Determination of stability of substance or substance mixture

Abstract

The invention relates to a computer-implemented method for determining the stability of a substance or a substance mixture. In one embodiment, the method comprises the following steps: a data detection step (102) in which at least one measurement data set is received, each measurement data set representing a chemical property, in particular a phytochemical property, of a respective substance or substance mixture; a data evaluation step (106) comprising, for each measurement data set: determining (112) a starting value for the respective substance or substance mixture based on the measurement data set; and quantizing (114) a change in the respective substance or substance mixture relative to the initial value by means of a mathematical distance measure; and a data output step (108) in which, for each measurement data set, a change in the respective substance or substance mixture is graphically presented.

Description

Technical Field

[0001] The present invention relates generally to the technical field of analysis, and in particular to pharmaceutical analysis. Background Art

[0002] For example, the development of new products in the pharmaceutical, cosmetic, and food sectors faces an increasing number of requirements and stringent quality standards. Stability studies play a particularly important role in such developments, as they can cover a wide range of possible influencing factors (e.g., chemical or physical incompatibilities between molecules in the product, resistance to temperature and / or humidity, resistance to UV / VIS radiation, oxidation, interactions between different reactants, microbial contamination, biodegradation, etc.).

[0003] Each parameter in a product can only be characterized by a wide range of different technical methods based on its response to these and other factors. For example, it often requires multiple different HPLC methods (high performance liquid chromatography) for different analyses, wet chemistry colorimetric reactions, moisture determination, and physical parameter testing (e.g., breaking strength, friability, and disintegration time of tablets, flowability, bulk and tap density of powders, turbidity measurement of liquids, etc.).

[0004] To be able to characterize as many of these potential interactions as possible, it's common practice to plan a project using a "Design of Experiments" (DoE, synonymous with Statistische Versuchsplanung). This typically takes into account not only all possible qualitative components but also the quantitative proportions of the substances present. Furthermore, consideration may be given, for example, to which packaging material (primary and / or secondary or other) will best protect the product from variations. Packaging material refers to the material used to package the product. If all of these aspects are included in the test plan, the number of product variants will be multiplied by the number of packaging material options to be tested.

[0005] A distinction is usually made between primary and secondary packaging. Primary packaging for drugs and medicinal products includes containers or components made of glass, rubber, plastic, aluminum, composite materials and films. These materials are in direct contact with the medicinal product and must therefore meet certain requirements regarding safety, effectiveness and reliability. Primary packaging manufacturers must meet the expectations of the pharmaceutical manufacturer and be able to demonstrate that their production processes comply with the rules of an integrated quality management system (QMS) and Good Manufacturing Practice (GMP) and thus meet the required quality standards. Secondary packaging materials are outer packaging that has no direct contact with the items to be packaged or the medicinal product or other substances and mostly has a protective and control function.

[0006] This results in a project design involving a large number of samples and numerous analytical methods, ideally requiring comprehensive implementation for all samples. However, because this experimental design can yield significant information gains, its implementation is desirable to ultimately identify the optimal product with the longest product lifespan. However, the labor and equipment costs required using traditional analytical methods are prohibitive. Such extensive studies can only be undertaken with considerable difficulty using a given pool of staff and equipment, as the required time and financial investment would be prohibitive and would significantly extend product development timelines. The alternative is to significantly reduce measurement costs, but this could result in unacceptable product quality.

[0007] Besides the immense effort required to collect large datasets in stability studies, interpreting such vast data sets presents a significant challenge. This is because "classic" raw data analysis, which typically evaluates only a small subset of the total data (e.g., examining only the three strongest signals in an HPLC chromatogram rather than the entire "fingerprint range"), can mask or fail to capture weak trends or anomalies. Modern chemometric methods can, in principle, offer a remedy. These include, for example, principal component analysis, correlation analysis, distance metrics, partial least squares regression, support vector machines, and neural networks. Such methods are able to transform complex raw datasets containing many variables (here, the test parameters) into a small number of latent variables that still accurately describe the dataset. Therefore, implementing chemometric methods to simply and efficiently evaluate datasets to address these issues is essential. Due to the aforementioned personnel, financial, and time constraints, product stability testing currently involves fewer parameters. This can lead to the failure to detect certain other unmeasured parameters that may affect stability, potentially resulting in low-quality products. Therefore, it is highly beneficial to characterize products from a holistic stability perspective, rather than based on a single, selected parameter that may not accurately reflect the overall stability of the substance or mixture of substances that the product or its packaging is composed of. Consequently, important parameters may be overlooked, resulting in suboptimal products. The present invention addresses these issues by comprehensively testing the properties of a product, substance, or mixture of substances (hereinafter referred to as a "mixture of substances") and enabling specific conclusions to be drawn about the stability of the entire product. Therefore, the present invention eliminates the need to analyze multiple parameters.

[0008] Therefore, the object of the present invention is to provide a method for determining the stability of a substance mixture in order to overcome the above-mentioned disadvantages of the prior art. In addition, the calculation method can also be used for monitoring processes in which multiple monitoring parameters are detected in parallel. Summary of the Invention

[0009] The present invention includes a computer-implemented method for determining the stability of a substance mixture and for monitoring a process for detecting multiple monitoring parameters in parallel. The method includes at least one data detection step, in which at least one measurement data set is received, and at least one temporally consecutive data detection step, in which at least one further measurement data set is received. The first data detection step includes at least one starting value measurement data set. Each measurement data set can represent the chemical characteristics, in particular the phytochemical characteristics, of the corresponding substance mixture. The method can also include a data evaluation step. The evaluation is to process the (raw) data of the experiment using (raw) data consisting of at least one starting value measurement data set and at least one further measurement data set to generate specific knowledge gains. The data evaluation step includes, for each measurement data set, determining the starting value of the corresponding substance mixture based on the measurement data set. In addition, the data evaluation step also includes quantifying the change of the corresponding substance mixture relative to the starting value by a mathematical distance metric, which is performed by detecting at least one further temporally consecutive data detection step, using the data detection step to detect at least one further measurement data set. The method can also include a data output step, in which, for each measurement data set, the change of the corresponding substance mixture is graphically represented.

[0010] The substance mixtures according to the invention include both individual substances and combinations or substance mixtures as well as chemical and / or biological products. The terms "substance or substance mixture", "combination" and "product" are to be understood as equivalent terms within the scope of the present invention.

[0011] In the sense of the present invention, the stability of a substance mixture refers to a measure of the time period over which the changes in the corresponding substance mixture remain within predefined limit values. This includes in particular chemical and / or physical stability. In a stability study, stability can be checked according to the method of the present invention. Therefore, a stability study is understood to be an experiment to check whether different substance mixtures are generally stable or the speed at which their stability decreases. As an alternative or in addition, it can be provided that different substance mixtures, in particular formulations, are compared with each other within the scope of such an examination. Stability comprises changes in a substance or substance mixture over a time period measured between at least two time points. The first time point comprises a first data detection step with at least one starting value measurement data set, and the second time point comprises a second and all other data detection steps with at least one additional measurement data set.

