Method and apparatus for change detection for differential analysis of data by mass spectrometry

Abstract

The present disclosure relates to a method for change detection for differential analysis of data by mass spectrometry, and is to provide a change detection method, a change detection device, and a program stored in a computer-readable recording medium to execute the change detection method for deep learning-based differential analysis of mass spectrometry data.

Description

Change detection method and device for differential analysis of data by mass spectrometry

[0001] The present disclosure relates to a change detection method and device for differential analysis of data obtained by mass spectrometry. More specifically, the present disclosure relates to a change detection method and device used for differential analysis of data obtained by mass spectrometry to identify changes in intracellular proteins.

[0002] Proteomics is a field that studies the proteome, which is a collection of proteins within a cell. Its goal is to understand how living organisms work and to identify the mechanisms of disease by studying the structure, function, expression level, interaction, and modification of proteins.

[0003] Differential proteomics is the study of changes in protein abundance or activity under different conditions or samples. Its primary purpose is to study changes in cellular proteins associated with disease onset and progression, physiological processes, and environmental changes. In particular, analyzing changes in protein abundance in response to physiological or pathological changes is a crucial process for identifying biomarkers for disease.

[0004] Quantitative proteomics is a field of proteomics that measures the amount of proteins in a sample. Differential proteomics analysis can be performed using quantitative proteomics data measured using mass spectrometry (MS). Liquid chromatography-tandem mass spectrometry (LC-MS / MS) is an analytical chemistry technique that combines the physical separation capabilities of liquid chromatography with the mass analysis capabilities of mass spectrometry. Liquid chromatography separates a mixture of multiple components, while mass spectrometry provides spectral information that can help identify each separated component.

[0005] When liquid chromatography-tandem mass spectrometry is applied, peptides degraded from proteins are separated through liquid chromatography, the peptides undergo an ionization process to become precursor ions, and the precursor ions are fragmented into smaller fragment ions. At this time, the mass-to-charge ratio (m / z) of the precursor ions and fragment ions can be measured in the form of mass spectra (MS1, MS2, respectively).

[0006] In tandem mass spectrometry, DIA (Data Independent Acquisition) is an analysis method that acquires multiple tandem mass spectra by moving a fixed window. By obtaining tandem mass spectra of fragment ions for a wide range of precursor ions, information can be extracted and quantified from rich data. Since the data obtained from the DIA analysis method includes all mass spectral data obtained from the mass spectral measurement step of the precursor ion (MS1) or the mass spectral measurement step of the fragment ion (MS2), fragment ions with a specific m / z are selected, the intensity of the signal over time is tracked to generate a graph (extracted-ion chromatogram, XIC), and after smoothing and denoising the generated graph, the peak area of ​​the graph is measured to estimate the amount of precursor ion or fragment ion (quantity estimation), and the data obtained at the peptide level are integrated (peptide / protein-level aggregation) to confirm the total amount or expression level of the protein from which the peptide is derived, and the difference of the mass spectral data can be performed through a process of increasing reliability through data normalization and comparison to maintain data consistency (cross-run normalization).

[0007] However, the existing method requires a complex series of processes (including parameter settings for specific cases) to differentiate mass spectral data, and each process relies on experience or intuition, making it prone to human error and potentially reducing the accuracy of the results and the efficiency of the process.

[0008] Accordingly, the present disclosure provides a novel method for differential analysis of mass spectrometry data, which converts mass spectrometry data into two-dimensional peak map data and then detects differences (Change Detection) using a deep learning method, bypassing the existing complex method.

[0009] The present disclosure provides a change detection method for differential analysis of data by deep learning-based mass spectrometry, a change detection device, and a program stored in a computer-readable recording medium for executing the change detection method.

[0010] Another purpose of the present disclosure is to provide a change detection method for differential analysis of mass spectrometry data that detects differences by a deep learning method without going through a complex series of conventional processes for differential analysis of mass spectrometry data.

[0011] Another purpose of the present disclosure is to quickly and accurately analyze differences by using a deep learning method used in computer vision after converting mass spectrum data into two-dimensional peak map data.

[0012] Another object of the present disclosure is to provide a change detection method for differential analysis of mass spectrometry data that reduces dependence on experience, intuition, or variable parameters in differential analysis of mass spectrometry data, thereby increasing the accuracy and reliability of the analysis.

[0013]

[0014] The purposes of the present disclosure are not limited to those mentioned above, and other purposes and advantages of the present disclosure not mentioned above can be understood through the following description and will be more clearly understood through the embodiments of the present disclosure. Furthermore, it will be readily apparent that the purposes and advantages of the present disclosure can be realized by the means and combinations thereof set forth in the claims.

[0015] The present disclosure provides a change detection method for differential analysis of data by deep learning-based mass spectrometry, a change detection device, and a program stored in a computer-readable recording medium for executing the change detection method.

[0016] A change detection method for differential analysis of deep learning-based mass spectrometry data according to one embodiment of the present disclosure may include a step of deriving a first peak map from first data obtained by mass spectrometry under a first condition, a step of deriving a second peak map from second data obtained by mass spectrometry under a second condition, and a step of extracting features of the first peak map and the second peak map using a first artificial neural network, and detecting a quantitative change between the first data and the second data based on the features of the first peak map and the second peak map.

