Method of storing and querying measurement data

By transforming N-dimensional measurement data into 1-dimensional coordinates using domain-specific resolutions and tolerances, and building a prefix tree for storage, the method addresses storage inefficiencies and enables efficient random access, reducing resource needs and improving analysis speeds.

WO2026234231A1PCT designated stage Publication Date: 2026-11-12BRUKER DALTONIK GMBH & CO KG +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2026/063407
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-09
Filing Date
2026-05-08
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Existing methods for storing and querying measurement data from analytical systems like mass spectrometers and imaging systems face challenges due to large data sizes, inefficient storage, and difficulty in random access, especially with data formats that do not consider the inherent multi-dimensional structure of the data, leading to increased resource requirements and slow analysis speeds.

Method used

A method that transforms N-dimensional measurement data into 1-dimensional coordinates using a bijection parameterized by domain resolutions and tolerances, builds a prefix tree from these coordinates, and serializes it for storage, enabling efficient storage and random access across physical measurement domains.

Benefits of technology

This approach reduces storage requirements and enables significant speed-ups in data processing by preserving the multi-dimensional structure of the data, allowing for efficient pruning and traversal during querying, thus improving analysis speeds.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2026063407_12112026_PF_FP_ABST
    Figure EP2026063407_12112026_PF_FP_ABST
Patent Text Reader

Abstract

A method of storing measurement data of a measurement system in a database on a non-transitory tangible computer-readable storage medium comprises i) obtaining M unique N- dimensional coordinates being associated with the measurement data, ii) transforming the M unique N-dimensional coordinates with a bijection to 1-dimensional coordinates, iii) building a prefix tree derived from the M unique 1-dimensional coordinates, iv) serializing the prefix tree, and v) writing the serialized prefix tree to the non-transitory tangible computer-readable storage medium. A method of querying a database for a subset of measurement data comprises i)receiving at least one query defining at least one N- dimensional constraint, ii) transforming the N-dimensional constraint, using the same bijection, into at least one corresponding 1-dimensional constraint, iii) querying the prefix tree with the at least one 1-dimensional constraint, iv) retrieving at least one subset of 1-dimensional coordinates from the prefix tree, v) and converting the at least one retrieved subset of 1-dimensional coordinates back into corresponding N-dimensional coordinates being associated with the measurement data using an inverse of the bijection.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] TITLE

[0002] METHOD OF STORING AND QUERYING MEASUREMENT DATA

[0003] TECHNICAL FIELD

[0004] The present invention relates to a method of storing measurement data of a measurement system in a database according to claim 1, to a method of querying a database comprising measurement data of a measurement system according to claim 15, to a computer program product causing a processor to perform these methods according to claim 25, and to a measurement system comprising a processor configured to perform these methods according to claim 26.

[0005] PRIOR ART

[0006] Analytical measurement systems such as mass spectrometers (MS) acquire billions of measurement data per sample, requiring much disk space for storage. Currently, storing measurement data of common liquid or gas chromatography (LC or GC) separated MS samples typically requires hundreds of GB per measurement system per day. In other applications such as imaging, e.g. Mass Spectrometry Imaging (MSI), storing measurement data from just a single sample can already require hundreds of GBs of disk space. As a result, use-cases such as clinical studies that require large sample cohorts with many replicates can yield tera- or even petabytes of unprocessed data. Storing such data efficiently is non-trivial and requires massive resources and even small gains in storage efficiency have the potential to dramatically reduce these resources. Equally, processing these measurement data in its entirety becomes challenging due to RAM constraints, especially when even single samples are already very large, for instance MSI data. The solution generally is to require processing in smaller batches and multiple (modular) steps which frequently duplicates (parts of) the data, yet again requiring more disk space, and at the same time worsening analysis speeds. Fast-paced technological advancements for such measurements systems introduce both new techniques as well as more complex combinations of techniques, resulting in even larger data sets and thereby exacerbating the problem.Different data formats can be used to store data from various measurement systems. For instance, TIMS-TOF (Trapped Ion Mobility Separation, Time-Of-Flight) mass spectrometers use TDF (TIMS-TOF data format). This format is optimized for serialization during acquisition, e.g. the data is ordered in the same way that analytical separation occurs. For LC-TIMS-TOF MS this is first in the retention time (RT) which are stored by thousands of frame indices, second in the ion mobility (IM) domain which are stored by hundreds of scan indices and last in the m / z (mass over charge) domain which are stored as hundreds of thousands time-of-f light (TOF) indices. In the case of imaging data, the first order are typically hundreds or thousands of X coordinates of the rasterized imaging sample and the second order are the hundreds or thousands of Y coordinates of the rasterized imaging sample. In the case of MSI, the third domain can be e.g. the IM domain stored as scan indices and fourth can be the m / z domain stored as TOF indices. While this format is sensible from an acquisition perspective, it is not ideal from a storage or analysis perspective, as it imposes a serial order that is determined by acquisition rather than by the structure of the measurement data itself.

[0007] By storing the data sequentially as exemplified above, only the density and correlations in the last dimension are fully considered. However, spectrometric and imaging measurement data is inherently structured in multiple dimensions, resulting in repetitive or at least comparable values in all these dimensions. For LC-TIMS-TOF data, this is exemplified by extracted ion chromatograms (XICs) or mobilograms (XIMs) of individual ion species that follow various distributions rather than being a list of sparse values. Such data representation shows clear gaussian-like distributions illustrative of the continuous analytical separation. For imaging, data form clear structures in the X and Y dimensions of the rasterized image as these naturally correlate with the tissue that is being captured within the image. As these structures, i.e. changes in density, are distributed over multiple dimensions which are naturally stored sequentially, the resulting serial representation fails to preserve multi-dimensional, physically meaningful neighbourhoods and the entropy is not optimized.

[0008] Furthermore, the sparsity of the data means that not all domains are represented by fixed length arrays. For instance, in TDF the first domains are all dense, whereas the last dimension, i.e. the TOF domain, is sparse, i.e. only non-zero values are present. This means that while querying the data this variability in array sizes must be considered, making it difficult to perform efficient random access. For instance, searching in the TOF domain ismost easily done with a binary search on a per scan basis. This is reasonable when data is also accessed sequentially such as for MS2 spectra as acquired with data dependent acquisition (DDA) but is prohibitive for e.g. XIC / XIM extraction or for image retrieval for MSI data. In such cases thousands of frames need to be accessed, even though only a tiny fraction, typically less than 0.001% as TOF ranges are typically selected at ppm (parts per million) precision, of each individual frame needs to be accessed. While techniques such as caching and good heuristics can ameliorate this to some extent, the lack of a data representation that reflects resolution-dependent structure across multiple dimensions prevents efficient pruning of irrelevant data regions and true random access has the potential to speed up naive approaches by up to 5 orders of magnitude.

[0009] SUMMARY OF THE INVENTION

[0010] It is an object of the present invention to overcome the drawbacks of the prior art. In particular, in a first aspect, it is an object to provide an improved method of storing measurement data, especially a method that reduces the storage requirements for measurement data. In a second aspect, it is an object to provide an improved method of querying measurement data, in particular enabling random access across one or more physical measurement domains.

[0011] The first object is achieved with a computer-implemented method as claimed in claim 1. That is, in a first aspect a computer-implemented method of storing measurement data of a measurement system in a database on a non-transitory tangible computer-readable storage medium is provided. The method comprises the steps of i) obtaining M unique N-dimensional coordinates being associated with the measurement data, ii) transforming the M unique / V-dimensional coordinates with a bijection to 1 -dimensional coordinates, whereby M unique 1 -dimensional coordinates are obtained, iii) building a prefix tree from the M unique 1 -dimensional coordinates, iv) serializing the prefix tree, and v) writing the serialized prefix tree to the non-transitory tangible computer-readable storage medium.

[0012] That is, in the first aspect the present invention provides a method of storing measurement data of a measurement system to a non-transitory tangible computer-readable storage medium. In particular, the method according to this first aspect preferably enables the provision of a database that is then stored, i.e. written to the non-transitory tangible computer-readable storage medium. In particular, the prefix tree preferably represents at least a structural part of the database, which can then be stored on the non-transitorytangible computer-readable storage medium in a compact and efficiently accessible form.

[0013] In a further aspect, the present invention also provides a method of querying such a prefix tree or database, preferably with random access across one or more physical measurement domains, respectively, as will be explained in greater detail further below.

[0014] The non-transitory tangible computer-readable storage medium preferably is a permanent storage medium such as a disk, a cloud, etc.

[0015] The measurement data preferably is spectrometric measurement data and / or spatial measurement data. Thus, explanations made for "measurement data" preferably likewise apply to "spectrometric measurement data" and "spatial measurement data", and vice versa.

[0016] Each N-dimensional coordinate preferably comprises N spectrometric indices and / or N spatial indices acquired from a spectrometric component or from a spatial component of the measurement system, respectively.

[0017] The measurement data is preferably associated with one or more physical measurement domains that each have an inherent resolution and / or tolerance. Such domains include, for example, various biochemical and / or physical properties such as hydrophobicity, collisional cross section (CCS), mass over charge ratio (m / z), and / or spatial coordinates that each have an inherent resolution and / or tolerance such as X and Y coordinates in a rasterized image, etc. It should be noted that for reasons of simplicity said "physical measurement domains" are sometimes just referred to as "domains".

[0018] In particular, each spectrometric component and / or spatial component of the measurement system preferably performs a continuous discrimination of the measurement data, in particular the spectrometric and / or spatial measurement data, along a respective physical measurement domain having an associated resolution and / or tolerance.

[0019] The measurement data preferably defines a sparse occupancy of an N-dimensional measurement space while comprising localized structures and / or clusters whose extent along each physical measurement domain is defined by the resolution and / or tolerance associated with the respective physical measurement domain. Preferably, at least one physical measurement domain performs time of flight separation of ions, although otherphysical measurement domains are likewise conceivable, see further below.

