Identifying and quantifying probable principal ions in ion-mobility domain of mass spectrometry data

Abstract

Various embodiments herein disclose machine learning-based extraction and quantification of mass spectrometry data This data may be collected using high-resolution mass spectrometry (HR-MS) instruments. Also, reference to the deep learning-based identifying and quantifying peak properties of mass spectrometry data in drift time domain, as may be understood by a person skilled in the art or otherwise as described herein. Quantifying the properties of the peaks in the mass spectrometry data in the Drift time domain is provided. The properties of the peak may comprise of the m / z value associated with the peak, the retention time value associated with the peak, the drift time value associated with the peak, collision cross-section (CCS) value of peak using Mason-Schamp equation, the area under the peak, and the Signal to noise ratio (SNR) for the peak.

Description

IDENTIFYING AND QUANTIFYING PROBABLE PRINCIPAL IONS IN ION-MOBILITY DOMAIN OF MASS SPECTROMETRY DATATECHNICAL FIELD

[0001] The present subject matter, in general, relates to mass spectrometry and more particularly to data in the ion mobility domain of mass spectrometry.BACKGROUND OF THE INVENTION

[0002] Ion mobility spectrometry-mass spectrometry coupled with chromatography is an analytical technique that provides a multi-dimensional platform for analysis of pure samples and complex samples. The method’s ability to distinguish between structural isomers that are inseparable in spectral and chromatographic dimension, have been reported in literature for analyzing different analyte types such as proteins, lipids and metabolites. This has become one of the methods of choice in the analytical field due to its high sensitivity and selectivity to retrieve structural information allowing the unequivocal identification of compounds.

[0003] Figure 1 illustrates a typical hardware configuration in which the sample first passes through chromatography, where compounds are separated based on their chemical properties. The eluted compounds then enter the ion mobility spectrometer (IMS), which separates ions based on their size, shape, and charge. Finally, the ions are analyzed by the mass spectrometer (MS), which determines their mass-to-charge ratios, enabling detailed characterization of the sample components.

[0004] The standard time-dispersive IMS experiment routinely relies upon an experimental sequence where one ion packet is released and the slowest moving ion population isallowed to arrive at the detector prior to the subsequent ion packet release. With the standard drift tube ion mobility system this translates to ion utilization efficiencies as low as ~1%. This efficiency can get reduced even more if the length of the drift tube increases further.

[0005] To address the issue, multiplexing (also called as overpulsing) of the ion beam is an alternate strategy to maximize ion utilization efficiency by selectively injecting new ion packets into the separation path before the slowest ion population has arrived at the detector. The underlying ion modulation strategy is governed by a pseudo random sequence derived from a Hadamard matrix with unique autocorrelation properties. The length of each pseudo random sequence scales with the power of 2 where a 4-bit sequence corresponds to a pseudo-random sequence with 15 elements (2n-l, where n is the bit order of the sequence). While effective Hadamard demultiplexing for reduced pressure ion mobility experiments has been realized for lower bit order sequences (4-6), the increased separation capacity of SLIM and extended arrival times require the use of higher bit-orders (i.e. >= 7) along with a new class of deconvolution strategies. PRS is then used to construct the NxN Hadamard matrix; where N is the number of binary elements of the PRS. The Hadamard matrix is used to construct the inverse Hadamard matrix. The inverse Hadamard matrix is then used for convolutional measurement data to extract a data array (or vector) in which the once decomposed spectra accumulate (or demultiplex). Since the different pulses of the original multiplexed data are non-ideal, the demultiplexed data has significant residual noise. Other demultiplexing algorithms can be used and Hadamard is one such example.

[0006] A US patent US 9576778 B2 proposes a method that multiplexed data acquisition is beneficial in increasing the signal to noise (SNR) ratio.

[0007] The results from ion mobility spectrometry-mass spectrometry coupled withchromatography are compiled in a file containing mass scans and frames. Each mass scan is obtained from the mass spectrometer (103) and is characterized by a unique pair of drift time and retention time. Scans with the same retention time are present as a group in the file to form a frame. Essentially, a frame contains data of molecules that are eluted together, while a mass scan contains data of molecules that are not separable in the chromatographic and ion-mobility dimensions.

[0008] Figure lb illustrates conventional data processing methods, which typically process the data in the reverse order of how it was separated during acquisition. When processing data in the ion-mobility dimension, it is done frame by frame using the conventional data processing methods. This approach is inefficient because the data is sparse and contains significant noise, leading to longer processing times and potential inaccuracies in the results.

[0009] There are multiple ion-mobility techniques developed by multiple companies. Drift Tube and SLIM (Structures for Lossless Ion Manipulations) are based on demultiplexing of data using PRS. The widely used tool named PNNL Preprocessor (Bilbao et. al.) deal frame by frame for demultiplexing of data.