[0017] In one embodiment, the first data and / or the second data are copies of data obtained by a plurality of mass spectrometry methods repeatedly performed under the same conditions or mass spectrometry data for the same conditions, and may include at least one of mass spectrum data (MS1) of precursor ions and mass spectrum data (MS2) of fragment ions.

[0018] In one embodiment, the first peak map and / or the second peak map may include mass spectrum data (MS1) information of precursor ions and mass spectrum data (MS2) information of fragment ions.

[0019] In one embodiment, the first peak map and / or the second peak map may be expressed by fusing the characteristics of data obtained by a plurality of mass spectrometry methods using a second artificial neural network.

[0020] In one embodiment, the second artificial neural network is a feature fusion model, and the feature fusion model may be a neural network including one or more combination blocks and a bottleneck.

[0021] In one embodiment, the first artificial neural network comprises a change detection model, the change detection model comprises a convolutional Siamese network, and the convolutional Siamese network is capable of deriving a feature vector from the first peak map and / or the second peak map.

[0022] In one embodiment, the convolutional Siamese network may include a convolutional layer for deriving the feature vector, and a pooling layer for reducing the dimension of the feature vector while maintaining important information.

[0023] In one embodiment, the convolution layer may share weights for mass analysis data for the same conditions when analyzing mass spectrum data (MS1) information of precursor ions and mass spectrum data (MS2) information of fragment ions included in the first peak map and / or the second peak map, but weights for precursor ions and fragment ions may be separated and processed separately.

[0024] In one embodiment, the first artificial neural network further includes an output model, and the output model can derive a differential analysis result by using one or more convolutional layers and fully connected layers arranged in series.

[0025] A change detection device for differential analysis of deep learning-based mass spectrometry data according to one embodiment of the present disclosure includes a memory storing one or more commands, and at least one processor executing the one or more commands stored in the memory, wherein the at least one processor derives a first peak map from first data obtained by mass spectrometry under a first condition, derives a second peak map from second data obtained by mass spectrometry under a second condition, extracts features of the first peak map and the second peak map using a first artificial neural network, and detects a quantitative change between the first data and the second data based on the features of the first peak map and the second peak map.

[0026] In one embodiment, the first data and / or the second data are copies of data obtained by a plurality of mass spectrometry methods repeatedly performed under the same conditions or mass spectrometry data for the same conditions, and may include at least one of mass spectrum data (MS1) of precursor ions and mass spectrum data (MS2) of fragment ions.

[0027] In one embodiment, the first peak map and / or the second peak map may include mass spectrum data (MS1) information of precursor ions and mass spectrum data (MS2) information of fragment ions.

[0028] In one embodiment, the first peak map and / or the second peak map may be expressed by fusing the characteristics of data obtained by a plurality of mass spectrometry methods using a second artificial neural network.

[0029] In one embodiment, the second artificial neural network is a feature fusion model, and the feature fusion model may be a neural network including one or more combination blocks and a bottleneck.

[0030] In one embodiment, the first artificial neural network comprises a change detection model, the change detection model comprises a convolutional Siamese network, and the convolutional Siamese network is capable of deriving a feature vector from the first peak map and / or the second peak map.

[0031] In one embodiment, the convolutional Siamese network may include a convolutional layer for deriving the feature vector, and a pooling layer for reducing the dimension of the feature vector while maintaining important information.

[0032] In one embodiment, the convolution layer may share weights for mass analysis data for the same conditions when analyzing mass spectrum data (MS1) information of precursor ions and mass spectrum data (MS2) information of fragment ions included in the first peak map and / or the second peak map, but weights for precursor ions and fragment ions may be separated and processed separately.

[0033] In one embodiment, the first artificial neural network further includes an output model, and the output model can derive a differential analysis result by using one or more convolutional layers and fully connected layers arranged in series.

[0034] One embodiment of the present disclosure includes a program stored on a recording medium to cause a computer to execute a method according to one embodiment of the present disclosure.

[0035] One embodiment of the present disclosure includes a computer-readable recording medium having recorded thereon a program for executing a method according to one embodiment of the present disclosure on a computer.

[0036] One embodiment of the present disclosure includes a computer-readable recording medium having recorded thereon a database used in one embodiment of the present disclosure.

[0037] According to one embodiment of the present disclosure, there is an effect of providing a change detection method for differential analysis of deep learning-based mass spectrometry data.

[0038] In addition, according to one embodiment of the present disclosure, there is an effect of providing a change detection method for differential analysis of mass spectrometry data without going through a complex series of existing processes for differential analysis of mass spectrum data.

[0039] Another purpose of the present disclosure is to have the effect of quickly and accurately analyzing differences by using a deep learning method used in computer vision after converting mass spectrum data into two-dimensional peak map data.

[0040] In addition, according to one embodiment of the present disclosure, there is an effect of increasing the accuracy and reliability of analysis by reducing the dependence on experience, intuition, or variable parameters in differential analysis of mass spectrum data.

[0041] In addition, according to one embodiment of the present disclosure, there is an effect of being able to comprehensively and simultaneously process data by analyzing a two-dimensional peak map including both MS1 and MS2 data.

[0042] Additionally, according to one embodiment of the present disclosure, there is an effect that enables more in-depth analysis of mass spectrometry data through comprehensive and simultaneous processing of data.