[0012] In addition, the method can also be used to monitor production processes or reactions (see Application Example 3). These include chemical or biological production processes and reactions, such as extraction, tableting, capsule filling, and mixing processes. For example, the equilibrium state in an extraction process can be monitored using the new method and the time point of the equilibrium state can be determined without prior calibration of the method. In this method, the equilibrium state is displayed by the calculated distance value asymptotically approaching a plateau, such as Figure 8 and Figure 9 shown.

[0013] The chemical profile of a substance mixture refers to the totality of the chemical compounds in the substance mixture and / or the chemical properties or reactivity of the substance mixture based thereon. The phytochemical profile of a substance mixture refers to the totality of the chemical compounds of the plant components in the substance mixture and / or the properties of the substance mixture based thereon.

[0014] The aforementioned challenges and requirements define the criteria that new analytical concepts must meet. The method according to the present invention allows rapid and extensive investigation of parallel substance mixtures and their comparison with one another, even over time. This enables advantageous relative comparisons of product variants and, therefore, the selection of promising candidates, for example, in pharmaceutical research.

[0015] In the context of the present disclosure, drugs refer to: (a) all substances or combinations of substances / mixtures that are drugs with therapeutic or preventive properties for human or animal diseases, or (b) all substances or combinations of substances / mixtures that can be used in or on the human or animal body or can be infused into a human or animal in order to restore, correct or affect the physiological functions of a human or animal or to make a medical diagnosis through pharmacological, immunological or metabolic effects.

[0016] A particular advantage of the method according to the present invention is that during product development, a significantly reduced selection of possible product candidates can be generated with minimal effort, so that, for example, only these candidates subsequently require further investigation using other specialized classical analytical methods. By focusing on certain promising product candidates, more detailed knowledge can be efficiently acquired to ultimately select the optimal product variant. In particular, the proposed method according to the present invention allows for analytical processing of large-scale product comparisons within a reasonable timeframe and with very low personnel and equipment requirements. The method according to the present invention is particularly advantageously applicable to complex mixtures of substances, but is also applicable to products with small numbers of individual components and compounds.

[0017] By comprehensively evaluating complex measurement data sets that represent as complete a (phyto)chemical profile as possible for all samples in the stability study, the method according to the invention makes it possible to identify those substance mixtures that have varied the least and show the smallest changes compared to their starting values ​​and can therefore be considered the most stable.

[0018] It is preferably provided that the method, in particular the data evaluation step, comprises the use of a machine learning model ("machine learning model"), which is preferably generated or trained by unsupervised ("unsupervised") and / or supervised machine learning ("supervised machine learning"). Unsupervised machine learning is a mathematical or information technology method in which a data processing device / computer processes data without external additional information and attempts to find a solution independently. An example of this is principal component analysis (abbreviated PCA). Supervised machine learning is a mathematical or information technology method in which a data processing device / computer performs processing with the aid of external information (e.g. training data, in particular concentration information used in the experiment). In this case, the independent search for a solution is usually carried out after calibration with known data.

[0019] It is particularly preferably provided that the at least one measurement data set comprises data obtained by near-infrared spectroscopy.

[0020] By using near infrared spectroscopy (NIR), samples of mixtures of substances can be measured without complex sample processing or destruction to generate a measurement data set. Near infrared spectroscopy, abbreviated as NIR spectroscopy or NIRS, is a physical analysis technique based on spectroscopy in the short-wave infrared range. In NIRS, detection is preferably performed in the range of about 760-2500 nm (equivalent to about 13160 cm -1 Up to 4000cm -1 ) is carried out in the near-infrared range of the atmosphere. This technique detects the combined vibrations and overtone vibrations of all molecules in the sample and is therefore able to detect them in their entirety. In order to exploit these data, mathematical methods (chemometrics) are used in particular, since the information contained in these spectra (such as the presence / absence, concentration, etc. of certain chemical compounds) is usually hidden from the observer. Near-infrared spectroscopy can also be advantageously applied to liquid samples (for example to measure transmittance, i.e. a measuring light beam passes through the sample over a defined layer thickness and the unabsorbed light is captured by a detector; or to measure transreflectance, i.e. a measuring light beam impinges on the sample. The reflected light is detected by a detector. The absorbance is given by the difference between the incident and reflected light) and to solid samples (for example to measure diffuse reflectance or transmission).

[0021] One aspect of the present method is to provide for a machine-based and reproducible evaluation of measurement datasets. Datasets obtained, for example, using near-infrared spectroscopy have the advantage of containing more information per observation date than datasets obtained using other analytical techniques, as conventional analytical techniques only observe a few isolated signals simultaneously. Near-infrared spectroscopy allows for a more comprehensive evaluation of a wide range of signals.

[0022] It is preferably provided that at least one measurement data set additionally or alternatively comprises: data obtained by UV / VIS spectroscopy, data obtained by Raman spectroscopy, a (U)HPLC fingerprint, a GC fingerprint, a peak table from chromatography, and / or at least one physical, biological or chemical parameter, in particular sugar content, decomposition rate, color, breaking strength, disintegration time, friability, density, viscosity, refractive index and / or optical rotation angle.

[0023] UV / VIS photometry refers to a spectroscopic method belonging to optical molecular spectroscopy, which uses ultraviolet (UV) and visible (VIS) electromagnetic waves, wherein the light sources of UV and visible light emit in the wavelength range from about 200 nanometers to about 800 nanometers.

[0024] In Raman spectroscopy, the material to be investigated is illuminated with monochromatic light, preferably a laser. In the spectrum of the scattered light from the sample, other frequencies are observed in addition to the incident frequency (Rayleigh scattering). The frequency differences from the incident light correspond to the material's characteristic energies, such as rotational, vibrational, phonon, or spin-flip processes. The resulting spectrum allows conclusions to be drawn about the substance being investigated.

[0025] (U)HPLC (ultra-high performance liquid chromatography) is a standard analytical method (liquid chromatography) that can not only separate substances but also identify and quantify them using standard samples. (U)HPLC can also analyze non-volatile substances. It can be connected to sensitive detection systems such as mass spectrometry (MS or the combined term UHPLC MS / MS).

[0026] In the analysis, the results can be fully recorded as patterns ("fingerprints") These "fingerprints" can be compared with databases or results from other analyses using the same method to identify the substance being analyzed.

[0027] The method according to the invention advantageously allows for combining and integrating different analytical data sets into the evaluation. For example, at least one measurement data set of a substance mixture can include data sets obtained, for example, by UV / VIS spectroscopy, Raman spectroscopy, (U)HPLC fingerprints (particularly via coupling with various detectors, such as diode array detectors (DADs), refractive index detectors, electrochemical detectors, UV / VIS detectors, mass spectrometry detectors (MS), light scattering detectors (ELSDs), GC fingerprints (particularly coupled with various detectors, such as flame ionization detectors (FIDs), MS), or generally peak tables from all conceivable chromatographic methods. In other words, the method according to the invention can be applied to a variety of measurement techniques that assign one or more X values ​​(e.g., wavelength, m / z ratio (mass / charge ratio), retention time, measurement point) to one or more Y values ​​(e.g., intensity, absorbance, voltage, measured value). Therefore, the method can be used in a particularly versatile and flexible manner.