[0020] For the purposes of the present application, “spectrometric measurement data” preferably refers to measurement data acquired by at least one spectrometric component of a measurement system, wherein the spectrometric measurement data preferably comprises values associated with spectrometric indices along at least one physical measurement domain. Said physical measurement domain preferably corresponds to an analytical property that is discriminated by the spectrometric component, for example mass-to-charge ratio, time-of-flight, ion mobility and / or collisional cross section, wavelength, frequency and / or energy. The spectrometric indices preferably are continuous or discretized in accordance with an associated resolution and / or tolerance.

[0021] For the purposes of the present application, “spatial measurement data” preferably refers to measurement data associated with positions in a spatial coordinate system, wherein each position is defined by at least two spatial indices (for example X and Y coordinates of a rasterized sample, optionally a Z coordinate), and wherein the spatial measurement data preferably comprises at least one measured value that is associated with each respective position, optionally together with one or more spectrometric indices.

[0022] For the purposes of the present application, a “continuous measurement system” preferably refers to a measurement system configured to acquire measurement data over a progression of at least one acquisition parameter to produce a plurality of measurement events for successive parameter values. The acquisition parameter preferably corresponds to at least one physical measurement domain, for example retention time, ion mobility, time-of-flight, and / or spatial position, and wherein the measurement data is thereby acquired as a sequence of measurement events that is continuous in the sense of the acquisition process, optionally while being discretized into indices in accordance with an associated resolution and / or tolerance.

[0023] For the purposes of the present application, a “resolution-defined neighbourhood” and a “tolerance-defined neighbourhood”, respectively, preferably refers to a neighbourhood in the N-dimensional measurement space defined by per-domain bounds derived from the associated resolution and tolerance, respectively, of each physical measurement domain, such that two N-dimensional coordinates preferably are within the same neighbourhood if and only if they differ by at most the respective per-domain bound in each corresponding domain. For instance, in LC-TIMS-TOF measurement data, multiple ions from the same ionspecies frequently fall within 3 seconds in the RT domain, within 0.03 1 / Ko in the IM domain and within 10 parts per million (ppm) for the m / z domain. These resolutions represent a conceivable example of a resolution-defined neighbourhood. In the frame / scan / TOF index domain (see further below), they can be expressed e.g. as 4 / 30 / 4. A tolerance-defined neighbourhood, by contrast, preferably corresponds to per-domain bounds that are selected independently of the intrinsic acquisition resolution and reflect an admissible deviation for comparison or querying purposes, for example arising from calibration uncertainty, instrumental drift or user-defined query windows. As a non-limiting example, a tolerance-defined neighbourhood in the m / z domain may correspond to a ±20 ppm window around a target mass, even if the effective acquisition resolution corresponds to a smaller ppm spacing.

[0024] For the purposes of the present application, a “local coordinate neighbourhood” preferably refers to a set / region of N-dimensional coordinates defined around a reference coordinate by per-domain bounds in each corresponding physical measurement domain, wherein the per-domain bounds are preferably derived from and in particular correspond to the associated resolution and / or tolerance of the respective physical measurement domain.

[0025] For the purposes of the present application, a “positional numeral system” preferably refers to a representation of an integer coordinate in a base x, wherein the coordinate is expressed as an ordered sequence of digits and each digit position has a weight determined by a power of the base x, such that the higher-order digit positions correspond to higher powers of x and thereby encode the most significant part of the represented coordinate.

[0026] For the purposes of the present application, “common high-order digits” preferably means that two or more 1 -dimensional coordinates, when represented in base x, share an identical sequence of leading (most significant) digits of at least length one, such that the coordinates have a common prefix in said positional numeral system and are therefore preferably represented on a common branch of the prefix tree up to the depth corresponding to the shared leading digits.

[0027] Each domain’s resolution preferably bounds and defines the smallest meaningful distinctions along that domain and thereby defines the extent of local neighbourhoods in the N-dimensional measurement space.

[0028] Thus, the measurement data is preferably generated by at least one continuousmeasurement system, for instance, a mass spectrometer setup comprising several separation techniques such as liquid chromatography (LC), trapped ion mobility separation (TIMS), time-of-f light (TOF), etc. In the case of mass spectrometry imaging (MSI) measurement system, additional measurement data such as the X and Y coordinates in a rasterized plane can be acquired. Measurement data from other continuous measurement systems are of course likewise conceivable. Each of the physical measurement domains is preferably characterized by an associated resolution and / or tolerance, such as ppm windows in the m / z or TOF domain, scan step sizes in the IM domain, sampling intervals in the RT or spatial domains. Due to the continuous measurement across multiple physical measurement domains, such measurement data is typically sparse and exhibits many structures and / or clusters across these domains, wherein said structures and / or clusters are bounded and defined by the resolution of the respective physical measurement domains.

[0029] The measurement data can be obtained from at least one continuous measurement system while the measurement system performs at least one measurement, i.e. while being acquired, the measurement data can immediately be converted to the format according to the invention. However, it is likewise conceivable that the measurement data is obtained from at least one archived format, i.e. previously acquired measurement data can be converted to the format according to the invention. In other words, the method according to the invention can be coupled to at least one measurement system and / or can be configured to convert and / or archive existing formats. Again, in other words, the measurement data can be obtained from a stored traditional raw format and can then be converted and stored with the method according to the invention, or the measurement data can be directly converted and stored with the method according to the invention. In any case, the method according to the invention enables a conversion of measurement data in a lossless manner to save disk space fore.g. long-term archival. Hence, in a first aspect, the method according to the invention enables savings in disk space, i.e. reduces the storage requirements for measurement data. Moreover, and as will be explained in greater detail below, in a second aspect the method according to the invention also enables more efficient processing and querying, especially with random access across one or more physical measurement domains.

[0030] The measurement data can be obtained from a variety of samples such as biological or chemical samples, e.g. blood plasma, tissue such as brain tissue, lung tissue, etc. However, other samples are likewise conceivable.A measurement system preferably is an apparatus, such as a mass spectrometer or imaging system, configured to acquire measurement data along one or more physical measurement domains.

[0031] The coordinates associated with the measurement data preferably are integer coordinates and / or are countable. These coordinates are preferably indices, in particular the above-mentioned spectrometric and / or spatial indices, resulting from a discretization of continuous physical measurement domains, wherein the discretization is determined by the resolution and / or tolerance of the respective domain. For instance, in the event of the measurement data originating from a measurement system comprising a TOF mass spectrometer, the coordinates are TOF indices which are directly related to m / z values and whose spacing is preferably determined by the timing resolution of the TOF separation, thereby defining the effective resolution in the m / z domain. In the event of the measurement data originating from a measurement system comprising a TIMS cell, the coordinates comprise scan indices which are directly related to ion mobility values and their associated CCS values and whose discretization is preferably defined by the scan step size of the mobility separation. More generally expressed, the measurement data corresponding, for instance, to LC-TIMS-TOF measurement data can be provided in the coordinates, in particular in the spectrometric and / or spatial indices:

[0032] 1) frame index related to retention time (RT) and associated hydrophobicity,

[0033] 2) scan index related to ion mobility (IM) and associated CCS, and

[0034] 3) TOF index related to m / z

[0035] In any case, the present invention is based on the insight that instead of storing the measurement data as M unique / V-dimensional coordinates, i.e. as multidimensional coordinates, they can instead be converted to 1 -dimensional coordinates. To this end, the measurement data is preferably structured and / or clustered across the N physical measurement domains in a manner that is preferably bounded and defined by the resolution and / or tolerance of the respective physical measurement domains, such that many of the N-dimensional coordinates that lie within a resolution-defined neighbourhood share common high-order digits after transformation when expressed in a positional numeral system. Furthermore, the bijective transformation is preferably configured to use the per-domain resolutions and / or tolerances. Since the prefix tree represents these 1D coordinates, the prefix tree contains the exact same information as these coordinates in the lossless case.Importantly, the above effect is not achieved by an arbitrary or purely mathematical one-dimensional mapping, even if bijective. In contrast to generic bijections that preserve uniqueness but are agnostic to the physical meaning of the underlying dimensions, the transformation according to the present invention is preferably explicitly parameterized by the resolutions and / or tolerances of the respective physical measurement domains. This resolution-dependent parameterization ensures that coordinates falling within physically meaningful, resolution-defined and / or tolerance-defined neighbourhoods in the N-dimensional measurement space are preferably deliberately mapped to one-dimensional coordinates sharing long common prefixes. These shared prefixes are therefore preferably not incidental, but are induced to reflect the structure of spectrometric and / or spatial measurement data as determined by the measurement process itself. As a consequence, the resulting prefix tree simultaneously reduces entropy for storage and enables efficient pruning and traversal during querying, since resolution-consistent neighbourhoods can be represented, accepted, or rejected at the prefix level rather than through exhaustive inspection of individual coordinates.

[0036] The M unique / V-dimensional coordinates are preferably obtained as a collection, such as a set, a list, an array, a vector, etc. as it is well-known in the art. The collection preferably reflects a sparse occupancy of the N-dimensional coordinate space due to the structured nature of the measurement data.

[0037] In the step of obtaining the M unique / V-dimensional coordinates, values associated with these M unique coordinates are preferably also obtained. In such a case, the values are preferably stored alongside these coordinates. Values of the / V-dimensional coordinates that lie within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space are preferably comparable. In other words, values associated with coordinates that lie within the same resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space typically differ only by small amounts and therefore exhibit similar value characteristics. This reflects the fact that the underlying physical measurement varies smoothly within the resolution or tolerance of the respective measurement domains. These values preferably are intensity values, and wherein preferably the intensity values represent an ion count.