[0043]

[0044] In addition to the above, the specific effects of the present disclosure are described together with the specific matters for carrying out the disclosure below.

[0045] FIG. 1 is a flowchart of a change detection method for differential analysis of data using deep learning-based mass spectrometry according to one embodiment of the present disclosure.

[0046] FIG. 2 is an example diagram of a peak map according to one embodiment of the present disclosure.

[0047] FIG. 3 is an example diagram of a configuration of a neural network model according to one embodiment of the present disclosure.

[0048] FIG. 4 is an exemplary diagram showing the configuration of a feature fusion model according to one embodiment of the present disclosure.

[0049] FIG. 5 is a diagram showing the configuration of a change detection model in one embodiment of the present disclosure.

[0050] FIG. 6 is an example diagram of a Siamese Network of Multilayer Autoencoders (SNMA) using a multilayer autoencoder according to one embodiment of the present disclosure.

[0051] Figure 7 is a flowchart illustrating a general differential analysis experimental procedure based on mass spectrometry in the field of proteomics.

[0052] Figure 8 is a flowchart showing a general protein identification experimental procedure based on mass spectrometry in the field of proteomics.

[0053] Figure 9 is a general flowchart for generating a graph of the intensity of a specific mass-to-charge ratio (m / z) over time from mass spectral data obtained from a DIA analysis method.

[0054] Figure 10 is a typical graph (XIC) showing signal intensity over time by extracting information on major fragment ions after collecting the entire mass spectrometry data.

[0055] Figure 11 is an image of the mass analysis data of Figure 10 converted into a two-dimensional peak map.

[0056] Figure 12 is a two-dimensional peak map image of mass analysis spectral data of precursor ions and fragment ions associated with a specific peptide sequence.

[0057] FIG. 13 is a block diagram of a change detection device for differential analysis of data by deep learning-based mass spectrometry according to one embodiment of the present disclosure.

[0058] To clarify the technical idea of ​​the present disclosure, embodiments of the present disclosure will be described in detail with reference to the attached drawings. In describing the present disclosure, if a detailed description of a related known function or component is determined to unnecessarily obscure the gist of the present disclosure, the detailed description will be omitted. Components having substantially the same functional configuration among the drawings are given the same reference numbers and symbols as possible even if they are shown in different drawings. For convenience of explanation, devices and methods are described together when necessary. Each operation of the present disclosure does not necessarily have to be performed in the described order and may be performed in parallel, selectively, or individually.

[0059] The terms used in the embodiments of this disclosure have been selected from widely used, current terms, taking into account the functions of the present disclosure. However, these terms may vary depending on the intentions of those skilled in the art, precedents, the emergence of new technologies, etc. Furthermore, in certain cases, the applicant may arbitrarily select terms, and in such cases, their meanings will be described in detail in the description of the relevant embodiments. Therefore, the terms used in this specification should not be defined simply as names of terms, but rather based on their meanings and the overall content of the present disclosure.

[0060] Throughout this disclosure, singular expressions may include plural expressions unless the context clearly dictates otherwise. Terms such as "comprise" or "have" should be understood to indicate the presence of a feature, number, step, operation, component, part, or combination thereof, but do not preclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof. In other words, when it is said throughout this disclosure that a part "comprises" a certain component, unless specifically stated otherwise, this does not mean that other components may be included, but rather that other components may be excluded.

[0061] Expressions such as "at least one" modify the entire list of elements, and do not modify individual elements of the list. For example, "at least one of A, B, and C" and "at least one of A, B, or C" refer to only A, only B, only C, both A and B, both B and C, both A and C, all of A, B, and C, or any combination thereof. In addition, terms such as "unit" and "module" described in the present disclosure mean a unit that processes at least one function or operation, which may be implemented in hardware or software, or a combination of hardware and software.

[0062] Throughout this disclosure, when a part is said to be "connected" to another part, this includes not only cases where the parts are "directly connected," but also cases where the parts are "electrically connected" with other elements intervening. Furthermore, when a part is said to "include" a component, this does not exclude other components, but rather includes other components, unless otherwise specifically stated.

[0063] The expression "configured to" as used throughout this disclosure can be used interchangeably with, for example, "suitable for," "having the capacity to," "designed to," "adapted to," "made to," or "capable of." The term "configured to" does not necessarily mean something is "specifically designed to" in hardware. Instead, in some contexts, the expression "a system configured to" can mean that the system, together with other devices or components, is "capable of." For example, the phrase "a processor configured (or set) to perform A, B, and C" may mean a dedicated processor (e.g., an embedded processor) for performing those operations, or a generic-purpose processor (e.g., a CPU or application processor) that can perform those operations by executing one or more software programs stored in memory.

[0064] Throughout this disclosure, a peptide is a polymer in which amino acid units are linked, and has an N-terminal with an amino group and a C-terminal with a carboxyl group, which can indicate the directionality of the peptide.