[0010] However, there is a lack of accurate extraction of good quality quantifiable peaks and their quantification in intensity vs drift time domain of mass spectrometry data. This may lead to inaccuracies in the results.SUMMARY OF THE INVENTION

[0011] Accordingly, the present invention provides a method for identifying and quantifying probable principle ions in the ion mobility domain, the method comprising the steps of receiving peaks in chromatographic dimension made from the mass spectrometry datahaving data in ion mobility, chromatography and spectral dimensions. Receiving chromatography coupled mass spectrometry data for multiple retention times and drift times may take place. Further step involves grouping of peaks obtained from chromatographic dimension for multiple drift times. Further, building Regions of interest (ROIs) in ion mobility dimensions from the grouping of the peaks may take place. Further step involves applying a specially trained first deep learning method to classify ROIs as peak or noise. Further applying a specially trained second deep learning method to mark peak boundaries may take place. Hence, quantifying the properties of the peaks representing respective probable principal ions in the ion mobility dimension may take place.

[0012] Further, another aspect of the present disclosure is providing a first deep learning method that helps leverage knowledge of a trained expert and allow the application thereof to multiple scenarios. Accordingly the method comprises the steps of fetching a set of training chromatography coupled mass spectrometry data files as the first pre-identified dataset of a first pre-identified data files mass spectrometry data for multiple drift time. Further, identifying a training regions of interest (training ROI) within each of the first pre-identified data files takes place. Further step involves randomly selecting from a predefined number of ROIs, being from a spread out range of retention time, drift time and m / z values in said first pre-identified data file. Steps above for each of the first pre-identified data file chosen may be repeated and then identifying the ROIs as containing at least one true peak, as such, based on identification by a user, otherwise labeling said ROI as noise ROI may take place.

[0013] Another aspect of the disclosure herein is a second deep learning method that comprises the steps of fetching the ROIs containing at least one true peak. Further, marking the start and end boundary of all the peaks within the confines of the ROI by a human user, a traineddevice or an artificial intelligence module takes place.

[0014] In an embodiment, the present invention reveals an innovative approach for processing ion mobility spectrometry-mass spectrometry data coupled with chromatography, collected using high-resolution mass spectrometry (HR-MS) instruments. Since the data is sparse and comprises noise, this invention employs a weighted time accumulation method to combine multiple frames and then process them. Given that for a specific compound or substance in the sample it is typically normally distributed across retention time and drift time dimensions, this method enhances the true signal through weighted time accumulation. This amplification of the true signal significantly improves the accuracy of data analysis, providing more reliable results.

[0015] For a specific mass by charge value and retention time being shared by more than one compound / element, being able to distinguish between one compound / element from the other is an aspect of this invention. The present invention helps identify small and large molecules with better accuracy using mass spectrometry.

[0016] Each compound can be characterized by a unique combination of the values of retention time, drift time, and mass to charge ratio. However, during experimental measurements, dispersion is observed around these characteristic values thus resulting in gaussian-like peaks of mass / charge ratio vs. intensity, drift time vs. intensity and retention time vs. intensity. The various data processing algorithms differ in terms of how this dispersion is accounted for. The present invention uses weighted averaging of the dispersion in the retention time domain in order to form regions of interest in the drift time domain. This dramatically reduces the noise in the intensity vs. drift time domain and in turn reduces the residual noise after demultiplexing.

[0017] In the process, a Convolutional Neural Network (ConvNet / CNN) may take in an inputimage or tensors, assign importance (learnable weights and biases) to various aspects / objects in the image and be able to differentiate one from the other. Further methods for further noise reduction and elimination have been proposed.BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS

[0018] A clear understanding of the key features of the invention summarized below may be had by reference to the appended drawings and their respective captions, which illustrate the deep learning-based extraction and quantification of mass spectrometry data, although it will be understood that such drawings depict preferred embodiments herein and, therefore, are not to be considered as limiting its scope with regard to other embodiments which the invention is capable of contemplating. Accordingly:

[0019] Figure la illustrates the schematic of a typical hardware configuration of a chromatographic separation unit hyphenated with ion mobility (IM) unit and a mass spectrometer (MS) and the direction of flow of injected material. Details of each of the hardware components and those of the connections between them have been omitted for brevity. Furthermore, the details may vary from vendor to vendor.

[0020] Figure lb illustrates the schematic representation of a conventional data processing pipeline. It illustrates the sequential consideration of various dimensions for data processing and analysis. The specific algorithms employed to process data for each dimension may vary across different software platforms.

[0021] Figure 2 illustrates the conceptual representation of data processing pipeline that shows how the different dimensions are sequentially taken into account for data analysis and processing as per an embodiment herein.

[0022] Figure 3 depicts a flow chart of a method for Identifying and quantifying probable principal ions in ion-mobility domain using deep learning as per an embodiment herein.