[0038] The step of transforming the M unique / V-dimensional coordinates with a bijection to 1-dimensional coordinates preferably comprises the application of at most N componentfunctions to each of the M unique / V-dimensional coordinates, whereby N*M transformed coordinates are obtained, and wherein preferably the transformation comprises the summation of the N component functions. Additionally or alternatively, the bijection is preferably configured such that / V-dimensional coordinates that lie within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space are mapped to 1 -dimensional coordinates that share common high-order digits when expressed in a positional numeral system.

[0039] That is, the component functions are preferably parameterized in accordance with the resolutions or index ranges of the respective physical measurement domains, such that coordinates lying within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space are preferably mapped to 1 -dimensional coordinates sharing common high-order digits when expressed in a positional numeral system.

[0040] That is, the M unique / V-dimensional coordinates are preferably converted with a bijection comprising a combination of at most N component functions to 1D coordinates, such that measurements that are within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space, as determined by the per-domain resolutions or tolerances, are mapped to 1 -dimensional coordinates that deliberately share common high-order digits and prefixes when represented in a positional numeral system, wherein this prefix sharing reflects resolution-defined structures or clusters rather than numerical proximity in the 1 -dimensional domain.

[0041] A bijective transformation f x , x2, ■■■,xn) = y has the advantage that it is lossless. To this end, component functions are preferably used to transform each of the N dimensions independently and preferably are summed together such that f x1x2,

[0042]

[0043] = / i(%i) + 1 - fnxn) = y, but other bijective transformations are likewise conceivable.

[0044] As already mentioned, the coordinates are preferably integer coordinates. To ensure a bijection, a component function preferably shifts the bits of a binary representation such that they occupy a unique set of bits for this dimension. For instance, for an N-dimensional coordinate x, the coordinate xtin the / -th dimension, expressed in binary as am• 2m-I - 1- • 21+ a0• 2° with ate {0, 1} can utilize the component function fi xt) = am■ 2m+bm+ ••• + di • 21+bl+ a0• 20+b° with bm> ••• > b > b0and bke {0, 1, 2, ... }, wherein the bit offsets are selected in accordance with the resolution and / or tolerance of the respective physical measurement domain, such that coordinates belonging to the same structureand / or cluster map to 1 -dimensional coordinates that share a common high-order bit prefix. If for all dimensions / and j all k + bkare different and cover all positive integers {0, 1, ... , Z}, the summation f

[0045]

[0046] {xvx2, ■■■,xn) = / i(%i) + f2(x2) H - H / „(%„) = y indeed ensures a bijection. Other component functions are likewise conceivable.

[0047] Converting the M unique N-dimensional integer coordinates to 1 -dimensional integer coordinates is in contrast to the solutions of the prior art, wherein measurement data such as mass spectrometry data in general and LC-TIMS-TOF measurement data in particular are typically stored in a matrix or array format. For example, MS spectra are typically stored as a list of m / z-intensity pairs with metadata describing other coordinates such as RT or IM. In the case of TIMS-TOF, raw measurement data comprises multiple sequential scans that each contain alternating TOF differences and intensities. Hence, when coupled to an LC that acquires a list of sequential arrays such measurement data can be interpreted as a list of compressed sparse row matrices (CSRs). The same applies to measurement data obtained from other continuous measurement systems. CSRs, however, do not explicitly exploit or systematically leverage resolution-defined and / or tolerance-defined structures and / or clusters that occur across multiple physical measurement domains, even though these are typically exhibited in e.g. LC or TIMS measurement data. In contrast, the present method explicitly uses the per-domain resolutions or tolerances to parameterize the coordinate transformation such that coordinates belonging to the same resolution-defined and / or tolerance-defined structure or cluster are mapped to 1 -dimensional coordinates sharing common high-order digits and prefixes.

[0048] It is preferred that a function f (xnx2, ...,xn) is used to transform / V-dimensional coordinates that lie within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space into 1 -dimensional coordinates that deliberately share common high-order digits when expressed in a positional numeral system.. Using a resolution-dependent bijective transformation parameterized by these domain-specific resolutions, such detections are mapped to 1 -dimensional coordinates that share long common prefixes. As a result, detections that form dense clusters in the N-dimensional measurement space are represented on common branches of the prefix tree, and as a result a more efficient serialization and / or querying is achieved. However, other functions are likewise conceivable.

[0049] As there are many clusters in e.g. sparse 3D LC-TIMS-TOF measurement data such as ion species detected in roughly Gaussian patterns for the RT (frame), IM (scan) and m / z (TOF)domain, wherein these patterns correspond to clusters whose extent is defined by the resolution and / or tolerance of the respective measurement domains, many points that belong to such resolution-defined and / or tolerance-defined clusters have similar prefixes as a result of the resolution- dependent coordinate transformation being configured such that coordinates within the same resolution-defined and / or tolerance-defined neighbourhood map to 1 -dimensional coordinates sharing common high-order digits. For MSI, ion species are detected in structured spatial patterns for the X and Y coordinates, which likewise form resolution-defined and / or tolerance-defined clusters that yield shared high-order digits and prefixes after transformation.

[0050] In the method according to the invention, the 1 -dimensional coordinates are then converted to a prefix tree. In other words, a prefix tree for the M unique 1 -dimensional coordinates is built to preferably represent these coordinates, wherein the tree topology reflects shared high-order digits and shared prefixes that are deliberately induced by the resolution-defined transformation of the structured and / or clustered measurement data.

[0051] In the step of building the prefix tree, the 1D coordinates are preferably represented in base x, particularly preferably in base 8. In particular, in the step of building a prefix tree from the M unique 1-dimensional coordinates each 1-dimensional coordinate is preferably expressed in base x, the selection of base x determining the branching topology of the prefix tree, such that 1 -dimensional coordinates sharing common high-order digits are represented on a common branch of the prefix tree. The base is preferably selected to shape the topology of the prefix tree according to a redundancy in the prefixes resulting from clustering present in the measurement data, and wherein said clustering is preferably defined by the resolution and / or tolerance of the respective measurement domains. The selection of the base preferably cooperates with the resolution-parameterized and / or tolerance-parameterized bijective transformation such that coordinates lying within a resolution-defined and / or tolerance-defined neighbourhood are mapped onto common branches of the prefix tree by sharing common high-order digits.

[0052] Since the M unique 1 -dimensional coordinates are all still unique after the bijective transformation, they can be represented by using a prefix tree. This can for instance be done by representing the 1D coordinates in a base x such as base 2 (binary), base 8 (octal) or base 10 (decimal), but other representations are likewise conceivable. Prefix trees and their representations, for instance binary trees, are well-known in the art of computer science, however their application here relies on the deliberate creation of shared prefixesby configuring the bijective transformation using the resolutions or tolerances of the physical measurement domains, in combination with the base selection, rather than on pre-existing sequential data.

[0053] Said preferable octal representation simplifies further implementations of the method according to the invention. The same is true for any representation with base x a multiple of 8, as such representations provide a controlled trade-off between tree depth and branching factor, but other representations are likewise conceivable.

[0054] The prefix tree can optionally be pruned to discard certain branches. This can be done on any criterium, including but not limited to criteria derived from sparsity patterns and / or density characteristics of the measurement data, such as branch density, wherein said sparsity and / or density criteria are preferably evaluated relative to local neighbourhoods whose extent along each axis corresponds to the resolution or tolerance of the respective physical measurement domains.

[0055] Noise in the measurement data is preferably discarded based on a local sparsity criterion. Additionally or alternatively, coordinates of the M unique 1 -dimensional coordinates are preferably not considered noise when they lie within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space.

[0056] For the purposes of the present application, a “sparsity pattern” preferably refers to the distribution of occupied versus unoccupied coordinate locations in the N-dimensional measurement space (or in its 1 -dimensional representation), including the presence of localized regions of high occupancy (structures and / or clusters) and regions of low occupancy, wherein said occupancy is preferably assessed relative to neighbourhoods whose extent along each physical measurement domain is defined by the associated resolution and / or tolerance.

[0057] For the purposes of the present application, a “local sparsity criterion” preferably refers to a rule for classifying coordinates as noise or non-noise based on the number and / or density of occupied coordinates within a local neighbourhood (e.g., within a window / hypercube) whose extent along each physical measurement domain is preferably determined by the associated resolution and / or tolerance, such that isolated coordinates lacking sufficient neighbouring occupancy within said neighbourhood can be treated as noise.Said step of pruning the prefix tree also serves the purpose of removing extremely sparsely populated structures, which can be interpreted as denoising the measurement data.

[0058] In this context, “extremely sparsely populated structures” are preferably assessed relative to an expected occupancy within a resolution- and / or tolerance-defined local coordinate neighbourhood, such that isolated coordinates lacking sufficient support within the bounds imposed by the associated physical measurement domain can be treated as noise.

[0059] For the purposes of the present application, a “local coordinate neighbourhood” preferably refers to a region in the N-dimensional measurement space defined around a reference N-dimensional coordinate by per-domain bounds in each corresponding physical measurement domain, wherein said per-domain bounds are derived from, and preferably correspond to, the associated resolution and / or tolerance of the respective physical measurement domain.

[0060] Said step of pruning the prefix trees also serves the purpose of reducing storage requirements, albeit at the cost of not being lossless anymore. This is because if the tree is pruned, at least one 1 D coordinate is lost and thereby also at least one of the M unique N-dimensional coordinates of the obtained measurement data.

[0061] Pruning can be done, for instance by utilizing the density of a branch, i.e. the number of present nodes divided by the total number of possible nodes with the same prefix, i.e. on the same branch, as criterion for pruning and thereby discarding prefixes corresponding to regions of the N-dimensional measurement space that are extremely sparsely populated or do not exhibit larger structures and / or clusters, wherein such extremely sparsely populated structures are assessed relative to an expected occupancy within a resolution-defined neighbourhood, such that isolated coordinates not supported within the resolution of the respective measurement domains can be treated as noise.