[0065] The artificial intelligence-related functions according to the present disclosure can be operated via a processor and memory. The processor may include a general-purpose processor such as a CPU, an AP, or a Digital Signal Processor (DSP); a graphics-only processor such as a GPU or a Vision Processing Unit (VPU); or an artificial intelligence-only processor such as an NPU. Furthermore, the processor may be controlled to process input data according to predefined operating rules or artificial intelligence models stored in memory. Furthermore, the artificial intelligence-only processor may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0066] The predefined operation rules or AI model are characterized by being created through learning. Here, "created through learning" may mean that the basic AI model is trained using a learning algorithm using a large amount of learning data, thereby creating a predefined operation rule or AI model configured to perform a desired purpose. This learning may be performed on the device itself on which the AI ​​according to the present disclosure is executed, or may be performed through a separate server and / or system.

[0067] An artificial intelligence model may be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values, and can perform neural network operations through operations between the operation results of the previous layer and the multiple weights. The multiple weights of the multiple neural network layers may be optimized based on the learning results of the artificial intelligence model. For example, the multiple weights may be updated so that the loss value or cost value obtained from the artificial intelligence model is reduced or minimized during the learning process. The artificial neural network may include, but is not limited to, a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), a restricted boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), or deep Q-networks.

[0068]

[0069] FIG. 1 is a flowchart of a change detection method for differential analysis of data using deep learning-based mass spectrometry according to one embodiment of the present disclosure.

[0070] [Characteristic Fusion Step]

[0071] Referring to Fig. 1, first, after receiving multiple data obtained by mass spectrometry under each condition, peak maps corresponding to the data are derived, and the characteristics of the peak maps corresponding to each condition can be integrated and expressed using a neural network (feature fusion step, S110). For example, in the feature fusion step (S110), convolutional neural networks can be used to integrate and express the characteristics of the peak maps corresponding to each condition. However, the convolutional neural network is only one example and is not limited thereto, and various artificial intelligence models capable of integrating and expressing the characteristics of the peak maps can be used.

[0072] In one embodiment, a peak map can be derived using data obtained by mass spectrometry under each condition. For example, the data obtained by mass spectrometry can be measured using a data-independent acquisition (DIA) method, which repeatedly measures a certain window interval. In another embodiment, the data obtained by mass spectrometry can be measured using a data-dependent acquisition (DDA) method, which performs subsequent analysis on specific ions.

[0073] In one embodiment, data obtained by mass spectrometry can be expressed as three numbers (tuples): mass-to-charge ratio (m / z), retention time (or retention time), and mass spectrum intensity. For example, a peak map can be expressed as mass spectrum intensity of ions corresponding to a corresponding point, with retention time (or retention time) and mass-to-charge ratio (m / z) as two axes. For example, the peak map can use a method of expressing intensity as color in a two-dimensional plane, as illustrated in FIG. 2. As another example, the peak map can use a three-dimensional representation (e.g., a three-dimensional height value in the z-axis direction) corresponding to the mass spectrum intensity corresponding to a corresponding point based on the two-dimensional plane. In one embodiment, data obtained by one or more mass spectrometry methods can be multiple data obtained by repeatedly performing the method under the same conditions, and can include at least one of mass spectrum data (MS1) of precursor ions and mass spectrum data (MS2) of fragment ions.

[0074] In one embodiment, the peak map may include mass spectral data (MS1) of precursor ions and / or mass spectral data (MS2) of fragment ions. In one embodiment, the peak map may be expressed in the form of a 2D heat map by grouping data obtained by mass spectrometry according to experimental conditions, repetitions, precursor windows, and MS levels (MS1, MS2) and then smoothing the data.

[0075]

[0076] [Change Detection Step]

[0077] Change detection is the process of comparing image data (peak maps) from different points in time to identify changes, and quantify or classify the identified changes into a specific type. Various artificial intelligence models can be used to detect changes in image data. For example, artificial intelligence models for image data change detection may include convolutional neural networks, autoencoders, recurrent neural networks (RNNs), convolutional Siamese networks, Siamese networks of multilayer autoencoders, single-stream networks utilizing U-Nets, or recurrent convolutional neural networks that combine RNNs and convolutional neural networks. However, the artificial neural networks that can be adopted for change detection are not limited to those listed as examples above, and various artificial neural networks can be used.

[0078] In one embodiment, the difference between peak maps corresponding to each condition can be analyzed to detect changes in peak maps corresponding to condition changes (change detection step, S130). In one embodiment, to analyze the differences between peak maps, additional features can be extracted or dimensionality can be reduced using a neural network. For example, convolutional Siamese networks can be used to extract additional features or reduce dimensionality. However, convolutional Siamese networks are only an example and are not limited thereto, and various artificial intelligence models capable of detecting changes in peak maps can be used.

[0079]

[0080] [Output Step]

[0081] Finally, based on the features extracted under each condition, the relative changes according to the conditions can be quantified and output (result output step, S150). In one embodiment, the result output step may include additional convolutional neural networks or fully connected networks. For example, one or more convolutional layers (220) and fully connected layers (230) arranged in series can be used to quantify the changes in features and derive differential analysis results. However, this is merely an example and is not limited thereto, and various artificial intelligence models capable of quantifying and outputting the detected changes may be used.

[0082]

[0083] FIG. 3 is an example diagram of the configuration of a neural network model (10) according to one embodiment of the present disclosure.

[0084] According to one embodiment of the present disclosure, the neural network model (10) may include a feature fusion model (100), a change detection model (200), and an output model (250). Details regarding the feature fusion model (100) will be described later with reference to FIG. 4.