[0023] Figure 4 illustrates the peaks obtained from the Intensity vs. Retention time dimension are binned according to their drift time (or any representative of separation in the ion-mobility dimension). Each bin contains all peaks with the same drift time, covering the entire range of m / z and Retention Time as per an embodiment herein.

[0024] Figure 5a illustrates peaks with different drift time values but similar m / z and retention values within the error range grouped together to form a Region of Interest (ROI) in the Intensity vs. Drift Time dimension as per an embodiment herein.

[0025] Figure 5b illustrates a representative plot displaying all peaks from Figure 5a, stacked according to their drift time and retention time. Multiple peaks from the chromatographic dimension are considered to form an ROI in the ion-mobility dimension.

[0026] Figure 5c illustrates a representative plot depicting the Region of Interest (ROI) in the ion mobility dimension using the area under the chromatographic peak (shown in Figure 5b) as the z-axis, as per an embodiment herein.

[0027] Figures 6a and 6b illustrate the Regions of Interest (ROI) formed from peaks in the chromatographic dimension. Each point in these ROIs represents a specific peak from the Intensity vs. Retention Time dimension as per an embodiment herein. Figure 6a depicts the ROI for multiplexed data, while Figure 6b shows the ROI for standard mode data as per an embodiment herein.

[0028] Figure 6c presents the ROI after demultiplexing, an additional step necessary only for multiplexed data as per an embodiment herein.

[0029] Figure 7 illustrates the grouping of ROIs into two categories by the first deep learning model as per an embodiment herein. One group consists of all ROIs that contain onlyresidual noise, while the other group includes all ROIs that each contain at least one dominant peak along with residual noise as per an embodiment herein.

[0030] Figure 8 illustrates the smoothening of regions of interest (drift time vs intensity plot) that has been classified as peaks in the previous step as per embodiment herein.

[0031] Figure 9 illustrates the peak boundary marked for a ROI (shown in Figure 8) by the second deep learning model as per an embodiment herein.

[0032] Figure 10 illustrates a representative block diagram for the system for identifying a list of probable principal ions with respective quantifying properties, present in chromatography coupled mass spectrometry data as per an embodiment herein.DETAILED DESCRIPTION OF THE INVENTION

[0033] One or more systems, methods of identifying a list of probable principal ions with respective quantifying properties, present in ion mobility spectrometry-mass spectrometry coupled with chromatography data is described. In another embodiment, a non-transitory computer readable storage medium to tangibly store instructions, which when executed by a computer, cause the computer to perform steps as directed by the method herein, is described.

[0034] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well as the singular forms, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, and / or components, butdo not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0035] Unless, otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one having ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein. Further, references to deep-learning may be understood to refer to machine learning as well.

[0036] The following description is set forth for the purpose of explanation in order to provide an understanding of the invention. However, it is apparent that one skilled in the art will recognize that embodiments of the present invention, some of which are described below, may be incorporated into a number of different mass spectrometry systems and methods. Structures shown below in the diagram are illustrative of exemplary embodiments of the invention and are meant to avoid obscuring the invention.

[0037] Reference in the specification to “one embodiment”, “in one embodiment” or “an embodiment” etc. means that a particular feature, structure, characteristic, or function described in connection with the embodiment is included in at least one embodiment of the invention. The appearances of the phrase “in one embodiment” in various places in the specification are not necessarily all referring to the same embodiment.

[0038] A person skilled in the art may realize that an ion mobility dimension refers to a plot of the intensity vs drift time plots. Further, chromatographic dimension may refer to intensityvs. retention time plots. Furthermore, spectral dimension refers to intensity vs. mass / charge plots.

[0039] In describing the invention, it will be understood that a number of techniques and steps are disclosed. Each of these has individual benefits and each can also be used in conjunction with one or more, or in some cases all, of the other disclosed techniques. Accordingly, for the sake of clarity, this description will refrain from repeating every possible combination of the individual steps in an unnecessary fashion.

[0040] Various embodiments herein disclose machine learning-based extraction and quantification of ion mobility spectrometry-mass spectrometry coupled with chromatography data. This data may be collected using high-resolution mass spectrometry (HR-MS) instruments. Also, reference to the deep learning-based identifying and quantifying peak properties of mass spectrometry data in drift time domain, as may be understood by a person skilled in the art or otherwise as described herein.

[0041] Identifying the properties of the peaks in the ion mobility spectrometry-mass spectrometry coupled with chromatography data in the ion-mobility domain is provided. The properties of the peak may comprise of the m / z value associated with the peak, the retention time value associated with the peak, the drift time value associated with the peak, collision cross-section (CCS) value of peak calculated using Mason-Schamp equation, the area under the peak, and the Signal to noise ratio (SNR) for the peak.

