to 6, wherein when at least two matching scores based on two different elements are calculated in the calculating step, several score cards are provided following the calculating step, said method further comprising, for each score card, a step of filtering (1050) said score card according to a contrast of the considered score card.
8. Method according to any one of claims 1 to 7, further comprising a step of processing (1062) the score card before applying the grouping method, said processing step using at least one of the following algorithms: - standardization algorithm, - algorithm for processing the diagonal elements of the score card, - application of a square function.
9. A method according to any one of claims 1 to 8, wherein the score map is a two-dimensional matrix, the clustering method determining the at least one precursor ion and its associated fragment ions from matching scores located outside a diagonal of the score map.
10. A method according to any one of claims 1 to 9, wherein the clustering method uses at least one unsupervised learning algorithm.
11. A method according to any one of claims 1 to 10, wherein the clustering method is based on at least one of the following algorithms: - a hierarchical ascending clustering algorithm, - a k-nearest neighbors algorithm, - a k-means algorithm, - a stochastically distributed neighbor embedding algorithm.
12. Method according to any one of claims 1 to 11, further comprising, after the reconstruction step (1070), a step of analyzing (1090) the individual mass spectrum of the determined precursor ion and the associated fragment ions to characterize the chemical composition of the determined precursor ion and the chemical composition of the fragment ions associated with the precursor ion considered.
13. Method according to any one of claims 1 to 12, in which the method comprises a step of processing (1015) the acquired mass spectra comprising at least one of the following calculations: - filtering the mass spectrum considered by a Gaussian or normal kernel, - normalization of the mass spectrum considered, - smoothing of the mass spectrum considered, - returning to the baseline.
14. A method according to any one of claims 1 to 13, wherein the method comprises a step of processing (1025) the series of values abundance calculation comprising at least one of the following calculations: - filtering the given series of abundance values by a Kalman filter; - filtering the given series of abundance values by a Fourier filter, - filtering by moving average of the given series of abundance values.
15. A method according to any one of claims 1 to 14 comprising a compressed acquisition step to reduce the number of acquisitions.
Citation Information
Patent Citations
Data Independent Acquisition of Product Ion Spectra and Reference Spectra Library Matching
US20170032948A1