[0028] Furthermore, multidimensional datasets can be processed in the corresponding analytical methods (e.g., HPLC-DAD, HPLC-MS). Furthermore, large data matrices of test parameters (e.g., physical, biological, or chemical analyses, but not limited thereto) collected separately (i.e., using different techniques) can be combined for evaluation and subjected to new methods. For example, measurement results of sugar content, disintegration rate, tablet color, and breaking strength, as datasets measured at multiple time points, can be combined into a data matrix and evaluated.

[0029] It can also be provided that the quantification of the changes in the various substance mixtures is based on the totality of the components of the various substance mixtures.

[0030] By considering the totality of all components of the substance mixture being investigated, sustainable results can be generated in a particularly simple manner. In stability studies, all spectral changes of all components relative to the starting value should be specifically considered. Unlike conventional marker analysis (e.g., by HPLC), this allows for a holistic view of all components of a complex mixture (e.g., by near-infrared spectroscopy), thus reflecting fundamental changes in the entire sample.

[0031] The variation of material mixtures can be calculated using mathematical distance metrics.

[0032] Preferably, provision is made for the mathematical distance measure to be selected from the following group: Euclidean distance, Mahalanobis distance, Manhattan distance, Pearson distance and / or Gower distance.

[0033] By selecting a mathematical distance metric from the aforementioned group, it is possible to quantify the changes in the substance mixture being studied over time, while fully handling the complexity of the underlying dataset. Thus, choosing a suitable mathematical strategy to quantify the extent of sample changes over time allows for efficient quantification of changes in the corresponding substance mixture. In particular, it is possible to obtain a very good comparative overview of stability study datasets without having to examine many individual parameters separately.

[0034] It is particularly advantageously provided that the mathematical distance measure can be selected by the user.

[0035] By allowing the user to select the mathematical distance metric, the method of the present invention becomes more flexibly usable and adjustable. In particular, the user can use his or her experience to select an appropriate mathematical distance metric for determining the stability of a substance mixture using the method of the present invention.

[0036] In particular, it can be provided that the user can iteratively select multiple mathematical distance metrics in order to quantify the change in the respective substance mixture, wherein the quantification is performed based on each selected mathematical distance metric. This allows the user to compare the results of different selectable mathematical distance metrics with one another. This makes the method for determining the stability of a substance mixture particularly flexible and versatile. Providing this possibility also makes it possible to modify the unintentional selection of mathematical distance metrics for quantifying the change in various substance mixtures.

[0037] Preferably, the method further comprises a data preprocessing step, which may include performing scattered light correction on the at least one measurement data set, particularly when the data set includes near-infrared measurement data. Furthermore, the data preprocessing step may include centering, normalizing, and / or scaling the at least one measurement data set. Furthermore, the data preprocessing step may include performing a principal component analysis.

[0038] As used herein, data preprocessing refers to the mathematical manipulation of raw data in preparation for the actual evaluation. For example, it can include correction for scattered light, increasing the signal-to-noise ratio, or simply reformatting the detected measurement data set or raw data so that the input data can be correctly processed by the algorithm in the downstream evaluation. Appropriate data preprocessing can be particularly useful for improving the quality of the results of the method of the present invention. Data preprocessing is preferably performed after the data detection step of the method of the present invention.

[0039] By appropriate data preprocessing, which can include, in particular, data volume reduction, any possible parameter interactions in the complex overall data set can be automatically captured and taken into account. In other words, the accuracy or estimation quality of the method according to the present invention can be improved, in particular, by performing a scattered light correction (e.g., near-infrared spectral data) before the actual data evaluation.

[0040] Data normalization or scaling, along with upstream principal component analysis, can also remove unwanted noise from the dataset information, further improving the quality, validity, and reliability of the conclusions. Noise refers to interference with a broad, nonspecific spectrum that can overlap or mask the desired information-bearing signal. An example of noise is unwanted scattered light, which randomly falls on the detector of a near-infrared spectrometer along with the desired excitation light and is measured in addition to it. Applying specific mathematical techniques can help separate the noise from the usable information, thereby increasing the signal-to-noise ratio and making the assessment more robust and reliable.

[0041] It can also be provided that the data output step includes: presenting the changes in the corresponding substance mixture as a box plot; and / or presenting the mean, median, 0.25 / 0.75 quantiles, highlighting possible outlier candidates; and / or performing at least one statistical test, in particular a t-test, a Wilcoxon rank sum test, a one-way analysis of variance and / or a Kruskal-Wallis test.

[0042] This design of the data output step allows for the simultaneous investigation of very large measurement data sets, particularly across multiple samples, in a short time and with increased interpretability. Advantageously, this also provides intuitive evaluation. This means that the technology can be used by users without advanced mathematical knowledge, and no separate programming is required for each investigation. For example, in the data output step, the user can output the variations in the investigated substance mixture as a boxplot, providing a simple and intuitive overview. User-friendliness is further enhanced by highlighting the mean, median, 0.25 / 0.75 quantiles, and outliers. For example, the measurement data set detected by the user can be acquired directly from the measuring instrument or introduced into the method according to the aforementioned data preprocessing step, and an easily interpretable evaluation can be immediately generated.

[0043] Furthermore, it can be provided that the determination of the starting value for the respective substance mixture is based on metadata of the respective measurement data set.

[0044] By using metadata when determining the starting value, investigation time or processing time can be advantageously saved. The starting value is determined in a very short time based on the metadata.

[0045] It is preferably provided that the data evaluation step further comprises performing an additional principal component analysis on the measurement data set, which analysis does not involve quantifying the variations by means of a mathematical distance measure, wherein the result of the additional principal component analysis is displayed in the data output step.

[0046] The additional principal component analysis, especially the qualitative analysis that is not included in the distance calculation, supports the user in evaluating the output results after the data export step. Thus, during the data export step, the user receives both a quantitative analysis, especially including distance metrics, and a qualitative analysis, enabling further conclusions to be drawn clearly and intuitively.

[0047] It can be provided in particular that the substance mixture comprises a solid and / or liquid and / or gaseous substance mixture.

[0048] Furthermore, it may be provided that the substance mixture includes biological, chemical, plant, animal, human substances or substance mixtures, pharmaceutical compositions, phytodrugs, chemical and / or biological drugs, cells, cell therapy products (e.g. gene therapy products, such as CAR T cells (chimeric antigen receptor T cells)), NK cells (natural killer cells), somatic cell therapy products, biotechnologically processed tissue products / tissue engineering products, tissues, stem cells, stem cell products or preparations, such as CD34+ cells, CD19+ cells, CD20+ cells, HEK295 cells, TCR alpha / beta cells, TCR gamma / delta cells, CD3+, CD4+, CD8+, CD133+ cells), blood, blood products, organs, medicinal teas, extracts, in particular verbena extracts, drops, tablets, coated pills, capsules, powders, granules, solutions, suspensions, slurries, foods, in particular meat or minced meat, fruit juice, in particular orange juice, nutritional supplements, cosmetics, lotions, ointments and / or creams, as well as packaging, packaging materials, films, in particular polyethylene, polyvinyl chloride, etc.

[0049] The method for determining the stability of a substance mixture is applicable to a wide range of situations. Thus, a substance mixture can advantageously be studied in any aggregation state.