[0042] The embodiments herein also explain the use of certain deep learning techniques for peak identification in the intensity vs drift time plots obtained from peaks formed in retention time vs intensity domain and specially designed feature vectors as input to the method to classify peak and noise in high-resolution mass spectrometers (HR-MS). Thedeep learning technique uses a feature vector (X) as an input and predicts the class (Y). The feature vector of an ROI can be the intensity vector, normalized intensity vector (values normalized between -1 and 1), a suitably smoothened, normalized intensity vector or a derivative vector of one of the above vectors, typically obtained by a finite difference method. A person skilled in the art may realize the use of alternative feature vectors with corresponding changes to suit the working of the embodiments herein. The input X can also be a tuple based on a suitable combination of the feature vectors described above. X can have the size of n x 1, n x 2 or n x 3 depending on whether the input comprises of one, two or three feature vectors, respectively, from the ones described above. The output Y is a class label such as peak or noise.

[0043] Various embodiments herein also relate to the detection of peaks and their boundaries in the intensity vs drift time plots obtained from peaks formed in retention time vs intensity domain. The same is further improvised by reduction in the residual noise left after demultiplexing, when dealing with multiplexed data received at the start of the process. Further, the leftover noise may be considerably reduced using deep learning techniques. The deep learning method applies to the ROIs formed in intensity vs drift time dimension, extracts features of the ROIs, submits them to a specially trained deep learning method, predicts peaks and their boundaries and calculates the median, mean or weighted average m / z, median, mean or weighted average retention time, median, mean or weighted average drift time, collision cross-section using Mason-Schamp equation, area under the curve for each peak and, signal to noise ratio. The deep learning tool is trained to perform the task of detecting peaks and their boundaries. The peaks are of variable heights and widths. The present invention uses the peaks formed in the intensity vs retention time domain to form regions of interest in the drift time vs intensity plot. The area under the peak or maximumheight of intensity vs retention time peaks along drift time values is used to form the ROIs. Further, the intensity is normalized in the range of -1 and 1, thereby zooming along the y-axis. A deep learning method classifies the Region of interest (ROIs) into peak and noise. A separate deep learning method then marks the peak boundaries and allowing the calculation of (i) representative drift time as the median, mean or weighted average m / z of the peak, (ii) the peak area as area under the curve between the peak boundaries and in terms of area of Intensity vs Retention Time dimension*milliseconds (intensity*minutes*milliseconds), (iii) CCS (collision cross-section), and (iv) Signal to Noise Ratio (SNR).

[0044] The deep learning methods herein may use the normalized intensity vector, normalized demultiplexed intensity vector if multiplexed data acquisition and a vector of first derivatives of the intensity vector of an ROI as inputs and provide classification of the ROI as peak or noise and further provide peak boundaries if classified as a peak. In one embodiment, the method may be primarily based on convolutional neural networks (CNN). A Convolutional Neural Network (ConvNet / CNN) is a Deep Learning method which can take in an input image, assign importance (learnable weights and biases) to various aspects / objects in the image and be able to differentiate one from the other.

[0045] The Deep learning-based method as per an embodiment herein may be trained by using real examples of ROIs that are labelled by human experts. Labeled data is not available in the public domain and needs to be generated specially by using real life HR-MS data. The ROIs are generated from the real life, ion-mobility spectrometry-mass spectrometry coupled with chromatography data available in public domain that has been generated by different LCMS instruments.

[0046] The list of peaks that represent probable elements / compounds present in the sample andthe properties are calculated that will be utilized by analysts for detailed analysis of the study. The peak formed in the Intensity vs drift time dimension uses the area under the peaks formed in the Intensity vs Retention Time dimension, therefore, the area under the peak detected in intensity vs drift time represents the ion volume of the probable principal ion. The drift time values are then used to calculate the collision cross-section (CCS) values using the Mason-Schamp equation. The CCS value can be used to annotate the elements / compounds in the sample. The presence of noisy peaks data results in noisy inaccurate analysis hence, defeating the purpose of study.. Thus, the present invention significantly reduces the noisy peaks in the data and calculates the accurate drift time thereby significantly improving the quality of the analysis and annotation of biological molecules. This in turn improves the outcome of the results from the study.

[0047] Figure 2 illustrates the conceptual representation of data processing pipeline that shows how the different dimensions are sequentially taken into account for data analysis and processing as per an embodiment herein. Centroiding of the data is required in the Intensity vs m / z dimension (201), then formation of true peaks in the Intensity vs Retention Time dimension (202) and finally dealing with the Intensity vs drift time dimension (203).