[0062] Prior to serializing the prefix tree, the prefix tree is preferably split into at least one subtree, and wherein each of the subtrees provides an independent prefix tree.

[0063] Large lists of coordinates lead to large prefix trees. To avoid having to deal with the whole prefix tree at once, it can be split into at least one independent smaller subtree by partitioning the list of coordinates, wherein each subtree preferably corresponds to a coherent prefix region induced by the structure and / or clustering of the measurement data.Said splitting of the prefix tree is very beneficial to reduce memory usage and querying, as only those subtrees whose prefixes are relevant for a given operation, typically a structure or cluster, need to be accessed as will be explained in greater detail below.

[0064] Preferably, the structures and / or clusters present in the measurement data correspond to localized, resolution-defined and / or tolerance-defined neighbourhoods in the N-dimensional measurement space. Owing to the resolution-defined and / or tolerance-defined transformation and the resulting prefix structure, one-dimensional coordinates that belong to the same structure and / or cluster are predominantly represented within a single subtree when the prefix tree is split. This limits fragmentation of individual structures and / or clusters across multiple subtrees and ensures that measurement data belonging to the same resolution-defined and / or tolerance-defined neighbourhood can be processed, stored, and queried largely within a single subtree.

[0065] Each of the at least one subtree can then individually be used as independent prefix tree in the further method steps such as the serialization and the writing to the non-transitory tangible computer-readable storage medium. That is, any explanations made herewith regarding a prefix tree preferably likewise apply to at least one subtree that emerged from the prefix tree upon splitting and vice versa.

[0066] After splitting the prefix tree into the at least one subtree, the at least one subtree is preferably indexed, particularly preferably with another prefix tree.

[0067] The measurement data is particularly preferably split into said subtrees such that different structures and / or clusters present in the measurement data is mapped to different branches of the prefix tree and thereby to different subtrees. Said structures and / or clusters particularly preferably correspond to localized, resolution-defined and / or tolerance-defined neighbourhoods in the N-dimensional measurement space, such that 1 -dimensional coordinates belonging to a same structure and / or cluster are predominantly contained within a single subtree. Additionally or alternatively, the at least one subtree is particularly preferably indexed such that, when querying a subset of the measurement data with at least one query, only those subtrees whose prefixes are consistent with the query can be accessed.

[0068] When subtrees are used to split up a larger prefix tree, the subtrees can be constructedsuch that all of the subtrees coincide with independent branches of the original prefix tree, thereby preserving the prefix-based partitioning of the 1 -dimensional coordinate space and in turn preserving the original structures and / or clusters in the measurement data. This can for instance be done by splitting the original prefix tree at a certain branch depth, but other splitting methods are likewise conceivable.

[0069] To quickly access all subtrees, their addresses, either in memory or on a storage device, can be indexed, preferably with another prefix tree which is herein called a supertree. Other indexing methods are likewise conceivable.

[0070] The 1 D coordinates of this supertree can for instance be chosen as the smallest prefix present in each subtree, such that the supertree provides an index over prefix ranges of the original prefix tree. The addresses of these subtrees can then be used as values associated with these coordinates as described above. Other implementations of the supertree are likewise conceivable.

[0071] A supertree can then individually be used as independent prefix tree in the further method steps such as the serialization and the writing to the non-transitory tangible computer-readable storage medium. That is, any explanations made herein regarding a prefix tree preferably likewise apply to the supertree emerged from the prefix tree upon splitting and vice versa.

[0072] The prefix tree is preferably represented in base x. In the step of serializing the prefix tree, each node of the prefix tree is preferably represented as a string of x bits with bit / set to 1 in the event that said node has a direct descendent / . The thus serialized nodes are then preferably concatenated either breadth-first or depth-first while ensuring a lexicographical ordering of the 1D coordinates when the tree is traversed, such that the serialization order reflects the prefix structure of the data which in turn follows the structures and / or clusters of the measurement data without requiring any additional metadata.

[0073] There are many ways to serialize a tree. As a goal of this first aspect of the invention is to reduce storage requirements for measurement data, the prefix tree is preferably serialized such that a lexicographic ordering of coordinates when traversing the tree removes the need to store any metadata or explicit pointers between nodes, thereby encoding the tree topology implicitly in the serialized byte sequence. Other serializations are likewise conceivable though.In the step of serializing the prefix tree, the serialization preferably comprises the steps of encoding, for each node of the prefix tree, only the presence or absence of its possible direct descendants as bitflags, and concatenating the resulting encodings in an order consistent with a lexicographic traversal of the 1 -dimensional coordinates.

[0074] That is, the prefix tree is preferably serialized such that only structural information inherent to the tree topology is recorded. In particular, each node is represented solely by information indicating the presence or absence of its possible direct descendants. The nodes are written in an order consistent with a lexicographic traversal of the one-dimensional coordinates represented by the tree. Due to this ordering, the relative position of the serialized representations implicitly defines the hierarchical relationships between nodes, such that the topology of the prefix tree can be reconstructed from the serialized data itself, without storing explicit node identifiers, pointers, or other auxiliary metadata.

[0075] A tree can be traversed either depth-first or breadth-first in an order consistent with the lexicographic ordering of coordinates represented in base x. The order in which the nodes are visited is also the order in which they are concatenated for the tree serialization, ensuring that nodes sharing long prefixes are placed close to each other in the serialized representation.

[0076] In the step of serializing the prefix tree, each node is thus preferably represented exclusively by information indicating the presence or absence of its possible direct descendants, for example in the form of bitflags. The resulting encodings are concatenated in an order consistent with a lexicographic traversal of the one-dimensional coordinates represented by the prefix tree. As a result, the relative ordering of the serialized encodings implicitly defines the topology of the prefix tree, such that the complete tree structure can be reconstructed without storing explicit node identifiers, pointers, or other auxiliary metadata.

[0077] To serialize a node, a string of x bits is preferably used, with bit / set to 1 if and only if it has a descendent / . Unlike classical prefix trees operating on sequential symbols, the prefixes represented by the nodes here arise from shared high-order digits induced by the resolution-defined transformation of structured and / or clustered measurement data. As the tree is constructed by using 1D coordinates which have a fixed length L when represented in base x, corresponding to a fixed discretization depth of the underlying N-dimensional measurement space, leaf nodes with a representation of 0 occur only at depth L and hencedo not need to be explicitly serialized.

[0078] When a prefix tree is represented in an octal representation, the serialization of a single node is exactly one byte, simplifying its interpretation and computational implementation while directly controlling the size of the serialized on-disk representation.

[0079] This minimal serialization is enabled by the bounded branching factor imposed by the base-x representation, which limits each node to at most x possible direct descendants. Moreover, because the measurement data is preferably structured and / or clustered, nodes of the prefix tree typically correspond either to regions with very sparse occupancy or to dense structures following shared prefixes. As a result, the bitflag representations of nodes tend to contain either very few or very many set bits. The achieved compact serialization is therefore not incidental, but a direct consequence of the clustered nature of the measurement data in combination with the resolution-defined transformation and the selected tree topology.

[0080] Once the prefix tree is serialized, the resulting bytes are preferably written to the non-transitory tangible computer-readable storage medium. This can be done either directly, or after another compression step as explained now.

[0081] Prior to writing to the non-transitory tangible computer-readable storage medium, the serialized prefix tree is preferably compressed. This compression takes advantage of redundancies arising from the structured and / or clustered nature of the measurement data that are not fully captured by shared prefixes alone. In particular, regions of the prefix tree corresponding to dense structures or repeated structural patterns give rise to recurring bit patterns or regularities in the serialized representation, which can be exploited by conventional compression algorithms to further reduce storage requirements.

[0082] That is, once a prefix tree (or if applicable a subtree or supertree) is serialized, a resulting byte stream is preferably compressed before being written to the non-transitory tangible computer-readable storage medium.

[0083] Once a prefix tree (or if applicable a subtree or supertree) is serialized, any compression algorithm can be used to compress the resulting bytes, for instance with zstd (Zstandard) or LZ (Lempel-Ziv) compression as it is well-known in the art. Other compression algorithms are likewise conceivable.Depending on the density and amount of repetition present in the serialized prefix tree, which directly reflects shared prefixes and structural redundancy induced by clustered measurement data, this compression can significantly reduce space requirements.

[0084] For instance, in the case of LC-TIMS-TOF measurement data, the prefix tree itself can have many repetitive paths as different ion species that are related such as fragments from the same precursor or naturally occurring isotopes will have similar branches differing only in the TOF domain.

[0085] The values associated with the coordinates are preferably serialized by reordering the values such that their associated 1 D coordinates are sorted lexicographically.

[0086] Since values are associated with specific coordinates, this association needs to be retained when serializing both the values and the coordinates.

[0087] As the original / V-dimensional coordinates are unique, the bijective transformation also ensures the 1D coordinates are unique. Therefore, there is only one unique way to lexicographically sort these 1D coordinates. Since the coordinates have a unique lexicographically order, the same order can also be used for the values. As a result, the values are unambiguously associated with the correct coordinates, regardless of how the coordinates themselves or their prefix tree representation are serialized.

[0088] As the lexicographic ordering of the 1D coordinates preferably reflects shared prefixes induced by structured and / or clustered measurement data, values associated with coordinates that belong to the same structure and / or cluster are placed close to each other in the serialized value stream.

[0089] In addition to serializing the prefix tree, values associated with the coordinates may likewise be serialized. Preferably, the values are reordered such that their associated one-dimensional coordinates are sorted in lexicographic order. As a result, the index of a coordinate in the serialized prefix tree implicitly identifies the corresponding value by the same ordering, without requiring additional metadata to maintain the association between coordinates and values.