[0050] The computer program according to the invention comprises instructions which, when the program is executed on a computer, cause the program to carry out the above-described method.

[0051] The device according to the present invention can be, in particular, a measuring instrument or a server-computer and include means for implementing the above-described method. Furthermore, a mobile electronic device, such as a smartphone, a programmable memory controller, or a so-called "edge device," is also conceivable as a device according to the present invention. Furthermore, the method can be provided as a "Software as a Service" cloud solution or, in general, as a "serverless" application.

[0052] The technical advantages and designs described in conjunction with the method of the present invention are also applicable to the computer program of the present invention and the device of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Preferred embodiments of the present disclosure are described below with reference to the following drawings:

[0054] Figure 1 A general overview of possible steps of a method according to an embodiment of the invention is shown schematically.

[0055] Figure 2 An overview of possible steps of a method according to an embodiment of the invention is shown schematically.

[0056] Figure 3 Shown is a spectral overlay of a measured measurement data set of a near-infrared spectrometer with desaturation of the detector.

[0057] Figure 4 Shown are spectral overlays of measured measurement data sets of a near-infrared spectrometer without data preprocessing (top) and with exemplary data preprocessing (scattered light correction) (bottom).

[0058] Figure 5 Boxplot showing a comparison of the resulting Euclidean distances calculated based on the NIR spectra of orange juice stored for several days at room temperature (="RT") or 40°C (="40C") .

[0059] Figure 6 Shown are the mean values ​​+ / - standard errors ("SW" = starting value) of the resulting Euclidean distances calculated based on NIR spectra of ground meat samples stored for several days at room temperature (="RT") or 40°C (="40C").

[0060] Figure 7 The image of a minced meat sample at 40°C and room temperature (RT) shows changes over time. At 40°C, a clear change in the sample was detected (1). At room temperature, the marbling of the minced meat was still clearly preserved (2 and 3). However, using the method of the present invention, this qualitative decay can be more accurately quantified and tracked.

[0061] Figure 8 Shown are box plots comparing the resulting Euclidean distances calculated based on NIR spectra obtained from verbena extraction experiments performed within a limited time of 180 minutes.

[0062] Figure 9 Show the Figure 8 A more detailed kinetic evaluation of the data was performed, in which in particular the experimental rate constants (distance max and k) were determined and from which estimated times for 50%, 90% and 95% extraction were derived.

[0063] Figure 10A more detailed kinetic evaluation of the cell culture experimental data is shown, here of untreated HEK295 cells without additional stressor (= control). The time to reach 50% of the plateau value (K m ), 90% and 95% of the time.

[0064] Figure 11 A more detailed kinetic evaluation of the data from cell culture experiments under the influence of 3% ethanol solution added as a stress agent is shown. The time required to reach 50% of the plateau value (K m ), 90% and 95% of the time. Different survival rates and kinetics (see also Figure 12 ) can be correlated well here with preliminary experiments (Example 4), in which more concentrations were tested.

[0065] Figure 12 A more detailed kinetic evaluation of the data from cell culture experiments under the influence of 6% ethanol solution added as a stress agent is shown. The time required to reach 50% of the plateau value (K m ), 90% and 95% of the time.

[0066] Figure 13 A summary of the plateau values ​​determined from the cell culture experiments is shown. The distance values ​​determined for the 6% ethanol method are the largest, demonstrating the strong effect of this high concentration on cell culture. Therefore, this measurement method is considered the "least stable." The values ​​for the 3% and control methods are correspondingly lower. These results are also reflected in the previously performed cell viability tests.

[0067] Figure 14 The correlation between cell viability and the ethanol concentration (percentage) used in the preliminary experiment (A), the maximum observed Euclidean distance and the ethanol concentration (percentage) used in the main experiment (B), and the correlation between cell viability and the maximum observed Euclidean distance (C) are shown, linking the two analysis techniques. For each subplot (A), (B), and (C), a linear trend line with an R of 0.955, 0.999, or 0.999, respectively, has been added. 2 .

[0068] Figure 15 Boxplot showing the Euclidean distance of NIR measurements of medicinal tea formulations under different storage conditions (refrigerator (KS) / room temperature (RT) / climate cabinet at 40°C).

[0069] Figure 16 Comparison of stability measurements of medicinal tea formulations is shown. The measurements were performed using a validated reference method (photometric test in the UV / VIS range, referred to as GPP in the legend) and NIR using the new evaluation method of the present invention.

[0070] Figure 17 An example graphical display of a raw HPLC / MS data set from a 20% ethanol extraction of a medicinal plant mixture consisting of thyme, rosemary, and chamomile. These raw data were directly processed by the novel evaluation method.

[0071] Figure 18 Shows that Figure 17 Boxplots of Euclidean distances calculated for the HPLC / MS datasets extracted with 20% ethanol. Euclidean distances are plotted against weeks (observation period: 24 weeks) or, on the far right, after light treatment (SunT). Legend: KII = Climate Zone II; AC = Climate Zone AC; B = brown glass bottle; W = white glass bottle.

[0072] Figure 19 Show the basis Figure 17 Fitting of the determined kinetic function to the Euclidean distance calculated from the HPLC / MS raw data of the 20% ethanol extract of [Eta](R)]. Legend: KII = Climate Zone II; AC = Climate Zone AC.

[0073] Figure 20 Shown are the stability curves of a 20% extract of a medicinal plant mixture consisting of thyme, rosemary and chamomile against a representative single selection marker compound (m / z 329.17; retention time = 8.16 minutes) under different climatic conditions.

[0074] Figure 21 Boxplot showing the Euclidean distance of a medicinal plant powder mixture consisting of rosemary, thyme and chamomile before (time 0) and after 20 hours of irradiation with light (time 20). Legend: P1 = powder mixture packaged in paper bags; P2 = powder mixture stored open.

[0075] Figure 22 A bar graph showing reference data for wet-chemical photometric measurement of total polyphenols based on entry 2.8.14 of the European Pharmacopoeia. A mixture of medicinal plant powders was irradiated with light for 20 hours, either packed in paper bags (P1) or stored exposed (P2). The recovery is expressed as the percentage correspondence between the absorbance after irradiation and the starting value of the unloaded sample.

[0076] Figure 23 Shown is the fitting of the determined kinetic function to the Euclidean distance calculated based on NIR spectra (corrected for SNV) to the latent variables during 24 weeks of observation of a mixture of medicinal plant powders in different climate zones. Legend: KII = Climate Zone II; AC = Climate Zone AC

[0077] Figure 24Shown are the stability curves of the medicinal plant powder mixture under different climatic conditions for a representative single selection marker compound (m / z 329.17; retention time = 8.16 minutes).

[0078] Figure 25 Shown is the change in appearance of a medicinal plant powder mixture after storage in different climate zones for 24 weeks. The samples of the mixture stored in climate zones AC were noticeably darker in color.

[0079] Figure 26 Boxplot showing Euclidean distances calculated from a Hawthorn film sheet NIR dataset (SNV correction). Euclidean distances are plotted against weekly values ​​(observation period: 24 weeks). Legend: FT = film sheets in polyethylene bags; PP = film sheets in blister packs and folding boxes.