[0048] Figure 3 depicts a flow chart of a method for Identifying and quantifying probable principal ions in ion-mobility domain using deep learning as per an embodiment herein. In an embodiment, the steps may involve formation of peaks in the Intensity vs Retention time dimension (302), for multiple retention times and drift times, as an input. The ion mobility spectrometry-mass spectrometry coupled with chromatography data may consist of four dimensions comprising at least of m / z, retention time, drift time and intensity values. In one embodiment an indication may be provided regarding the input data being multiplexed data or non multiplexed data. More particularly the method may receive an input from a user or amachine relating to input data being multiplexed data. This might be done along with receiving peaks in chromatographic dimension made from the mass spectrometry data having data in ion mobility, chromatography and spectral dimensions

[0049] In an embodiment, the raw data or analytical data that is received from the high-resolution mass spectrometer is first converted into an open format such as the mzML format. In one embodiment, centroiding of data may be done if data is acquired in profile mode, else centroid data as obtained may be used. The method if worked upon non-centroided data, the accuracy of the process may reduce.

[0050] In one embodiment, further step may involve grouping of peaks (303) obtained from the intensity vs retention time dimension for multiple drift times. A bin contains all the peaks having the same drift time, covering the entire range of m / z and retention time. These bins assist in the formation of the ROIs in the Intensity vs drift time dimension.

[0051] As per an embodiment herein, a region of interest may be a range of drift time, such that the intensity values correspond to the drift time values. Further, each ROI may be a range of drift time values with their corresponding intensity values along the entire range of drift time, wherein different ROIs differ with respect to different values of median m / z and median retention time values. Peaks with different drift time values (belonging to different bins as discussed above) but similar m / z and Retention Time values within the error range are grouped together to form a ROI (304) in the Intensity vs. Drift Time dimension. This may be further understood with reference to Figure 5a, where multiple peaks (501-508) having m / z and Retention time in the error range are grouped together. The same peaks are shown in Figure 5b for a clear understanding. The area values of the Intensity vs Retention Time domain peaks are taken as the representative to form a peak in the intensity vs drift time domain as shown in Figure 5 c. ROIs formed in this process will be of constant length.

[0052] This may be further understood with reference to Figure 6a and 6b, plotting drift time for a small range of m / z value and retention time value, while the drift time range consists of discrete values from 0 to maximum drift time value recorded by instrument. The properties of the ROI formed depends on whether the data acquisition was done using multiplexing mode or standard(non-multiplexed) mode. Figure 6a shows the ROI of multiplexed data while Figure 6b shows the ROI of standard mode data.Each ROI formed is 2 domanial data with Drift time and intensity for a slice of m / z and retention time.

[0053] In one embodiment, the data is demultiplexed if it is acquired in the multiplexed mode. In some scenarios, Pseudorandom Sequence (PRS) which is used to multiplex the data may be used to demultiplex the data using Hadamard Transform. As demultiplexing using Hadamard Transform may increase the noise levels, although improving the SNR, we perform demultiplexing after weighted time accumulation to reduce the addition of noise by this step. Figure 6c shows the demultiplexed data of Figure 6a.. This may be true with other demultiplexing methods other than Hadamard transform performing the same function as Hadamard transform.

[0054] In one embodiment, the ROIs that are of size k (k being less than n) can be forced to be of the target size either by zero padding or padding with very low intensity of noise (upper limit of noise padded can be taken as 1% of maximum intensity of ROI) or by interpolating and resampling. Where the target size may be the size of the window that the deep learning model is able to process.

[0055] In one embodiment, the further steps may involve applying (305) a specially trained first deep learning model to classify ROIs as peak and noise. In another embodiment, additional class labels may be used. For example, ternary, quaternary etc. or in general, multivalent labels may be used. For example, we can choose the output variable to beternary where the 3 options are 1) Noise, 2) Peak and 3) Borderline cases (ROIs which have some gaussian characteristics but have, say low SNR).

[0056] The normalized intensity vector along with certain derived values may be submitted as features to apply a specially trained deep learning method to classify the ROI as peak or noise. Region of Interests formed in the previous steps may be given as an input to this step of ETL (Extract-Transform-Load) data pipeline. The input to deep learning model is multiplexed intensity vector, demultiplexed intensity vector and first derivative and other vectors calculated from either multiplexed intensity vector or demultiplexed intensity vector. Convolutional Neural Network may be used to classify peaks (good quality signal) and noise (artifact). The noise is discarded at this step and peaks are fed to the next convolutional neural network (segmentation) for detection of peak boundaries (start and end of peak) after smoothening the intensity vector as depicted.

[0057] In one embodiment, a step of smoothening of data that are classified as peaks may take place (example: moving average or any algorithmic smoothening) (as an optional step). This step may particularly help remove high frequency noise in the intensity vs drift time plot. This can be understood in detail from Figure 6a that contains drift time vs intensity plot without smoothening and Figure 8 containing the same plot but after removing very high frequency noise. Hence, preparing the data for accurate interpretation helps the second deep learning model to make informed decisions.