[0085] In one embodiment, the change detection model (200) may include a convolutional Siamese network (210) to analyze the differences between peak maps under each condition. For example, convolutional Siamese networks may be used to extract additional features or reduce dimensionality. In one embodiment, the output model (250) may include one or more convolutional layers (220) and fully connected layers (230) arranged in series. Details regarding this will be described later with reference to FIG. 5.

[0086]

[0087] FIG. 4 is an exemplary diagram showing the configuration of a feature fusion model (100) according to one embodiment of the present disclosure.

[0088] According to one embodiment of the present disclosure, the feature fusion model (100) may include a first fusion block (110), a first bottleneck (120), a second fusion block (130), and a tail (140).

[0089] In one embodiment, the first combining block (110) may include a first convolutional layer (111) for deriving feature information and a first pooling layer (112) for reducing the dimension of the feature information while maintaining important information.

[0090] In one embodiment, the data by one or more mass spectrometry methods input to the first combination block (110) may be multiple mass spectrometry data obtained by repeatedly performing under the same conditions (e.g., Condition A or Condition B) or replicas (Rep 1, Rep 2, …, Rep N) of mass spectrometry data for the same conditions.

[0091] In one embodiment, the first combining block (110) passes the feature values ​​() through a first convolution layer for one or more input mass analysis data. , , ..., ) can be derived, and the reduced value and feature value that passed through the first pooling layer (112) are connected to the result value ( , , ..., ) can be derived.

[0092] In one embodiment, the first bottleneck (120) may be used to increase computational efficiency by reducing the depth, parameters, or computational complexity for the first result value derived from the first combination block (110) using the convolutional layer (121) of the first bottleneck.

[0093] In one embodiment, the first bottleneck (120) takes the result value derived from the first combination block (110) as an input value and outputs the result value ( , , ..., ) can be derived.

[0094] In one embodiment, the second combining block (130) may include a second convolution layer (131) for deriving feature information and a second pooling layer (132) for reducing the dimension of the feature information while maintaining important information.

[0095] In one embodiment, the second combination block (130) passes the feature values ​​(by one or more mass spectrometry data) through a second convolution layer (131) to the input , , ..., ) can be derived, and the reduced value that has passed through the second pooling layer (132) with the feature value as the input value, the feature value that has passed through the second convolution layer (131), and the feature value that has passed through the first convolution layer (111) ( , , ..., ) is directly connected (skip connection) without going through the first bottleneck (120) and the second convolution layer (131) to obtain the result value ( , , ..., ) can be derived.

[0096] In one embodiment, the tail (140) can derive a two-dimensional peak map that fuses the characteristics of one or more data obtained by mass spectrometry corresponding to each condition from the result values ​​derived from the second combining block (130) using the convolution layer and the pooling layer within the tail (140). In one embodiment, the two-dimensional peak map derived from the tail (140) can be used as an input to the change detection model (200).

[0097] In this disclosure, a configuration for fusing the characteristics of a peak map is described by exemplifying the neural network structure of FIG. 4, but it is not limited thereto, and it is easily understood by those skilled in the art that other neural network structures for fusing the characteristics of multiple data can be used.

[0098]

[0099] FIG. 5 is a diagram showing the configuration of a change detection model (200) and an output model (250) in one embodiment of the present disclosure.

[0100] In one embodiment, the change detection model (200) may be used to detect changes by deriving features of two-dimensional peak maps corresponding to each condition. In one embodiment, the change detection model (200) may be implemented using a convolutional Siamese network (210). The convolutional Siamese network (210) may derive feature vectors using convolutional neural networks for each of two inputs. In one embodiment, the convolutional Siamese network (210) may include a first convolutional network (211) and a second convolutional network (212), and the weights of the first convolutional network (211) and the second convolutional network (212) may be shared. In one embodiment, the first convolutional network (211) and the second convolutional network (212) may include a convolution layer for deriving feature vectors, and a pooling layer for reducing the dimensionality of the feature vectors while maintaining important information.

[0101] In one embodiment, when analyzing the mass spectrum data (MS1) information of the precursor ion and the mass spectrum data (MS2) information of the fragment ion included in the two-dimensional peak map, the weights of the first convolutional network (211) and the second convolutional network (212) for each piece of information are shared, but the weights for the precursor ion and the fragment ion can be set separately.

[0102] In one embodiment, the output model (250) may derive differential analysis results for feature vector data derived from the convolutional Siamese network (210). In one embodiment, the output model (250) may include one or more convolutional layers (220) and fully connected layers (230) arranged in series.

[0103] In another embodiment, the convolutional Siamese network (210) may be replaced with a Siamese network (300) using a multilayer autoencoder, a single-stream network using U-Net (not shown), or a recurrent convolutional neural network (not shown) that combines an RNN and a convolutional neural network. For example, the single-stream network using U-Net may include an encoder composed of convolutional layers and pooling layers to extract features from an input image, and a decoder to generate an output using the convolutional layers. The recurrent convolutional neural network may include a convolutional layer to extract features from an input image, and an RNN layer to learn sequential features of time-series data. Details regarding the Siamese network (300) using a multilayer autoencoder will be described later with reference to FIG. 6.

[0104]

[0105] FIG. 6 is an example diagram of a Siamese Network of Multilayer Autoencoders (SNMA) (300) using a multilayer autoencoder according to one embodiment of the present disclosure.