[0080] Figure 27 The stability curves of hawthorn film sheets for a representative dimeric proanthocyanidin (m / z 577.14; retention time = 2.69 minutes) under different climatic conditions and different packaging materials are shown. Legend: FT = film sheets in polyethylene bags; PP = film sheets in blister packs and folding boxes. DETAILED DESCRIPTION

[0081] like Figure 1 As shown, data detection 102 is first performed. Here, at least one measurement data set is received, which can be generated by measuring one or more measuring devices. Each measurement data set includes the chemical characteristics of the substance mixture to be studied, in particular the phytochemical characteristics. The measurement data set or multiple measurement data sets may in particular include data obtained by near-infrared spectroscopy. By near-infrared spectroscopy (NIR), samples of substances or substance mixtures can be measured without complex sample processing to generate measurement data sets. This technology detects the combined vibrations and overtone vibrations of all molecules in the sample and is therefore able to detect them in their entirety. The advantage of data sets obtained, for example, by near-infrared spectroscopy (NIR) is that these data sets contain more information in the observation date compared to data sets obtained by other analytical techniques, because traditional analytical techniques only observe a few isolated signals at the same time. In other words, near-infrared spectroscopy enables a more extensive evaluation of various signals.

[0082] Alternatively or additionally, the measurement data set or data sets may include data detected by UV / VIS spectroscopy, Raman spectroscopy, (U)HPLC fingerprints (particularly from coupling with various detectors, such as DAD, MS, ELSD), GC fingerprints (particularly from coupling with various detectors, such as FID, MS), or generally from peak tables obtained from all conceivable chromatographic methods. In other words, the method can be applied to various measurement techniques that assign one or more X values ​​(e.g., wavelength, m / z ratio, retention time, measurement point) to one or more Y values ​​(e.g., intensity, absorbance, voltage, measured value). The method can therefore be used in a particularly versatile and flexible manner.

[0083] In a further optional step, data preprocessing 104 can be performed. This can include, in particular, performing a scattered light correction on at least one measurement data set, in particular when the data set includes near-infrared measurement data; performing a centering, normalization and / or scaling on at least one measurement data set; and / or performing a principal component analysis. Appropriate data preprocessing can be used in particular to improve the quality of the results of the method according to the invention. By means of data preprocessing, which can in particular include a reduction in the data volume, possible parameter interactions in the complex overall data set are automatically captured and taken into account. In other words: by performing a scattered light correction, in particular before the actual data evaluation (for example in the case of near-infrared spectral data), the accuracy of the method according to the invention or the quality of the evaluation can be improved.

[0084] In the next step, data evaluation 106 is performed for each measurement data set. Here, a start value calculation 112 and a distance metric calculation 114 are performed for each measurement data set. In the start value calculation 112, a start value is determined based on the measured measurement data, from which changes in the substance mixture to be investigated are observed and quantified within the scope of the distance metric calculation 114. The distance metric calculation 114 can be performed, in particular, for all components of the substance mixture to be investigated in order to increase the robustness and validity of the results.

[0085] The distance metric used in distance metric calculation 114 can be selected, in particular, from the group consisting of: Euclidean distance, Mahalanobis distance, Manhattan distance, Pearson distance, and / or Gower distance. By selecting a mathematical distance metric from this group, it is possible to quantify the changes in the substance or substance mixture being studied over time, while also being able to comprehensively account for the complexity of the underlying dataset. Thus, selecting a suitable mathematical strategy for quantifying the extent of sample changes over time allows for efficient quantification of the changes in the respective substance or substance mixture. In particular, this allows for a very good comparative overview of the stability study dataset without having to examine many individual parameters individually.

[0086] It is particularly advantageously provided that the mathematical distance measure can be selected by the user.

[0087] By allowing the user to select the mathematical distance metric, the method of the present invention becomes more flexible and adjustable. In particular, the user can use his experience to select an appropriate mathematical distance metric to determine the stability of a substance or substance mixture using the method of the present invention.

[0088] In particular, it can be provided that the user selects a plurality of mathematical distance measures for quantifying the changes of the respective substance or substance mixture, wherein the quantification is performed based on each selected mathematical distance measure. Thus, the user can compare the results of different selectable mathematical distance measures with one another.

[0089] Alternatively or additionally, the user can be provided with the ability to successively select different mathematical distance metrics for quantifying the change in the respective substance or substance mixture. This makes the method for determining the stability of a substance or substance mixture particularly flexible and versatile. Furthermore, by providing this option, the inadvertent selection of the mathematical distance metric for quantifying the change in the respective substance or substance mixture can be modified.

[0090] Optionally, within the scope of data evaluation 106 , a PCA calculation 110 can be performed, which precedes the calculation of the starting values ​​112 and the calculation of the distance metric 114 . The abbreviation PCA stands for principal component analysis (German: Hauptkomponentenanalyse). Thus, data evaluation 106 can be performed both with the raw data and after data scaling and / or data reduction to latent variables, for example, by PCA calculation 110 .

[0091] exist Figure 1In the final step shown, data output 108 is performed. For each measurement data set, the corresponding substance mixture's variation is graphically displayed. Data output 108 may specifically include: displaying the variation of the corresponding substance mixture as a boxplot; and / or displaying the mean, median, 0.25 / 0.75 quantiles, highlighting possible outlier candidates; and / or performing at least one statistical test, particularly a t-test, Wilcoxon rank sum test, one-way analysis of variance, and / or a Kruskal-Wallis test. This design of data output 108 enables the simultaneous investigation of very large measurement data sets, particularly across multiple samples, in a reduced time and with increased interpretability. Furthermore, it advantageously provides intuitive evaluation. This means that this technology can be used by users without advanced mathematical knowledge, and each study does not require individual programming. For example, during the data output step, the user can output the variation of the studied substance mixture as a boxplot, providing a simple and intuitive at-a-glance understanding. User-friendliness is further enhanced by highlighting the mean, median, 0.25 / 0.75 quantiles, and outliers. For example, measurement datasets collected by the user can be directly acquired from the measurement device or introduced into the method following the above-mentioned data preprocessing steps and immediately generate easily interpretable evaluations.

[0092] Figure 2 An overview of possible steps of a method according to an embodiment of the present invention is schematically shown. Figure 2 The present invention will be described.

[0093] The raw data from one or more measuring devices are aggregated in step 202. As described above, in order to detect the raw data (data detection step), different measurements can be performed on the same substance mixture or different samples of the substance mixture.

[0094] In step 204, a decision is made as to whether preprocessing is required. If preprocessing is not required, a distance metric is selected in the next step 211. To select a distance metric, one can choose from the above-mentioned distance metrics. After selecting a distance metric, a quantitative and / or qualitative evaluation is performed according to the method design. Within the scope of the quantitative evaluation, in step 212, a starting value or zero value is first identified for each measurement data set. In another step 214, the changes in the substance mixture to be studied are determined based on the selected distance metric, for example, over time. For example, in step 216, the results are grouped and displayed in a box plot. In addition, within the scope of the quantitative evaluation, accompanying statistical tests can be performed in step 217. Within the scope of a possible qualitative evaluation 316, for example, through an additional principal component analysis, the user can obtain useful conclusions beyond the quantitative evaluation. In step 400, the conclusions or results are presented and output graphically.