[0058] In one embodiment, as explained above, the further step may include applying (307) a specially trained deep learning model to mark peak boundaries as may be seen in Figure 8 is the smoothened ROI, and Figure 9 describes the peak boundary detection from the second deep learning model. The specific deep learning model used here is Convolutional Neural network (CNN) that takes intensity and derivative of intensity as input and outputs bit mask(a vector that contains 0 or 1) of same length as intensity vector, which in further is processed to conclude the borders of each peak(s) present in Region of Interest (ROI). These specially trained deep learning models were trained on the real life data annotated by human experts. It may be pertinent to note here that existing solutions in this domain use statistical and mathematical algorithms to detect between noise and peaks. Further, more than 50% of the classified signal as peaks by other software are not good quality signals and thus require human expert intervention to go through thousands of signals. However, the disclosure presented herein for our deep learning models requires little to no human intervention that makes the process of classification of good and bad quality signals. Lack of this functionality of detecting peak boundaries automatically requires a huge amount of human expert time to go through all peaks and mark the peak boundaries. The specially trained deep learning method thus involves collection of multiple data points using one or more human experts. Once this information is collated, the same may be used to build a module that may then be used at various locations, instead of needing a human expert every time a list of probable principal ion needs to be identified.

[0059] Further, median, mean or weighted average of drift time, mean, median and weighted average of m / z value, mean or weighted average of retention time, signal to noise ratio and area under the curve are recorded along with other statistics for each peak. Thus, quantifying (308) the properties of the peaks representing respective probable principal ion in the chromatography coupled mass spectrometry data in the Drift time dimension takes place. The process is repeated for all the Regions of Interest formed. Steps shown in 302-308 may be automated on a computing device by writing software programs in languages such as, for example Python or C++. Additionally, public domain libraries such as TensorFlow, Numpy, Scipy and Keras may be used to deploy the steps of 101-106 on acomputer. A person skilled in the art may use other public domain libraries with obvious modifications for suitability with the teachings of the embodiments herein.

[0060] Figure 5a, 5b and 5c provide examples for formation of regions of interest (ROIs) from drift time vs intensity domain peaks. HR-MS data is sparse and thus ROIs are essentially regions of non-zero intensity values. Certain tunable parameters may be used to build ROIs, which may then be subjected to ad hoc filters to eliminate instances of noise that can be identified readily.

[0061] Figure 6a and 6c of an embodiment provides an illustrative example of a multiplexed ROI formed and the demultiplexed ROI respectively. Further, Figure 7 of this embodiment illustrates example ROIs that are classified into peak or noise by using a first deep learning-based method. The classification is solely based on the shape of the ROI and not based on the value of intensity. The figure shows a total of 8 ROIs, 4 containing at least 1 species peak and the other 4 comprising pure noise. The method takes the sequences of smoothened, normalized multiplexed intensity values (normalized between -1 and 1), smoothened, normalized demultiplexed intensity values (normalized between -1 and 1) and first derivative and demultiplexed intensity vector, stacked into a 3 x k matrix, with k being a predetermined value. ROIs that are shorter than k are padded with zeros or noise (noise will be maximum of 2% of the maximum intensity) on either side. We apply a convolutional neural network with several layers but other architectures of neural networks may be applicable. In an embodiment the method may be first trained with real life data comprising of ROIs that are labeled by humans as peak or noise. Alternatively, a special purpose application or a method using computer programs may be used for the said purpose.

[0062] Figure 6a and Figure 8 explains the effect of smoothening either by using movingaverage or any dedicated algorithm to remove the high frequency noise from the peak. Smoothening the peak before giving it as an input to a specially trained second deep learning model helps to detect peak boundaries more accurately.

[0063] Figure 9 explains the identification of peak boundaries in ROIs by using a deep learning-based method. Once an ROI is classified as a peak, it may contain one or more peaks and the peak boundaries need to be identified correctly. An ROI obtained from an arbitrarily chosen real life mass scan has been used as an illustrative example. The method takes the sequences of smoothened demultiplexed intensity, normalized intensity values (normalized between -1 and 1), and its first derivative , stacked into a 2 x k matrix, with k being a predetermined value. ROIs that are shorter than k are padded with zeros on either side while ROIs that are longer than k are interpolated and resampled or forcefully broken using a sliding window approach. We apply a convolutional neural network with several layers but other architectures of neural networks may be applicable. The method is first trained with real life data comprising of ROIs that are labeled by humans as peak or noise along with accurately marked peak boundaries. In the present example, the method detects one peak within the ROI and marks the peak boundaries.

[0064] A person skilled in the art may realize upon reading the disclosure herein, the processing of data in intensity vs retention time dimensions and then forming the Region of Interests (ROIs) in intensity vs drift time dimension may have multiple advantages associated with it. Data is demultiplexed if multiplexing is performed during acquisition. Noise removal is efficient as noise removal happens in each dimension independently. Hence the problem associated with the current algorithm that tries to process data firstly in intensity vs drift time dimension along with demultiplexing if required, and then data is processed in intensity vs retention time dimension, can thus be overcome.