[0090] If coordinates are grouped into subgroups by forming subtrees, the values associated withthese coordinates are likewise split while still being associated with their respective coordinates of the subgroups.

[0091] If coordinates are pruned by pruning (sub)tree(s), the values associated with these coordinates are likewise pruned while still being associated with their respective coordinates of the subgroups.

[0092] The reordered (pruned) values associated with these (sub)tree(s) are then serialized by serializing their individual values to bytes and concatenating these bytes. For example, if values are represented by a u32, this can be serialized to 4 bytes in for instance big or little endian. Other serializations of individual values are likewise conceivable.

[0093] Prior to writing to the non-transitory tangible computer-readable storage medium, the serialization of the values associated with the coordinates is preferably compressed, preferably by a compression algorithm wherein the ordering of the values induced by the prefix tree groups values associated with coordinates belonging to the same structure and / or cluster. Due to the lexicographic ordering induced by the prefix tree, values associated with coordinates belonging to the same structure and / or cluster are placed close to each other in the serialized value stream. Since such values typically exhibit similar value characteristics, this ordering improves the effectiveness of subsequent compression.

[0094] Before compression of the values, positional bytes from each value are preferably concatenated to improve compression performance. Because values associated with coordinates belonging to the same structure and / or cluster tend to have similar high-order and / or low-order bytes, this byte reordering groups similar bytes together and can further improve compression efficiency.

[0095] That is, once values associated with the coordinates are serialized, the resulting byte stream is preferably compressed before being written to the non-transitory tangible computer-readable storage medium.

[0096] Once values associated with the coordinates are serialized, any compression algorithm can be used to compress the resulting bytes, for instance with zstd (Zstandard) or LZ (Lempel-Ziv) compression as it is well-known in the art. Other compression algorithms are likewise conceivable.Depending on the similarity of value characteristics within clusters induced by the ordering of the associated coordinates, this compression can further reduce storage requirements.

[0097] Since many values are small, they often can be compressed to save storage space.

[0098] Since many compressors take bytes and are unaware of structures bytes such as u32, it is often beneficial to do a byte reordering such that the most significant bytes of each value are grouped together.

[0099] For instance, the serialized values can be reordered such that the first byte of all u32 values are concatenated, then the second byte of all u32 values, then the third byte of all values, etc.

[0100] Thus, during serialization of the values, positional bytes from each value are preferably concatenated together rather than first concatenating the bytes per value before concatenating all values.

[0101] In a second aspect, a computer-implemented method of querying a database for a subset of measurement data preferably with random access across one or more physical measurement domains is provided. The database comprises measurement data of at least one measurement system and has been provided by the steps of i) obtaining M unique N-dimensional coordinates being associated with the measurement data, ii) transforming the M unique / V-dimensional coordinates with a bijection to 1 -dimensional coordinates, whereby M unique 1 -dimensional coordinates are obtained, and iii) building a prefix tree from the M unique 1 -dimensional coordinates. The method of querying the database comprises the steps of i) receiving at least one query defining at least one N-dimensional constraint, ii) transforming the N-dimensional constraint, using the same bijection, into at least one corresponding 1 -dimensional constraint, iii) querying the prefix tree with the at least one 1-dimensional constraint, iv) retrieving at least one subset of 1 -dimensional coordinates from the prefix tree, and v) converting the at least one retrieved subset of 1 -dimensional coordinates back into corresponding N-dimensional coordinates being associated with the measurement data using an inverse of the bijection.

[0102] Said database preferably corresponds to a database that is generated in the method according to the first aspect of the invention explained above. Hence, it is preferred that the method of querying the database is preferably performed in a database comprisingmeasurement data of at least one measurement system that have been converted to a prefix tree in the method according to the first aspect of the invention. Thus, any explanations made herein with regard to the method according to the first aspect and / or of constructing a prefix tree preferably likewise apply to the method of querying the database according to the second aspect and vice versa.

[0103] For instance, the measurement data preferably is spectrometric measurement data and / or spatial measurement data, and wherein each N-dimensional coordinate preferably comprises N spectrometric indices and / or N spatial indices acquired from a spectrometric component and / or from a spatial component of the measurement system. Furthermore, each spectrometric component and / or spatial component of the measurement system preferably performs a continuous discrimination of the measurement data along a respective physical measurement domain having an associated resolution and / or tolerance, and wherein the measurement data preferably defines a sparse occupancy of an N-dimensional measurement space while comprising localized structures and / or clusters whose extent along each physical measurement domain is defined by the resolution and / or tolerance associated with the respective physical measurement domain, etc.

[0104] When querying the database, at least one query defining at least one N-dimensional constraint is received. Each N-dimensional coordinate preferably being defined by N spectrometric indices and / or N spatial indices associated with the measurement data as mentioned earlier.

[0105] Said N-dimensional constraint is transformed into at least one corresponding 1 -dimensional constraint using the same bijection as was used in the provision of the database when transforming the M unique / V-dimensional coordinates with a bijection to 1 -dimensional coordinates. That is, the one or more N-dimensional constraints are preferably transformed using the same resolution-defined and / or tolerance-defined transformation as used during storage, into corresponding one-dimensional constraints that can be expressed as one or more prefixes and / or one-dimensional intervals. Then, the prefix tree is queried with the at least one 1 -dimensional constraint, at least one subset of 1 -dimensional coordinates from the prefix tree are retrieved, and the at least one retrieved subset of 1 -dimensional coordinates are converted back into corresponding N-dimensional coordinates being associated with the measurement data using an inverse of the bijection. During querying, i.e. traversal of the prefix tree, only those branches whose prefixes are consistent with the transformed one-dimensional constraint are visited.A query may define a fixed coordinate in one physical measurement domain and a coordinate range in another. For example, in an LC-TIMS-TOF measurement system, a query may fix a target m / z value and define a coordinate range corresponding to a ±20 ppm window around that value, while leaving the retention time and ion mobility domains unconstrained or bounded by broader coordinate ranges. More generally, queries may fix one point or a small range in one or more dimensions and retrieve all coordinates, and optionally values, from the remaining dimensions. When transformed using the bijection, such constraints correspond to one or more prefixes and / or 1 -dimensional intervals in the 1 -dimensional coordinate space, allowing the prefix tree to be traversed only along branches consistent with these constraints. In an LC-TIMS-TOF measurement system, this can for instance correspond to an XIC or XIM of a specific m / z value within a small ppm window

[0106] The querying is very efficient, as the prefix tree can efficiently be traversed. Since each node corresponds to a prefix, it can quickly be checked if this prefix satisfies the query. If the prefix of a node indeed satisfies the query, all descendent nodes can then be processed in the same manner. Moreover, if the prefix of a node does not satisfy the query, none of its descendants will satisfy the query either. As a result, large parts of the tree do not need to be traversed at all, saving a lot of time.

[0107] Once the query is completed, the subset of measurement data, i.e. a subset of the N-dimensional coordinates, can then be used for further processing. It is possible that the subset is empty, yielding zero N-dimensional coordinates.

[0108] As mentioned earlier, after building the prefix tree, the prefix tree is preferably split into at least one subtree, and wherein each of the subtrees provides an independent prefix tree, and wherein, after splitting the prefix tree into the at least one subtree, the at least one subtree is preferably indexed, preferably with another prefix tree. The measurement data is preferably split into said subtrees such that different structures and / or clusters present in the measurement data is mapped to different branches of the prefix tree and thereby to different subtrees. Additionally or alternatively, the at least one subtree is preferably indexed such that, when querying a subset of the measurement data with the at least one query, only those subtrees whose prefixes are consistent with the query can be accessed. Said structures and / or clusters preferably correspond to localized, resolution-defined and / or tolerance-defined neighbourhoods in a N-dimensional measurement space defined by themeasurement data, such that 1 -dimensional coordinates belonging to a same structure and / or cluster are predominantly contained within a single subtree.

[0109] As also mentioned earlier, the at least one N-dimensional constraint preferably comprises at least one fixed coordinate and / or at least one coordinate range defined with respect to the resolution and / or tolerance associated with at least one physical measurement domain. In the step of transforming the N-dimensional coordinates, the 1 -dimensional constraint preferably comprises at least one prefix and / or at least one 1D interval, such that, in the step of querying the prefix tree, only those parts of the prefix tree whose prefixes are consistent with the 1 -dimensional constraint are traversed. By defining the at least one N-dimensional constraint in a resolution-defined and / or tolerance-defined manner, the number of higher-order prefixes satisfying a given query is limited, thereby restricting traversal to a comparatively small number of branches and enabling efficient random-access querying.

[0110] As also mentioned earlier, after building the prefix tree, the provision of the database preferably further comprises the steps of serializing the prefix tree and writing the serialized prefix tree to a non-transitory tangible computer-readable storage medium. In this case, the method of querying the database preferably further comprises the step of deserializing the prefix tree before querying. In other words, if a prefix tree is stored on a non-transitory tangible computer-readable storage medium, it needs to be read from storage and is preferably deserialized.

[0111] If the prefix tree was compressed after serialization, it needs to be decompressed after deserialization. Hence, in this case the method of querying the database preferably further comprises the step of decompressing the deserialized prefix tree.

[0112] As mentioned earlier, in the step of obtaining the M unique / V-dimensional coordinates in the provision of the database, values associated with these M unique coordinates are preferably also obtained. In this case, the method of querying the database preferably further comprises the step of retrieving the values associated with the coordinates when performing the query.