[0095] If data preprocessing is determined to be necessary in step 204, then a decision is first made in step 2041 as to whether the measurement dataset needs to be cropped. If cropping is necessary, this is performed in step 2042. A decision is then made in step 2043 as to whether saturation correction is necessary, whereby saturation correction is performed in step 2044. Furthermore, a decision is made in step 2045 as to whether scattered light correction is necessary. If scattered light correction is necessary in step 2045, this is performed in step 2046. A decision is then made in step 2047 as to whether a smoothed derivative of the measurement dataset needs to be performed, whereby smoothed derivatives are performed in step 2048. Furthermore, a decision is made in step 2049 as to whether latent variables need to be calculated, and a scaling is selected in step 2050. For example, scaling can be selected from "None," "UV" (= "univariate," not to be confused with "ultraviolet"), or "Pareto." It goes without saying that only some of the aforementioned measures may be performed in the optional preprocessing.

[0096] The PCA calculation may then be performed in step 210. The distance metric may then be selected as described above.

[0097] Example

[0098] The experiments described below were conducted using a reference implementation in the R programming language. The reference implementation consists of the following steps:

[0099] Calculate the distance of each observation / sample to its starting value ("zero-month value") using an appropriate distance metric (e.g., Euclidean, Mahalanobis, Manhattan, Pearson, Gould, etc.). This distance calculation can be performed on the raw data or on preprocessed latent variables (e.g., PCA, data scaling, etc.). The calculated distance metric also allows the determination of the product's decomposition kinetics. For example, in this way, the half-life can also be used to identify the most stable variant.

[0100] - Presents the calculated distances in an interpretable boxplot, including automatic descriptive statistics (calculate and display mean, median, display 0.25 / 0.75 quantiles, annotate possible outlier candidates) and determines / performs the most appropriate statistical test (t-test / Wilcoxon rank-sum test / one-way ANOVA / Kruskal-Wallis test) to compare means.

[0101] - Ability to track changes in samples over time, including estimates of statistical significance (e.g., whether a sample has degraded further or reached a plateau after n months)

[0102] - Use principal component analysis (PCA) to transform the spectral data into latent variables in order to be able to additionally estimate different types of decomposition or rearrangement mechanisms via score plots. Two cases can be distinguished here:

[0103] A) PCA can be calculated from the raw data, and its scores then fed into the distance calculations and subsequent boxplots and statistics. These scores can then also be displayed graphically.

[0104] B) The actual distance calculation (and boxplots and related statistics) can also be done without upstream PCA - then it is still calculated and displayed, but does not involve the actual critical distance calculation.

[0105] - The created application allows intuitive evaluation without requiring programming or deeper mathematical knowledge from the end user.

[0106] The corresponding calculation method implemented in one embodiment of the present invention is described below:

[0107] In this example, the starting point is a user-prepared raw data table containing the samples to be studied and meta-information (e.g., the climate zone or month in which the samples were measured). Each row represents a sample, and each column represents an "attribute," such as the intensity or breaking strength value at the measurement wavelength, the concentration of the substance present, or other collected parameters. The number of columns may vary depending on the measurement method and the availability / relevance of other meta-information. A fragment of an exemplary raw data table is shown here:

[0108] Table 1: An exemplary fragment of the raw dataset for Application Example 1. It includes NIR spectra and meta-information at different time points for a stability study of orange juice under different climatic conditions.

[0109]

[0110] Especially in NIR spectroscopy, it may sometimes be desirable to use only specific wavenumber ranges for evaluation in order to be able to cut and reassemble them as desired.

[0111] For NIR spectroscopy, during the measurement, it may happen that the sensitivity of the detector is exceeded in certain wavenumber ranges (e.g., around 5400 cm -1 to 4900cm -1 Since detector saturation does not contain relevant information and may impair the quality of the evaluation, these areas can be deliberately cut out and excluded from further analysis if desired.

[0112] In NIR spectroscopy, if unwanted scattered light occurs during the measurement, a baseline shift of the individual spectra can occur. This can be eliminated by a scattered light correction. One possibility is the so-called vector normalization (SNV normalization), which works as follows: First, the sample matrix P is transformed and then the baseline is corrected according to the formula Calculate the correction value for each value in each column. Here: x is the value in the column, Where is the mean of the column and σ is the standard error of the column values. The resulting matrix is ​​then converted back so that the samples are listed again in each row. In addition, in embodiments of the present invention, for example, multiplicative scattered light correction (MSC) can also be used to perform scattered light correction in the NIR range. Both of these approaches are well-known standard methods. Figure 4 An exemplary NIR spectral overlay before and after data preprocessing is shown in .

[0113] Furthermore, it may be advantageous if the NIR spectrum (or other raw data matrix) is, if necessary, differentiated and / or subjected to a smoothing algorithm before the actual analysis. For these requirements, the Savitzky-Golay algorithm (see citation

[16] ) is used, which automatically performs both data processing steps.

[0114] If necessary, a principal component analysis (PCA) can be performed on the prepared raw data before the actual distance calculation. The subsequent distance calculation will then be performed based on the determined score values. This step may be useful if you have to deal with highly noisy raw data. Before PCA, the data can also be centered, UV (univariate) or Pareto scaled (centering: UV Scaling: Pareto scaling: Unlike scattered light correction, there is no data matrix transformation before calculation).

[0115] The user can then select the desired distance metric for the actual computation. Typically, this will be the Euclidean distance, but other distance metrics are also conceivable, notably Manhattan distance, Pearson distance, Gower distance, or Mahalanobis distance.

[0116] After this preprocessing, the application now determines, from the raw data metadata, a starting value or "zero value" (in other words, a single data point at time 0) for each included sample. This refers to the sample's chemical / physical / biological state, particularly (but not limited to) sugar content, decomposition rate, color, breaking strength, disintegration time, friability, density, viscosity, refractive index, and / or optical rotation, e.g., under specific climatic conditions or targeted stress tests, such as by UV / VIS light irradiation, enhanced redox reactions, etc. The starting or zero value is user-defined or determined. The sample metadata clearly indicates which sample or sample batch was examined at time 0. If a batch has multiple copies of time 0, an average spectrum is calculated from these and stored in memory for later use. The so-called starting date for each corresponding sample then serves as a reference for distance calculations.

[0117] A distance calculation is then performed for each time point, where this can be determined from the metadata for each batch or for each sample method, as selected by the user above. The distance values ​​thus obtained can be sorted, merged, and presented graphically, allowing for a more detailed static evaluation.

[0118] In the described embodiment, an additional PCA is also calculated from the pre-processed raw data, which PCA is not included in the distance calculation but supports an alternative, but purely qualitative, interpretation of the raw data.