[0106] According to one embodiment of the present disclosure, a Siamese network (300) using a multilayer autoencoder may be a form in which two multilayer autoencoders are connected in a Siamese network structure. A Siamese network is a structure in which two separate neural networks share the same weights and are configured in parallel, and after input data is converted into each feature vector through the same neural network configured in parallel, the similarity between the two vectors can be calculated. An autoencoder (310) is a neural network that encodes (compresses) input data and then restores it. When an autoencoder (310) is configured with multiple layers, it is called a multilayer autoencoder. A multilayer autoencoder learns important features of input data to generate a compressed code, and in the process, it can perform tasks such as dimensionality reduction of data, noise removal, and feature extraction.

[0107] In one embodiment, a Siamese network (300) using a multilayer auto-encoder may include a first multilayer auto-encoder (350) and a second multilayer auto-encoder (360). In one embodiment, the first multilayer auto-encoder (350) and the second multilayer auto-encoder (360) include auto-encoders (310) having the same number of layers, and can compare latent variables, which are features extracted from two auto-encoders (310) of the same layer, and use the results for change detection.

[0108] In one embodiment, the difference (Diff) between the output values ​​from the autoencoder (310) of the same layer of the first multilayer autoencoder (350) and the second multilayer autoencoder (360) can be calculated. In one embodiment, the difference value (Diff) calculated in each layer can be transmitted to the MLP decoder (320) so that the difference between the two input data can be analyzed from the difference value. The MLP decoder (320) can correspond to the output model (250).

[0109]

[0110] Figure 7 is a flowchart illustrating a general differential analysis experimental procedure based on mass spectrometry in the field of proteomics.

[0111] Figure 7(a) shows the experimental procedure for differential analysis of data by mass spectrometry using an isotope labeling method, and Figure 7(b) shows the experimental procedure for differential analysis without using a label (label free).

[0112] Referring to Figure 7(a), a specific protein (or cell) is divided into a control group and an experimental group (sample), labeled with a heavy isotope and a light isotope, the labeled control group and the experimental group are mixed and digested, and liquid chromatography-tandem mass spectrometry is applied to the digested protein to obtain the mass-to-charge ratio (m / z) of the precursor ion and fragment ion, and the relative amount of the protein can be compared using the labeled isotope as a marker at each m / z.

[0113] Referring to Fig. 7(b), a specific protein (or cell) is divided into a control group and an experimental group (sample), each decomposed, liquid chromatography-tandem mass spectrometry is applied to the decomposed protein, and the peak intensity obtained as a result is measured or the number of spectra obtained for a specific protein is counted (spectral count) to compare the relative amount of the protein.

[0114]

[0115] Figure 8 is a flowchart showing a general protein identification experimental procedure based on mass spectrometry in the field of proteomics.

[0116] Protein identification is a method that can be used when you want to identify a protein separated from a mixed sample or when you want to identify a newly discovered protein.

[0117] Referring to Fig. 8, after decomposing several proteins into peptides, the peptides are ionized to generate multiple precursor ions, and mass analysis is performed to obtain mass spectral data for the multiple precursor ions. The mass spectral data is derived as a peak for m / z, and a peak requiring additional analysis is selected, a specific precursor ion corresponding to the selected peak value is fragmented to generate multiple fragment ions, and mass analysis is performed to obtain mass spectral data for the multiple fragment ions. The mass spectral data for the multiple fragment ions can be compared with a protein sequence database to evaluate the peptide sequence and to identify the desired protein through a process of evaluating how closely it matches the protein sequence (peptide scoring).

[0118]

[0119] Figure 9 is a general flowchart for generating a graph of the intensity of a specific mass-to-charge ratio (m / z) over time from mass spectral data obtained from a DIA analysis method.

[0120] Data Independent Acquisition (DIA) is an analytical method that acquires multiple tandem mass spectra by moving a fixed window. Data obtained through DIA analysis not only contain mass spectra of multiple precursor ions, but also the mass spectra of fragment ions for each precursor ion.

[0121] Figure 9(a) is a graph of multiple precursor ion mass spectra (MS1) measured using the DIA (Data Independent Acquisition) analysis method. As shown in the graph of Figure 9(a), data is collected according to time (X-axis) and m / z (Y-axis), and the change in m / z measured over time is displayed by moving a fixed window divided into sections of m / z values.

[0122] Figure 9(b) is a graph listing mass spectrum information (MS2) of fragment ions included in a specific m / z interval over time. Looking at the graph of Figure 9(b), the data includes not only time (X-axis) and m / z (Y-axis), but also intensity (Z-axis) for m / z (Y-axis).

[0123] Figure 9(c) is a graph of the MS2 signal of specific fragment ions. By selecting specific m / z values ​​(Y values) for the graph of Figure 9(b) and connecting them over time, an MS2 signal graph of a specific fragment ion can be derived (XIC). Looking at the graph of Figure 9(c), it includes the intensity (Z-axis) of a specific fragment ion versus time (X-axis).

[0124]

[0125] Figure 10 is a typical graph (XIC) showing signal intensity over time by extracting information on major fragment ions after collecting the entire mass spectrometry data.

[0126] An extracted ion chromatogram (XIC) is a graph of the signal intensity of ions with a specific mass-to-charge ratio of interest extracted from the entire mass spectrometry data set over time. However, an extracted ion chromatogram extracted from the entire mass spectrometry data set is difficult to interpret on its own and requires smoothing and noise removal steps.