[0113] That is, when there are values associated with the coordinates retrieved when performing a query, these values are preferably retrieved likewise. In particular, in the provision of the database, values associated with the M unique N-dimensional coordinates are preferably also obtained and serialized in an order corresponding to a lexicographic ordering of theassociated 1 -dimensional coordinates, and wherein the method of querying the database preferably further comprises retrieving the values associated with the retrieved 1 -dimensional coordinates by using an index of a retrieved 1 -dimensional coordinate in said lexicographic ordering to identify the associated value stored at a same index position. That is, the prefix tree is preferably serialized in an order consistent with a lexicographic traversal of the 1 -dimensional coordinates, such that the relative ordering of the serialized representations implicitly encodes the tree topology. As a result, the complete structure of the prefix tree can be reconstructed from the serialized representation itself, without requiring additional metadata such as explicit node identifiers, pointers, or auxiliary structural information.

[0114] If the values are written to the non-transitory tangible computer-readable storage medium in the provision of the database, the values associated with the prefix tree are preferably read from storage and deserialized.

[0115] If the values were compressed after serialization in the provision of the database, they are preferably decompressed after deserialization.

[0116] If the values were byte-reordered before compression in the provision of the database, they are preferably byte-reordered after decompression.

[0117] As the values are ordered in the same way as the 1D coordinates, an index of a 1D coordinate in the prefix tree suffices to retrieve its associated value, as a value has the same index as its associated 1 D coordinate.

[0118] As mentioned earlier, when providing the database, the prefix tree is preferably split into at least one subtree and an index, preferably a supertree, and the at least one subtree is written to a non-transitory tangible computer-readable storage medium. In this case, the method of querying the database preferably further comprises the step of querying the index such that only the subtrees with a prefix satisfying the query are deserialized and queried.

[0119] When a prefix tree was split into at least one subtree that was written to disk in the provision of the database, it often is not needed to deserialize all subtrees when performing a query. With the help of an index, preferably a supertree, only those subtrees with a prefix satisfying the query need to be deserialized.If the index is a supertree that was serialized to storage in the provision of the database, the supertree is preferably deserialized in the method of querying the database.

[0120] If the index is a supertree that was compressed in the provision of the database, the supertree is preferably decompressed after deserialization in the method of querying the database.

[0121] If the index is a supertree that was serialized to storage in the provision of the database, the values associated with the supertree are preferably first deserialized in the method of querying the database.

[0122] If the index is a supertree that was serialized to storage in the provision of the database, and the values associated with the supertree are compressed in the provision of the database, they are preferably decompressed in the method of querying the database.

[0123] If the index is a supertree that was serialized to storage in the provision of the database, and the values associated with the supertree are byte reordered before compression in the provision of the database, they are preferably byte reordered in the method of querying the database.

[0124] If there are values associated with the coordinates of the subtrees, they can also be retrieved with the query.

[0125] If the values associated with the coordinates of the subtrees are compressed in the provision of the database, they are preferably decompressed after deserialization in the method of querying the database.

[0126] If the values associated with the coordinates of the subtrees are byte reordered before compression in the provision of the database, they are preferably byte reordered after decompression in the method of querying the database.

[0127] As mentioned earlier, after splitting the prefix tree into the at least one subtree in the provision of the database, the at least one subtree can be indexed, preferably with another prefix tree. In this case, the supertree can be queried for only those subtrees whose prefix satisfies the query.The addresses of the subtrees satisfying the query can likewise be retrieved as mentioned earlier. That is, the values associated with the coordinates can be retrieved when performing a query.

[0128] As a result of the resolution-defined and / or tolerance-defined coordinate transformation and the subsequent prefix-based partitioning, one-dimensional coordinates corresponding to the same structure and / or cluster in the N-dimensional measurement space can be predominantly grouped within a common subtree. This reduces fragmentation of individual structures and / or clusters across multiple subtrees and ensures that measurement data belonging to the same resolution-defined and / or tolerance defined neighbourhood remain largely localized within a single subtree, facilitating coherent storage and efficient access. For each subtree whose prefix satisfies the query, the subtree can be queried as described above.

[0129] In any case, the measurement data is preferably mass spectrometry (MS) data and / or mass spectrometry imaging (MSI) data and / or liquid chromatography (LC) data and / or trapped ion mobility separation (TIMS) data and / or time-of-flight (TOF) data and / or combinations thereof. However, measurement data from other measurement systems are likewise conceivable.

[0130] The measurement data can be obtained from at least one measurement system while performing at least one measurement or the measurement data can be obtained from at least one archived format.

[0131] In a third aspect, a computer program product comprising a non-transitory tangible computer-readable storage medium whose contents cause a processor to perform the methods as described above is provided.

[0132] Any explanations made herein regarding the methods according to the first aspect and the second aspect preferably likewise apply to the computer program product according to the third aspect and vice versa.

[0133] In a fourth aspect, a measurement system, preferably a mass spectrometer optionally combined with a liquid chromatography (LC) unit and / or an ion mobility separator (IMS) is provided. The measurement system comprises a processor comprising instructions which, when carried out by the processor, causes the processor to carry out the methods asdescribed above.

[0134] Any explanations made herein regarding the methods according to the first aspect and the second aspect and the computer program product according to the third aspect preferably likewise apply to the measurement system according to the fourth aspect and vice versa.

[0135] The at least one measurement system is preferably configured to measure the measurement data mentioned above.

[0136] Thus, in summary it is noted that the present invention is based on the insight that both reduced storage requirements and efficient random access can arise from a deliberate coupling of a resolution-defined and / or tolerance-defined coordinate transformation with a prefix-tree representation. In particular, the bijective transformation from N-dimensional measurement coordinates to one-dimensional coordinates is not merely lossless, but is parameterized using the per-domain resolutions and / or tolerances of the underlying physical measurement domains. As a result, N-dimensional coordinates that lie within the same resolution-defined and / or tolerance-defined neighbourhood can be mapped to one-dimensional coordinates that share common high-order digits when expressed in a positional numeral system. These 1 -dimensional coordinates deliberately share prefixes that reflect physically meaningful structures and clusters present in the measurement data. Building a prefix tree over these 1 -dimensional coordinates therefore exploits this prefix redundancy to reduce entropy and storage requirements, and simultaneously enables efficient pruning and traversal during querying, since entire resolution-consistent neighbourhoods can be accepted or rejected by inspecting only a small number of high-order prefixes.

[0137] BRIEF DESCRIPTION OF THE DRAWINGS

[0138] Preferred embodiments of the invention are described in the following with reference to the drawings, which are for the purpose of illustrating the present preferred embodiments of the invention and not for the purpose of limiting the same. In the drawings,

[0139] Fig. 1 shows a schematic of a method of storing measurement data of a measurement system in a database on a non-transitory tangible computer-readable storage medium according to the invention;

[0140] Fig. 2 shows a schematic of a coordinate transformation with a bijection in a methodas illustrated in figure 1;

[0141] Fig. 3 shows a schematic of building a prefix tree in a method as illustrated in figure 1;

[0142] Fig. 4 shows a schematic of pruning a prefix tree in a method as illustrated in figure 1; Fig. 5 shows a schematic of splitting a prefix tree in a method as illustrated in figure 1 ; Fig. 6 shows a schematic of serializing a prefix tree in a method as illustrate in figure 1;

[0143] Fig. 7 shows a schematic of a method of querying a database for a subset of measurement data according to the invention;

[0144] Fig. 8 shows a schematic of querying a subset of measurement data by traversing only parts of a prefix tree in a method as illustrated in figure 7;

[0145] Fig. 9 shows a schematic of decompressing a prefix tree, a supertree and a subtree in a method as illustrated in figure 7;

[0146] Fig. 10 shows measurement data and a schematic of storing said measurement data in a method as illustrated in figure 1.

[0147] DESCRIPTION OF PREFERRED EMBODIMENTS

[0148] Aspects of the methods according to the invention are now illustrated with respect to the figures.

[0149] In particular, and as illustrated in figure 1, the method of storing measurement data of a measurement system in a database on a non-transitory tangible computer-readable storage medium can comprise the steps of:

[0150] S1: obtaining M unique / V-dimensional coordinates being associated with the measurement data and optionally also the associated values;

[0151] S2: transforming the M unique / V-dimensional coordinates with a bijection to 1 -dimensional coordinates, whereby M unique 1 -dimensional coordinates are obtained;

[0152] S3: building a prefix tree from the M unique 1 -dimensional coordinates to represent these 1D coordinates;

[0153] S4: optionally, after the step of building the prefix tree, pruning the prefix tree, for instance based on a branch density;

[0154] S5: optionally, prior to serializing the (pruned) prefix tree, the (pruned) prefix tree can be split into at least one subtree and optionally into an index such as a supertree;

[0155] S6: serializing the (pruned) prefix tree (or the subtree, index or supertree) and optionally the associated values;S6a: optionally, during serialization, the associated values can be byte-reordered;

[0156] S7: optionally, prior to writing to the non-transitory tangible computer-readable storage medium, the (pruned) prefix tree (or the subtree, index or supertree) and / or the associated values can be compressed;

[0157] S8: writing the (pruned) prefix tree (or the subtree, index or supertree) and optionally the associated values to the non-transitory tangible computer-readable storage medium.

[0158] Steps S1 to S8 are preferably performed in this order.

[0159] For illustrative purposes only, numerical examples of the method according to the invention are now illustrated with reference to figures 2 to 7.

[0160] That is, figure 2 schematically illustrates two numerical examples of a coordinate transformation with a bijection in step S2 of the method illustrated in figure 1. In particular, two bijective transformations f and g for a 3-dimensional coordinate a are given here. Both f and g use component functions and gtthat transform a coordinate xtby repositioning the bits such that the transformed bit coordinates of x , x2and x3are disjunct. The total transformation is then the summation of the component functions which indeed forms a bijection. A second coordinate b is also given that is adjacent to point a in a Euclidean metric. In 1D however, the distance d(f (a),f (b)) is larger than the distance d g(g),g(by).