[0119] Application Example 1: NIR Measurement of Liquid Food: Orange Juice

[0120] The subjects of the study are:

[0121] - REWE's 'Beste Wahl' orange juice (100% freshly squeezed juice with pulp, 8.5g sugar per 100ml (as per label))

[0122] - Open the package and store aliquots at room temperature (22°C)

[0123] - Store another aliquot at 40°C, relative humidity = 75%

[0124] As a measuring instrument, a near-infrared spectrometer MPAII (Bruker Optik GmbH) was used:

[0125] - Orange juice measurement in transmission;

[0126] - Wave number range: 12500cm -1 to 3950cm -1 (Spectral range with absorbance > 4.5 is not included in the evaluation)

[0127] Data preprocessing of NIR spectra:

[0128] - Calculate the first derivative using 9 smoothing points

[0129] The following were selected as calculation parameters:

[0130] -Distance metric: Euclidean distance

[0131] Figure 5 A box plot of the calculated distances over 5 days is shown, where day 0 is the "starting value" (="SW"). In the first two days (days 1 and 2), the changes in the two storage conditions are almost the same, with no statistically significant differences. However, starting from day 3, it can be clearly seen that the Euclidean distance of the samples at 40°C increases significantly, and continues to increase on day 4 as well (Kruskal-Wallis test gives p < 0.05, paired Wilcox test shows statistical significance on days 3 and 4, p < 0.05, all other days are insignificant). This is consistent with the commonly held assumption that storing food at high temperatures leads to chemical or microbiological changes (such as decomposition processes of individual components or bacterial growth).

[0132] Application Example 2: NIR Measurement of Solid Food: Minced Meat

[0133] The subjects of the study are:

[0134] -Freshly prepared ground pork from REWE

[0135] - Store aliquots at room temperature (22°C) immediately after purchase

[0136] - Store another aliquot at 40°C, relative humidity = 75%

[0137] As a measuring instrument, a near-infrared spectrometer MPAII (Bruker Optik GmbH) was used:

[0138] -Measure NIR spectra in diffuse reflectance as absorption spectra

[0139] - Wave number range: 12500cm -1 to 3950cm -1

[0140] Data preprocessing of NIR spectroscopy: performing SNV scattered light correction

[0141] As calculation parameter, select: Euclidean distance

[0142] Figure 6The mean values ​​of the calculated Euclidean distances are listed (error bars: standard error). In all cases, the largest changes occurred from day 0 to day 1. In general, as expected, minced meat also changed more at higher temperatures than at room temperature (statistically significant differences at p < 0.05 for all time points (Kruskal-Wallis or Wilcoxon rank sum test)), and plateau formation was not yet complete for the 40°C samples.

[0143] In parallel with the above measurements, photographic documentation was carried out ( Figure 7 ) (beginning and end of the observation period). In the samples stored at high temperature, a distinctly colored "ring" (marked (1)) was visible, and in addition, a liquid formed above the meat sample that was not present in the room temperature sample. In addition, the texture of the samples (clear separation of the meat part (marked (2)) and the fat (marked (3))) changed significantly, which was significantly less obvious in the comparative samples. These changes indicate food spoilage.

[0144] Application Example 3 - Online NIR Measurement of Verbena Extract

[0145] The subjects of the study are:

[0146] -150g cut verbena (including all dried above-ground plant parts)

[0147] - Extract in a beaker, stirring in 1800 ml of deionized water at 20 °C

[0148] As a measuring instrument, a near-infrared spectrometer MPAII (Bruker Optik GmbH) was used. The sample set to be measured was filtered before measurement.

[0149] - Wave number range: 12500cm -1 to 3950cm -1 (Spectral range with absorbance > 4.5 is not included in the evaluation)

[0150] NIR spectral data preprocessing: performing SNV scattered light correction

[0151] As calculation parameter, select: Euclidean distance

[0152] Figure 8A boxplot of the distances calculated using the new method over the time period 0-180 minutes during the extraction process is shown. Clearly, as the extraction progresses, the solution changes are greater at the beginning of the experiment and slowly reach a plateau at the end. However, the key novelty of the present calculation method lies in the fact that, in this experiment, rather than tracking a single lead substance representing the entire extraction process, all dissolved species are tracked across the entire NIR spectrum. Different chemical plant constituents may have different extraction rates / dissolution kinetics. If only a single lead substance is used to evaluate a "depleted" extraction, it is possible that this very substance, in terms of its solution kinetics, is not representative of all contained species. Consequently, an extraction may terminate prematurely because the lead substance no longer shows solution changes after a period of time, even though other undetected components have not yet fully entered solution. The present method eliminates this problem, as all species in solution are detected, and therefore the "total kinetics" of all dissolution processes can also be determined.

[0153] Therefore, testing identical extracts from individual production batches is particularly important in the production of pharmaceutical extracts, as the active ingredient content in the manufactured product should always be the same. Fluctuations in the active ingredient content can lead to "out-of-specification" events during subsequent quality control, which can necessitate the rejection of the production batch. The method of the present invention can significantly alleviate this requirement at minimal cost, and even achieve this for the first time.

[0154] Figure 9 Shows the Figure 8 The total kinetics were calculated from the raw data. The mean values ​​with standard errors of the individual measurements from the box plots are shown in dot form (from Figure 8 ), and the subsequent calculation of the nonlinear kinetic curve. To this end, the formula distance (t) = (distance max *t) / (k+t). Similarly, another appropriate mathematical description can be used. Here, distance(t) = distance at time point t; distance max = distance value in the platform; k = velocity constant.

[0155] Furthermore, from the kinetics of the experiments performed in this example, it can be inferred that the plant sample used was extracted to 50% after approximately 4 minutes, to 90% after approximately 35 minutes, and to near exhaustion at 95% after approximately 75 minutes. This also means that the extraction can be completed after 75 minutes, for example, without having to unnecessarily prolong it for several hours, which would entail considerable costs on an industrial scale.

[0156] Application Example 4 - Tracking the Effects of Stress Factors on Cell Culture

[0157] Prior to the main experiments described below, a preliminary experiment was performed that would show whether HEK295 cells

[0158] a) can be grown confluently in the NIR measurement vessel,

[0159] b) Can stress treatment with ethanol, and if necessary, at what concentration, lead to a corresponding decrease in viability / confluence? This was demonstrated, in particular, using the complex but well-established trypan blue staining. For this purpose, concentrations of 0-10% ethanol were tested.

[0160] Based on the results of this preliminary experiment, it was decided to use two different concentrations of ethanol in the main experiments, since here a clear effect on cell viability could be seen over a period of time that was technically easy to cover later.

[0161] Next the main experiment was performed which confirmed the reduction in viability / confluence by NIR measurements using the computational method of the present invention.

[0162] The subjects of the study are:

[0163] HEK295 cells, supplied by Microbify, were grown confluently in an NIR measurement vessel with a BrukerOptics stainless steel corner punch at a concentration of 2.2 × 105 cells / ml of culture medium, i.e. adherently grown cells forming a continuous monolayer / cell line

[0164] Dulbecco's Modified Eagle's Medium (DMEM) nutrient medium with high glucose content and HEPES buffer without phenol red indicator was used.

[0165] The culture medium was provided by ThermoFisher in a ready-to-use form under catalog number 21063045 and supplemented with 10% fetal bovine serum (v / v) and 1% penicillin-streptomycin solution (v / v).