[0127] Figure 10(a) is a graph of signal intensities for two major fragment ions extracted over time from the entire mass spectrometry data, and Figure 10(b) is a graph after smoothing and noise removal. In each graph, the X-axis represents dRT (delta retention time), which represents the difference in time it takes for proteins, peptides, etc. to pass through the column in the chromatograph, and the Y-axis represents the intensity of the ion signal.

[0128]

[0129] Figure 11 is an image of the data obtained by mass spectrometry in Figure 10 converted into a two-dimensional peak map.

[0130] Referring to Fig. 11, although it is difficult to quantify the amount of protein through each image, it is not difficult to detect differences through comparison between images. Therefore, the present disclosure provides a novel differential analysis method of mass spectrometry data that bypasses the existing complex method, converts mass spectrometry data into a two-dimensional peak map, and then detects differences in the images using a deep learning method. In one embodiment, since the two-dimensional peak map can be analyzed in a neural network in the same manner as a two-dimensional image, it can be easily analyzed using the deep learning method used for image comparison without going through the complex quantification and normalization process of the existing method. For example, as discussed above, a convolutional Siamese network used for image comparison can be used to detect changes in peak maps obtained under two conditions.

[0131] In one embodiment, by storing a peak map under existing conditions (Con A) and then deriving a peak map under new conditions (Con B), and then analyzing the two peak maps using the method according to the present disclosure, quantitative changes in peptides due to changes in conditions can be easily detected. Compared to existing methods that require quantification and normalization processes, the method according to the present disclosure has the effect of detecting quantitative changes more easily and accurately.

[0132]

[0133] Figure 12 is a two-dimensional peak map image of mass analysis spectral data of precursor ions and fragment ions associated with a specific peptide sequence.

[0134] Referring to Fig. 12, the images in the first row are two-dimensional peak map images for mass analysis spectrum data of precursor ions, and the images in the second row and thereafter represent two-dimensional peak map images for mass analysis spectrum data of fragment ions. ConA Rep1, …, ConB Rep3 written at the top of the images represent multiple mass analysis data obtained by repeatedly performing under specific conditions (or for samples under specific conditions) (e.g., Condition A or Condition B) or replicas (Rep 1, Rep 2, Rep 3) of mass analysis data for the same conditions. For example, ConA Rep1 may represent the first experimental data or the first replica performed under conditions A (or for sample A).

[0135]

[0136] FIG. 13 is a block diagram of a change detection device for differential analysis of data by deep learning-based mass spectrometry according to one embodiment of the present disclosure.

[0137] Referring to FIG. 13, a change detection device (400) for differential analysis of mass spectrometry data based on deep learning may include a transceiver (410), a memory (420), and a processor (430). In addition, according to one embodiment of the present disclosure, a function for differentially analyzing mass spectrometry data based on deep learning may be implemented as software modules, and these software modules may be stored in the memory (420) and executed by the processor (430). In one embodiment, the processor (430) may include a general-purpose processor such as a CPU, an AP, a DSP (Digital Signal Processor), a graphics-only processor such as a GPU, a VPU (Vision Processing Unit), or an artificial intelligence-only processor such as an NPU.

[0138] However, not all of the components illustrated in FIG. 13 are essential components of the change detection device (400). The change detection device (400) may be implemented with more components than those illustrated in FIG. 13, or with fewer components than those illustrated in FIG. 13. In addition, the transceiver (410), the memory (420), and the processor (430) may be implemented in the form of a single chip. In addition, in one embodiment, the change detection device (400) may include at least one transceiver (410), one memory (420), and one processor (430).

[0139] In one embodiment, the transceiver (410) can communicate with a terminal, server, or other electronic device connected wired or wirelessly to the change detection device (400).

[0140] Various types of data, such as programs and files, such as applications, can be installed and stored in the memory (420). The processor (430) can access and use data stored in the memory (420), or store new data in the memory (420).

[0141] The processor (430) controls the overall operation of the change detection device (400) and may include at least one processor, such as a CPU or GPU. The processor (430) may control other components included in the change detection device (400) to perform operations for operating the change detection device (400). For example, the processor (430) may execute a program stored in the memory (420), read a stored file, or store a new file. In one embodiment, the processor (430) may perform operations for operating the change detection device (400) by executing a program stored in the memory (420).

[0142]

[0143] An embodiment of the present disclosure may also be implemented in the form of a recording medium containing computer-executable instructions, such as program modules, executed by a computer. Computer-readable media may be any available media that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. Furthermore, computer-readable media may include both computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Communication media typically contains computer-readable instructions, data structures, or program modules, and includes any information delivery media.

[0144] The above description of the present disclosure is provided for illustrative purposes only, and those skilled in the art will readily appreciate that the present disclosure can be readily modified into other specific forms without altering the technical spirit or essential characteristics of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, components described as being single may be implemented in a distributed manner, and similarly, components described as being distributed may be implemented in a combined manner.

[0145] The scope of the present disclosure is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present disclosure.