[0161] Figure 3 schematically illustrates a numerical example of building a prefix tree in step S3 of the method illustrated in figure 1. Since the M 1 -dimensional coordinates obtained in step S2 are all unique, they can be indexed by using a prefix tree. Different representations of the coordinates can be used, each leading to different prefix trees. While any representation can be used, some representations are more efficient than others for serialization and / or accession downstream. Here, five 1 -dimensional coordinates are shown in both their binary and octal representation, as well as their associated values. The octal representation is used to create a prefix tree where the tree itself represents the coordinates. The values are then coupled to the leaves of the tree.

[0162] Figure 4 schematically illustrates a numerical example of pruning a prefix tree in step S4 of the method illustrated in figure 1. That is, the prefix tree can optionally be pruned to discard certain values. This can be done on any criterium, including but not limited to branch density. As an example, the prefix tree here can be pruned based on density. The crossed-out branch contains only one leaf whereas the branch diverges before the leaf level is reached.Figure 5 schematically illustrates a numerical example of splitting a prefix tree in step S5 of the method illustrated in figure 1. Large measurement data can lead to large prefix trees. To avoid having to deal with the whole prefix tree at once, it can be split at a certain depth (e.g. layer of the three). This creates a supertree and subtrees. Each of these can then individually be handled as independent trees. While this is an optional step, it is very beneficial to reduce memory usage and accession from disk downstream. In the example here, the tree is split at depth 4, leading to the supertree (t0) and 2 subtrees

[0163]

[0164] and t2).

[0165] Figure 6 schematically illustrates a numerical example of serializing a prefix tree in step S6 of the method illustrated in figure 1. To serialize the prefix tree, it is converted to bytes. There are many ways to do so. In the depicted example, an octal representation was chosen for the prefix tree. Each node is therefore converted to a byte, where a bit / is set to 1 if it has descent / . The thus serialized nodes are then preferably concatenated either breadth-first or depth-first while ensuring a lexicographical ordering of the 1 D coordinates when the tree is traversed. In this case, a breadth-first approach is depicted. If ordered lexicographically, no additional information is needed to ensure an unambiguous serialization. The sorting furthermore ensures an unambiguous ordering for the values, which can be serialized by any other means.

[0166] Figure 7 illustrates conceivable steps of a method of querying a database comprising measurement data that has been generated in a method as described above for a subset of measurement data according to invention:

[0167] s1: obtaining prefix tree (subtree, index or supertree) and optionally also associated values. s2: receiving at least one query defining at least one N-dimensional constraint.

[0168] s3: transforming the N-dimensional constraint, using the same bijection as used during storage, into at least one corresponding 1 -dimensional constraint.

[0169] s4: querying the prefix tree (subtree, index or supertree) with the at least one 1 -dimensional constraint. That is, when querying for at least one subset of measurement data in the database, the prefix tree is traversed only along branches whose prefixes are consistent with the transformed 1 -dimensional constraint. Since each node corresponds to a prefix, it can efficiently be determined whether the prefix satisfies the constraint. If the prefix of a node indeed satisfies the query, all descendent nodes can then be processed in the same manner. Moreover, if the prefix of a node does not satisfy the query, none of its descendants will satisfy the query either. As a result, large parts of the tree do not need to be traversed at all, saving a lot of time.s5: retrieving at least one subset of 1 -dimensional coordinates from the prefix tree that satisfy the at least one 1 -dimensional constraint and optionally retrieving values associated with the retrieved 1 -dimensional coordinates.

[0170] s6: converting the at least one retrieved subset of 1 -dimensional coordinates back into corresponding N-dimensional coordinates being associated with the measurement data using an inverse of the bijection.

[0171] s7: optionally deserializing at least part of prefix tree (subtree, index or supertree) and optionally the associated values. That is, if the database, i.e. the prefix tree (subtree, index or supertree) and possibly also the associated values have been written to a storage medium, the prefix tree (subtree, index or supertree) and possibly also the associated values are preferably obtained by reading and deserializing them from storage

[0172] s8: optionally decompressing the prefix tree (subtree, index or supertree) and optionally the associated values. That is, if the prefix tree (subtree, index or supertree) and optionally the associated values have been compressed, they are preferably decompressed after deserialization.

[0173] S9: optionally byte-reordering the associated values. That is, if the values were bytereordered before compression, they are preferably byte-reordered after decompression.

[0174] Steps s1-s6 are preferably performed in this order. Steps s7 to s9 might be required to perform step s4 and s5, depending on how step s1 was performed.

[0175] For illustrative purposes only, these steps are now illustrated by means of a numerical example with reference to figures 8 and 9.

[0176] That is, figure 8 schematically illustrates a numerical example of querying a subset of measurement data by traversing a prefix tree in step s2 of the method illustrated in figure 7. Here, a query is made for a single measurement datum with 3-dimensional coordinates x,y,z. The query for the 3-dimensional coordinate can also be expressed in 1 D by applying the same transformation as used when constructing the prefix tree. As the tree is traversed, the nodes in squared boxes 2 and 3 do not satisfy the prefix of the query. As a result, all (grey) nodes to the right of these two nodes do not need to be traversed as their prefix will also not satisfy the query, meaning only 10 nodes are visited instead of all 17 nodes of the tree. Similarly, a query is also made for a single measurement datum with 3-dimensional coordinates a,b,c. In this case only 6 of the 17 nodes of the tree need to be traversed to conclude this datum is not present.Figure 9 schematically illustrates a numerical example of reading measurement data that has been stored to disk after tree splitting and compression. That is, if the measurement data has been written to disk, it is preferably retrieved in several steps. In fact, if the measurement data has been stored in the form of a prefix tree that has been split into a subtree and an index such as a supertree, it is preferred that the supertree is read first. If it was compressed, it needs to be decompressed. The deserialized supertree can then be queried to find subtrees satisfying the query as shown in Figure 8. Finally, all subtrees (in this example only t2) can be retrieved by reading it from the appropriate memory addresses as defined in the supertree and deserialize it (after optional decompression).

[0177] For illustrative purposes only, figure 10 now illustrates a concrete execution of the method on real measurement data and clarifies how the design choices underlying the claimed approach translate into tangible storage benefits. In this example, measurement data from a human cancer cell line acquired on a TimsTOF instrument is used. The instrument combines multiple spectrometric components that each discretize a continuous physical measurement domain at a characteristic resolution: liquid chromatography provides frame indices along retention time, ion mobility separation provides scan indices, and the time-of-flight analyzer provides TOF indices associated with m / z, with the detector recording corresponding ion intensities.

[0178] From the archived raw data, a subset of 267 three-dimensional measurement coordinates is considered, each defined by a frame index, a scan index, and a TOF index, together with an associated intensity value. These coordinates occupy only a small fraction of their enclosing measurement space, which here spans a single frame index, 80 scan indices, and 307 TOF indices. This sparsity is typical for spectrometric data and illustrates the inefficiency of dense storage approaches, which allocate equal space to all dimensions regardless of occupancy. When storing each coordinate component and each intensity value as a 32-bit integer, the resulting representation already requires several kilobytes for this small subset alone.

[0179] Rather than storing these coordinates as independent multidimensional tuples, the example then applies a resolution-defined bijective transformation that maps each three-dimensional coordinate to a single 1 -dimensional integer. The transformation is constructed such that bit positions in the 1 -dimensional integer are assigned to each physical measurement domain in proportion to its effective resolution. In this dataset, the ion mobility domain has the coarsest resolution of roughly 30 scan indices, whereas the retention time and TOF domainhave resolutions of roughly 4 frame / TOF indices. As such, the least three significant bits are occupied by this scan domain and all groups of three bits are occupied by alternating the domains, as their resolutions now all have a comparable magnitude. This layout preserves uniqueness while ensuring that coordinates which are close in physically meaningful, resolution-defined neighbourhoods tend to share long high-order prefixes after transformation. Alternative layouts are possible, but this particular choice strikes a balance between interpretability, efficient storage and efficient querying.

[0180] The resulting 1 -dimensional coordinates are then represented in an octal numeral system and organized into a prefix tree. Because each octal digit corresponds exactly to three bits, and those bits are consistently associated with a single physical domain at each level, changes in prefixes directly reflect refinements along specific measurement dimensions. Coordinates that arise from the same ion species, and therefore form dense, resolution-defined clusters in the original measurement space, naturally follow common branches of the tree and diverge only at deeper levels. As a result, the prefix tree has five major branches, each branch associated with a single isotope as visible in Figure 10.

[0181] The prefix tree is then serialized in a minimal form by encoding, for each node, only the presence or absence of its direct descendants. Owing to the octal representation, each node is recorded as a single byte, as it can have at most 8 descendants. Since the tree is traversed and serialized in lexicographic order, its structure is implicitly encoded without storing explicit pointers or metadata. Importantly, prefixes corresponding to stable, frequently occurring structures appear early and repeatedly in the serialized stream, while fine-grained distinctions appear only near the leaves. As a consequence, the entire serialized prefix tree for the 267 coordinates requires only 546 nodes, and thus equally as many bytes, a small fraction of the size needed to store the raw coordinates themselves.

[0182] Intensity values associated with the coordinates are serialized in the same lexicographic order as the 1 -dimensional coordinates, preserving a strict one-to-one correspondence without additional indexing information. Because values belonging to the same physical structures are placed adjacent in the serialized stream, their numerical similarity is exploited automatically by subsequent compression.

[0183] Finally, both the serialized prefix tree and the serialized values are compressed using a standard general-purpose algorithm. For this data subset, the combined size of the coordinates and intensities is reduced to 644 bytes, which is roughly fifteen percent of thenaive representation. This reduction is not incidental, but directly results from aligning the coordinate transformation, tree structure, and serialization order with the resolution-defined structure inherently present in the measurement data.