[0065] Figure 10 illustrates a representative block diagram for the system for identifying a list of probable principal ions with respective quantifying properties, present in chromatography coupled mass spectrometry data as per an embodiment herein.

[0066] In one embodiment, the system may comprise of a region of interest (ROI) building module 802, configured to identify a region of interest (ROI) as long ROI, normal ROI or small ROI, wherein, an ROI of size larger than a target size is a long ROI, an ROI of size smaller than the target size is a small ROI and an ROI of size equal to the target size is a normal ROI. Further, a first deep learning module 806, may be present that is configured to Check if said ROI contains at least one true peak, as that caused due to the presence of a substance in the sample and not because of an instrument noise, wherein the checking is based upon a first deep learning method trained on a first pre-identified dataset. The first deep learning module may be a tangible storage media containing the information in the form of knowledge received using a human expert, a trained device or an artificial intelligence module, for performing said action. The same may be created by mapping the information received by running the first deep method over multiple data. Further, a second deep learning module 808, may be present, configured to mark a peak boundary of the peak based on a second deep learning method, trained on a first pre-identified dataset in the database (each for multiplexed and non-multiplexed data), configured to identify the ends of a curve associated with the said peak. The second deep learning module may be a tangible storage media containing the information / program in the form of knowledge received using a human expert, a trained device or an artificial intelligence module, for performing said action. The same may be created by mapping the information received by running the first deep method over multiple data. Further, peaks quantifying module 804, configured to identify median drift time of a peak, area under the curve as the said peak, within theidentified boundary. A pre-processing and post-processing data module 803 may be provided to help with cleaning of the data, using various smoothening, normalizing methodologies. They may be configured to group the retention time vs intensity peaks where every group has peaks that have same drift time and smoothening of the ROIs that are classified as peaks from said first deep learning model.

[0067] In various embodiments herein accurate identification of a list of probable principal ions with respective quantifying properties, present in chromatography coupled mass spectrometry data, of a sample, are provided. More advantageously, isotopes that have similar M / z as well as retention time values may now be separately identified. Furthermore, due to the unique methodology of cleaning data early in the process the processing needed in the process is considerably reduced.

[0068] More advantageously, the disclosure herein allows use of deep learning modules that identify true peaks and ends thereof, thereby reducing dependency on a trained professional in the field to identify probable principal ions. Once the deep learning modules have been put to practice they may eliminate the need of a human expert at the time of identification. These modules may also be modified from time to time with more data and improvements.

[0069] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and / or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of preferred embodiments, those skilled in the art willrecognize that the embodiments herein can be practiced with modification within the spirit and scope of the embodiments as described herein and the claims provided.

Claims

We / I claim:

1. A method of identifying a list of probable principal ions with respective quantifying properties, present in chromatography coupled mass spectrometry data, comprising the steps of: a. Receiving peaks in chromatographic dimension made from the mass spectrometry data having data in ion mobility, chromatography and spectral dimensions; b. receiving a chromatography coupled mass spectrometry data for multiple retention times and drift times; grouping of peaks obtained from chromatographic dimension for multiple drift times; c. building Regions of interest(ROIs) in ion mobility dimensions from the grouping of the peaks; d. applying a specially trained first deep learning method to classify ROIs as peak or noise; e. applying a specially trained second deep learning method to mark peak boundaries; and f. quantifying the properties of the peaks representing respective probable principal ions in the ion mobility dimension.

2. The method as in claim 1, comprising the steps of: a. receiving input relating to input data being multiplexed data, along with receiving peaks in chromatographic dimension made from the mass spectrometry data having data in ion mobility, chromatography and spectral dimensions; and b. demultiplexing the ion mobility data received from built regions of interest before applying the said specially trained first deep learning method and said specially trained second deep learning method for multiplexed data.

3. The method as in claim 1, wherein the chromatography coupled mass spectrometry data is a four dimensional data comprising at least of m / z, retention time, drift time and intensity values.

4. The method as in claim 1, wherein the chromatography coupled mass spectrometry data is centroided data.

5. The method as in claim 1, wherein a region of interest is a range of drift time, such that the intensity values corresponding to the drift time values are continuously non-zero or zero values.

6. The method as in claim 1, wherein each ROI is a range of drift time values with their corresponding intensity values along the entire range of drift time, wherein different ROIs differ with respect to different values of median m / z and median retention time values.

7. The method as in claim 1, further comprising the step of Smoothening of data that are classified as peaks after applying the first deep learning method.