Claims

1. In a change detection method for differential analysis of data by mass spectrometry performed by a computer device, A step of deriving a first peak map from first data obtained by mass spectrometry under a first condition; A step of deriving a second peak map from second data obtained by mass spectrometry under a second condition; and A step of extracting features of the first peak map and the second peak map using a first artificial neural network, and detecting quantitative changes between the first data and the second data based on the features of the first peak map and the second peak map, A change detection method for differential analysis of mass spectrometry data based on deep learning.

2. In paragraph 1, The first data and / or the second data, Data obtained by multiple mass spectrometry methods performed repeatedly under the same conditions or copies of mass spectrometry data for the same conditions. Containing at least one of mass spectrum data of precursor ions (MS1) and mass spectrum data of fragment ions (MS2), A change detection method for differential analysis of mass spectrometry data based on deep learning.

3. In paragraph 1, The first peak map and / or the second peak map Containing mass spectrum data (MS1) information of precursor ions and mass spectrum data (MS2) information of fragment ions, A change detection method for differential analysis of mass spectrometry data based on deep learning.

4. In paragraph 2, The first peak map and / or the second peak map This is an expression of the characteristics of data obtained through multiple mass spectrometry methods by fusing them using a second artificial neural network. A change detection method for differential analysis of mass spectrometry data based on deep learning.

5. In paragraph 4, The above second artificial neural network is a feature fusion model, The above feature fusion model is a neural network including one or more combination blocks and a bottleneck. A change detection method for differential analysis of mass spectrometry data based on deep learning.

6. In paragraph 1, The first artificial neural network includes a change detection model, the change detection model includes a convolutional Siamese network, and the convolutional Siamese network derives a feature vector from the first peak map and / or the second peak map. A change detection method for differential analysis of mass spectrometry data based on deep learning.

7. In paragraph 6, The above convolutional Siamese network is, A convolutional layer for deriving the above feature vector, and Includes a pooling layer to reduce the dimension of the feature vector while maintaining important information. A change detection method for differential analysis of mass spectrometry data based on deep learning.

8. In paragraph 7, The above convolutional layer is, In analyzing the mass spectrum data (MS1) information of the precursor ion and the mass spectrum data (MS2) information of the fragment ion included in the first peak map and / or the second peak map, For mass analysis data for the same conditions, weights are shared, but the weights for precursor ions and fragment ions are separated and can be processed separately. A change detection method for differential analysis of mass spectrometry data based on deep learning.

9. In paragraph 6, The above first artificial neural network further includes an output model, The above output model derives the differential analysis result by using one or more convolutional layers and fully connected layers arranged in series. A change detection method for differential analysis of mass spectrometry data based on deep learning.

10. Memory that stores one or more instructions; and At least one processor for executing one or more instructions stored in the memory, At least one processor, A first peak map is derived from the first data obtained by mass spectrometry under the first condition, A second peak map is derived from the second data obtained by mass spectrometry under the second condition, Extracting features of the first peak map and the second peak map using a first artificial neural network, and detecting quantitative changes between the first data and the second data based on the features of the first peak map and the second peak map. A change detection device for differential analysis of deep learning-based mass spectrometry data.

11. In paragraph 10, The first data and / or the second data, Data obtained by multiple mass spectrometry methods performed repeatedly under the same conditions or copies of mass spectrometry data for the same conditions. Containing at least one of mass spectrum data of precursor ions (MS1) and mass spectrum data of fragment ions (MS2), A change detection device for differential analysis of deep learning-based mass spectrometry data.

12. In paragraph 10, The first peak map and / or the second peak map Containing mass spectrum data (MS1) information of precursor ions and mass spectrum data (MS2) information of fragment ions, A change detection device for differential analysis of deep learning-based mass spectrometry data.

13. In paragraph 11, The first peak map and / or the second peak map This is an expression of the characteristics of data obtained through multiple mass spectrometry methods by fusing them using a second artificial neural network. A change detection device for differential analysis of deep learning-based mass spectrometry data.

14. In paragraph 13, The above second artificial neural network is a feature fusion model, The above feature fusion model is a neural network including one or more combination blocks and a bottleneck. A change detection device for differential analysis of deep learning-based mass spectrometry data.

15. In paragraph 10, The first artificial neural network includes a change detection model, the change detection model includes a convolutional Siamese network, and the convolutional Siamese network derives a feature vector from the first peak map and / or the second peak map. A change detection device for differential analysis of deep learning-based mass spectrometry data.

16. In paragraph 15, The above convolutional Siamese network is, A convolutional layer for deriving the above feature vector, and Includes a pooling layer to reduce the dimension of the feature vector while maintaining important information. A change detection device for differential analysis of deep learning-based mass spectrometry data.

17. In paragraph 16, The above convolutional layer is, In analyzing the mass spectrum data (MS1) information of the precursor ion and the mass spectrum data (MS2) information of the fragment ion included in the first peak map and / or the second peak map, For mass analysis data for the same conditions, weights are shared, but the weights for precursor ions and fragment ions are separated and can be processed separately. A change detection device for differential analysis of deep learning-based mass spectrometry data.

18. In paragraph 15, The above first artificial neural network further includes an output model, The above output model derives the differential analysis result by using one or more convolutional layers and fully connected layers arranged in series. A change detection device for differential analysis of deep learning-based mass spectrometry data.

19. A program stored in a computer-readable recording medium that causes a computer to execute any one of the methods of paragraphs 1 to 9.

Images