[0184] Overall, this example demonstrates how mapping resolution-defined multidimensional coordinates to shared-prefix 1 -dimensional representations enables compact storage. The same structural properties that reduce entropy during serialization also allow queries to traverse and discard entire regions of the data space by inspecting only a small number of prefixes, making the approach scalable for much larger measurement datasets.

Claims

36CLAIMS1. A computer-implemented method of storing measurement data of a measurement system in a database on a non-transitory tangible computer-readable storage medium, the method comprising the steps of:- obtaining M unique / V-dimensional coordinates being associated with the measurement data;- transforming the M unique / V-dimensional coordinates with a bijection to 1- dimensional coordinates, whereby M unique 1 -dimensional coordinates are obtained;- building a prefix tree from the M unique 1 -dimensional coordinates;- serializing the prefix tree; and- writing the serialized prefix tree to the non-transitory tangible computer-readable storage medium.

2. The method according to claim 1, wherein the measurement data is spectrometric measurement data and / or spatial measurement data preferably generated by at least one continuous measurement system, andwherein each N-dimensional coordinate comprises N spectrometric indices and / or N spatial indices acquired from a spectrometric component and / or from a spatial component of the measurement system.

3. The method according to claim 2, wherein each spectrometric component and / or spatial component of the measurement system performs a continuous discrimination of the measurement data along a respective physical measurement domain having an associated resolution and / or tolerance, andwherein the measurement data defines a sparse occupancy of an N-dimensional measurement space while comprising localized structures and / or clusters whose extent along each physical measurement domain is defined by the resolution and / or tolerance associated with the respective physical measurement domain, andwherein preferably at least one physical measurement domain performs time of flight separation of ions.

4. The method according to any one of the preceding claims, wherein in the step of obtaining the M unique / V-dimensional coordinates, values associated with these M unique37coordinates are also obtained, andwherein preferably values of the / V-dimensional coordinates that lie within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space are comparable, andwherein preferably these values are intensity values, andwherein preferably the intensity values represent an ion count.

5. The method according to any one of the preceding claims, wherein the step of transforming the M unique / V-dimensional coordinates with a bijection to 1 -dimensional coordinates comprises the application of at most N component functions to each of the M unique / V-dimensional coordinates, whereby N*M transformed coordinates are obtained, andwherein preferably the transformation comprises the summation of the / V component functions, and / orwherein the bijection is preferably configured such that / V-dimensional coordinates that lie within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space are mapped to 1 -dimensional coordinates that share common high-order digits when expressed in a positional numeral system.

6. The method according to any one of the preceding claims, wherein in the step of building a prefix tree from the M unique 1 -dimensional coordinates each 1 -dimensional coordinate is expressed in base x, the selection of base x determining the branching topology of the prefix tree, such that 1 -dimensional coordinates sharing common high-order digits are represented on a common branch of the prefix tree.

7. The method according to any one of the preceding claims wherein, after the step of building the prefix tree, the prefix tree is pruned, and wherein preferably, the pruning is based on a branch density corresponding to a sparsity pattern and / or density characteristics of the measurement data.

8. The method according to claim 7, wherein noise in the measurement data is discarded based on a local sparsity criterion, and / orwherein coordinates of the M unique 1-dimensional coordinates are not considered noise when they lie within a resolution-defined and / or tolerance-defined neighbourhood in the N-dimensional measurement space.

9. The method according to any one of the preceding claims, wherein, prior to serializing the prefix tree, the prefix tree is split into at least one subtree, and wherein each of the subtrees provides an independent prefix tree, andwherein, after splitting the prefix tree into the at least one subtree, the at least one subtree is preferably indexed, preferably with another prefix tree.

10. The method according to claim 9, wherein the measurement data is split into said subtrees such that different structures and / or clusters present in the measurement data is mapped to different branches of the prefix tree and thereby to different subtrees, and / or wherein the at least one subtree is indexed such that, when querying a subset of the measurement data with at least one query, only those subtrees whose prefixes are consistent with the query can be accessed.

11. The method according to claim 10, wherein said structures and / or clusters correspond to localized, resolution-defined and / or tolerance-defined neighbourhoods in the N-dimensional measurement space, such that 1 -dimensional coordinates belonging to a same structure and / or cluster are predominantly contained within a single subtree.

12. The method according to any one of the preceding claims, wherein in the step of serializing the prefix tree, the serialization comprises the steps of encoding, for each node of the prefix tree, only the presence or absence of its possible direct descendants as bitflags, and concatenating the resulting encodings in an order consistent with a lexicographic traversal of the 1 -dimensional coordinates.

13. The method according to any one of the preceding claims, wherein, prior to writing to the non-transitory tangible computer-readable storage medium, the serialized prefix tree is compressed.

14. The method according to any one of the preceding claims wherein values associated with the coordinates are serialized by reordering the values such that their associated 1D coordinates are sorted lexicographically, andwherein, prior to writing to the non-transitory tangible computer-readable storage medium, the values associated with the coordinates are compressed, preferably by a compression algorithm wherein the ordering of the values induced by the prefix tree groups values associated with coordinates belonging to the same structure and / or cluster, and wherein, preferably, during serialization of the values, positional bytes from eachvalue are concatenated.

15. A computer-implemented method of querying a database for a subset of measurement data, wherein the database comprises measurement data of at least one measurement system and has been provided by the steps of:- obtaining M unique / V-dimensional coordinates being associated with the measurement data;- transforming the M unique / V-dimensional coordinates with a bijection to 1- dimensional coordinates, whereby M unique 1 -dimensional coordinates are obtained;- building a prefix tree from the M unique 1 -dimensional coordinates; and wherein the method of querying the database comprises the steps of:- receiving at least one query defining at least one N-dimensional constraint;- transforming the N-dimensional constraint, using the same bijection, into at least one corresponding 1 -dimensional constraint;- querying the prefix tree with the at least one 1 -dimensional constraint;- retrieving at least one subset of 1 -dimensional coordinates from the prefix tree; and - converting the at least one retrieved subset of 1 -dimensional coordinates back into corresponding N-dimensional coordinates being associated with the measurement data using an inverse of the bijection.

16. The method according to claim 15, wherein the measurement data is spectrometric measurement data and / or spatial measurement data preferably generated by at least one continuous measurement system, andwherein each N-dimensional coordinate comprises N spectrometric indices and / or N spatial indices acquired from a spectrometric component and / or from a spatial component of the measurement system.

17. The method according to claim 16, wherein each spectrometric component and / or spatial component of the measurement system performs a continuous discrimination of the measurement data along a respective physical measurement domain having an associated resolution and / or tolerance, andwherein the measurement data defines a sparse occupancy of an N-dimensional measurement space while comprising localized structures and / or clusters whose extent along each physical measurement domain is defined by the resolution and / or tolerance associated with the respective physical measurement domain, andwherein preferably at least one physical measurement domain performs time of flight separation of ions.

18. The method according to any one of claims 15 to 17, wherein, after building the prefix tree, the prefix tree is split into at least one subtree, and wherein each of the subtrees provides an independent prefix tree, andwherein, after splitting the prefix tree into the at least one subtree, the at least one subtree is indexed, preferably with another prefix tree.

19. The method according to claim 18, wherein the measurement data is split into said subtrees such that different structures and / or clusters present in the measurement data is mapped to different branches of the prefix tree and thereby to different subtrees, and / or wherein the at least one subtree is indexed such that, when querying a subset of the measurement data with the at least one query, only those subtrees whose prefixes are consistent with the query can be accessed.

20. The method according to claim 19, wherein said structures and / or clusters correspond to localized, resolution-defined and / or tolerance-defined neighbourhoods in a N-dimensional measurement space defined by the measurement data, such that 1-dimensional coordinates belonging to a same structure and / or cluster are predominantly contained within a single subtree.

21. The method according to any one of claims 15 to 20, wherein the at least one N-dimensional constraint comprises at least one fixed coordinate and / or at least one coordinate range defined with respect to the resolution and / or tolerance associated with at least one physical measurement domain, and / orwherein in the step of transforming the N-dimensional coordinates, the 1-dimensional constraint comprises at least one prefix and / or at least one 1D interval, such that, in the step of querying the prefix tree, only those parts of the prefix tree whose prefixes are consistent with the 1 -dimensional constraint are traversed.

22. The method according to any one of claims 15 to 21, wherein, after building the prefix tree, the provision of the database further comprises the steps of serializing the prefix tree and writing the serialized prefix tree to a non-transitory tangible computer-readable storage medium, andwherein the method of querying the database further comprises the step of reading41and deserializing the prefix tree before querying.

23. The method according to any one of claims 15 to 22, wherein, in the provision of the database, values associated with the M unique N-dimensional coordinates are also obtained and serialized in an order corresponding to a lexicographic ordering of the associated 1 -dimensional coordinates, andwherein the method of querying the database further comprises retrieving the values associated with the retrieved 1 -dimensional coordinates by using an index of a retrieved 1 -dimensional coordinate in said lexicographic ordering to identify the associated value stored at a same index position.

24. The method according to any one of the preceding claims, wherein the measurement data is at least one of:- at least one of mass spectrometry (MS) data, mass spectrometry imaging (MSI) data, liquid chromatography (LC) data, trapped ion mobility separation (TIMS) data, time-of-flight (TOF) data or combinations thereof, or- obtained from the at least one measurement system while performing at least one measurement or obtained from at least one archived format.

25. A computer program product, comprising a non-transitory tangible computer-readable storage medium whose contents cause a processor to perform the method according to any one of claims 1 to 24.

26. A measurement system, preferably a mass spectrometer optionally combined with a liquid chromatography (LC) unit and / or an ion mobility separator (IMS), wherein the measurement system comprises a processor comprising instructions which, when carried out by the processor, causes the processor to carry out the method according to any one of claims 1 to 24.