8. The method as in claim 1, wherein the first deep learning method comprises the steps of: fetching a set of training chromatography coupled mass spectrometry data files as the first pre-identified dataset of: a. a first pre-identified data files mass spectrometry data for multiple drift times; b. identifying a training regions of interest (training ROI) within each of the first pre-identified data file; c. randomly selecting from a predefined number of ROIs, being from a spread out range of retention time, drift time and m / z values in said first pre-identified data file; d. repeating steps 8. a to 8.c above for each of the first pre-identified data file chosen in step 8. a; and e. identifying the ROIs as containing at least one true peak, as such, based on identification by a user, otherwise labeling said ROI as noise ROI.

9. The method as in claim 1, wherein the second deep learning method comprises the steps of: a. fetching the ROIs containing at least one true peak; and b. marking the start and end boundary of all the peaks within the confines of the ROI by a human user, a trained device or an artificial intelligence module.

10. The method as in claim 1, wherein the first deep learning method comprises the steps of, fetching the training chromatography coupled mass spectrometry data file from in-house and open sources.

11. The method as in claim 3 wherein, the first pre-identified set of data files is acquired from instruments of more than one vendor of mass spectrometry instruments.

12. The method as in claim 8, further comprising the step of, repeating steps 8. a to 8.e till nearly an equal number of examples for both peak and noise labeled ROI are identified.

13. The method as in claim 1 , wherein a target size of the ROI, n, is calculated based on (i) n that is configured to be large enough so that true peaks are rarely longer than n in real life mass scan data generated by mass scan instruments, and (ii) n is small enough to minimize the complexity of the first and the second deep learning method.

14. The method as in claim 1 , whereinROIs that are of size k are forced to be of a target size either by zero padding or by interpolating and resampling.

15. The method as in claim 1, wherein the output of the first deep learning method is binary or multivalent, with the output variable having options comprising at least presence of 1.) all Noise, 2.) at least one Peak and 3.) Borderline cases, wherein ROIs which have one or more gaussian characteristics but have low signal to noise ratio.

16. The method as in claim 1 , wherein, prior to identifying the presence of at least one true peak in said ROI, normalizing the intensity values in the ROI, by dividing all the intensity values in the ROI by the maximum intensity value in that ROI, ensuring all intensity values in the ROI lie between a value minus one and a value one.

17. The method of claim 1, wherein prior to identifying the presence of at least one true peak in said ROI, preprocessing of the identified ROI, using the steps of smoothening using a moving average and producing a smoothened intensity vector.

18. The method as in claim 2, wherein for the demultiplexed data taking the derivative of the smoothened demultiplexed intensity vector; producing a derivative vector; and concatenating said smoothened demultiplexed intensity vector, multiplexed intensity vector and said derivative vector to together form an input feature vector for the deep learning method.

19. A non-transitory computer readable storage medium to tangibly store instructions, which when executed by a computer, cause the computer to: a. receiving peaks in chromatographic dimension made from the mass spectrometry data having data in ion mobility, chromatography and spectral dimensions; b. receiving a chromatography coupled mass spectrometry data for multiple retention times and drift times; c. grouping of peaks obtained from chromatographic dimension for multiple drift times; d. building Regions of interest(ROIs) in ion mobility dimensions from the grouping of the peaks; e. applying a specially trained first deep learning method to classify ROIs as peak or noise; f. applying a specially trained second deep learning method to mark peak boundaries; and g. quantifying the properties of the peaks representing respective probable principal ions in the ion mobility dimension.

20. A computer implemented method for identifying a list of probable principal ions with respective quantifying properties, present in chromatography coupled mass spectrometry data, of a sample, the method further comprising instructions which when executed by the computer cause the computer to:a. receiving peaks in chromatographic dimension made from the mass spectrometry data having data in ion mobility, chromatography and spectral dimensions; b. receiving a chromatography coupled mass spectrometry data for multiple retention times and drift times; c. grouping of peaks obtained from chromatographic dimension for multiple drift times; d. building Regions of interest(ROIs) in ion mobility dimensions from the grouping of the peaks; e. applying a specially trained first deep learning method to classify ROIs as peak or noise; f. applying a specially trained second deep learning method to mark peak boundaries; and g. quantifying the properties of the peaks representing respective probable principal ions in the ion mobility dimension.

21. A system for identifying a list of probable principal ions with respective quantifying properties, present in chromatography coupled mass spectrometry data, of a sample, comprising: a. a region of interest(ROI) building module, configured to identify a region of interest(ROI); b. a first deep learning module, configured to check if said ROI contains at least one true peak, as that caused due to the presence of a substance in the sample and not because of an instrument noise, wherein the checking is based upon a first deep learning method trained on a first pre-identified dataset each for multiplexed and non-multiplexed data; c. a second deep learning module, configured to mark a peak boundary of the peak based on a second deep learning method, trained on a first pre- identified dataset in the database each for multiplexed and non-multiplexed data, configured to identify the ends of a curve associated with the said peak; and d. peaks quantifying module, configured to identify median drift time of a peak, area under the curve as the said peak, within the identified boundary.

Images