[0166] Cell culture for 48 hours directly in the measuring vessel before measurement

[0167] - Add various concentrations of ethanol: 0% (control), 3%, 6% (v / v respectively). These EtOH concentrations were chosen due to preliminary experiments that caused corresponding death / lysis responses in cell viability assays. Before the start of the experiment, the cells were already confluent, i.e., 100% viable.

[0168] -Measure cellular changes via NIR over a 380-minute period

[0169] As a measurement instrument, a near-infrared spectrometer microPHAZIR GP (Thermo Fisher Scientific Inc.) was used with a wavenumber range of 6266 cm-1 to 4172cm -1 .

[0170] Data preprocessing of NIR spectra: No special preprocessing was performed and the spectra were directly further processed.

[0171] As calculation parameter, select: Euclidean distance

[0172] The cells were extracted from the incubator and a chemical stressor (ethanol) was added at the concentrations described. Three independent replicates were prepared, each with 0%, 3% and 6% ethanol, and measured by NIR at close intervals (2 minutes each) over a total of 6 hours.

[0173] According to the method of the present invention, the Euclidean distance is calculated from the measured spectra, averaged over the measurement time, and adjusted for the kinetic curve (see Application Example 3; same approach). Figure 10 、 11 and 12, respectively, for control, 3% ethanol, and 6% ethanol treatment. The data points are the average of three independent replicates, each measured three times, so each point has nine measurement points. The error bars are the standard error of the mean. In addition, K m (time point at which 50% of the platform value is reached), K 90 (time point at which 90% of the plateau value is reached) and K 95 (time point at which 95% of the plateau value is reached).

[0174] These values ​​were reached very quickly in the control (K m ~2.4 minutes, K 95 ~43.7 minutes), where the plateau value itself is also the lowest value in the experiment. Obviously, removing the cells from the incubator and using them on the bench is also a low stress factor. m The values ​​were statistically identical (approximately 13 minutes at 3% EtOH and approximately 11 minutes at 6% EtOH), likely due to a uniform, concentration-independent initial response of the cells to ethanol addition.

[0175] However, when ethanol was added at different concentrations, a higher plateau was gradually reached, indicating a significantly higher stress load on the cells. This is consistent with the fact that over time the cells in the measuring vessel (as already observed in preliminary experiments) detached from the surface and ceased to become confluent. This can also be clearly seen in the bar graph ( Figure 13 ) (statistical significance, one-way ANOVA, p < 0.001; post hoc test: paired t-test, all p values ​​< 0.001), which is from Figure 10 、11 and the curve adjustment parameters in 12 are generated.

[0176] The key innovation here is that, in order to demonstrate cellular stress, there is no need to first identify, isolate, and quantify a representative single substance from a cell suspension, nor is there a need to set up and perform complex cell staining and counting assays that, for example, cannot be designed for online measurement. Instead, very high sample quantities can be measured directly spectroscopically without complex external sample preparation and evaluated cost-effectively using the method of the present invention. Furthermore, it has been demonstrated that the method can also be successfully used to track the stability / viability of living cells.

[0177] The cell viability observed at the end of the preliminary experiment (linearly interpolated due to slightly different EtOH concentrations used) was compared to the maximum observed Euclidean distance (R 2 >95%) (see Figure 14 and the table below).

[0178] Table 2: Comparison of distance metrics calculated as Euclidean distance and colorimetric cell viability assay results (Trypan blue staining).

[0179] Preliminary experiments: cell viability Main experiment: Maximum Euclidean distance 0% ethanol 100％ 0.125 2.5% ethanol 68％ (Not measured) 3% ethanol 70.84% ​​(interpolation) 0.194 5% ethanol 65.5％ (Not measured) 6% ethanol 41.38% (interpolation) 0.272 7.5% ethanol 22.5％ (Not measured) 10% ethanol 0％ (Not measured)

[0180] Application Example 5: Comparison of medicine (herbal tea) with photometric reference method

[0181] The subjects of the study are:

[0182] - Medicinal tea: Bad Heilbrunn Liver and Gallbladder Tea (1 filter bag contains 1.75 grams, according to the manufacturer: 0.61 grams of peppermint leaves, 0.35 grams of dandelion, 0.26 grams of Javanese turmeric, 0.18 grams of yarrow; unspecified amounts of fennel, chamomile flowers, cumin, and licorice root)

[0183] - Extract 3 tea bags in 150 ml of deionized water under reflux (boiling temperature about 100°C) in a beaker

[0184] As a measuring instrument, a near-infrared spectrometer MPA II (Bruker Optik GmbH) was used. The sample to be measured was filtered before the measurement.

[0185] - Wave number range: 12500cm -1 to 3950cm -1 (Spectral range with absorbance > 4.5 is not included in the evaluation)

[0186] Data preprocessing of NIR spectroscopy: performing SNV scattered light correction

[0187] As calculation parameter, select: Euclidean distance

[0188] As a reference method, a validated photometric method was used for the determination of total polyphenols (measurement of absorbance at 760 nm after reaction of the solution with molybdate tungstate reagent and 29% sodium carbonate solution, according to European Pharmacopoeia entry 2.8.14).

[0189] The aim of this study was to investigate whether spectroscopic analysis of tea samples at day 0 and day 10 could demonstrate similar recovery / stability to a validated photometric reference method for total polyphenols (GPP) in solution when stored at 4°C refrigerator (KS), 20°C (room temperature RT), and 40°C.

[0190] Figure 15 Shown are the Euclidean distances determined from NIR measurements at day 0 and day 10. It grew the least in the case of samples in the refrigerator (KS), grew moderately when stored at room temperature (RT), and grew the most at 40°C, as intuitively expected.

[0191] and Figure 16 Comparison of the recovery or stability of the mesophotometric method (labeled "GPP" in the figure legend) further demonstrates good agreement between the new calculation method based on the distance metric and this reference. Its slightly lower value compared to the reference method is understandable because the new method detects all substances in the mixture, while the reference method only detects a subset.

[0192] To calculate the percent stability of NIR data, a hypothetical blank spectrum without analyte is taken and its Euclidean distance from the average starting spectrum is calculated. This distance is then used to normalize all other distances within this interval. For the reference method, total polyphenols are determined in mg / 100g and broken down accordingly after 10 days, also normalized to a percentage value.

[0193] Application Example 6 - Comparison of Extraction Mixtures in Different Climate Zones by HPLC / MS

[0194] The subjects of the study are:

[0195] - ethanol-water extract of a mixture of medicinal plants of thyme, rosemary and chamomile (1:1:1) by means of a 20% ethanol-water mixture

[0196] -Storage conditions:

[0197] AC (accelerated conditions): temperature = 40°C, relative humidity = 75%

[0198] KII (climate zone II): temperature = 25°C, relative humidity = 60%

[0199] Stored for 24 weeks

[0200] - Additional independent samples: irradiated at 150 klx for 20 hours in a Suntester tester in a brown glass / white glass bottle (at Figure 18 Indicated by "B" or "W" in the legend)

[0201] Equipment used: Suntest CPS+ (Atlas Material Testing Technology GmbH)

[0202] - Wavelength: 320nm to 800nm

[0203] - Irradiation dose: 3000klx*h

[0204] As measuring instruments, the following equipment was used:

[0205] -Chromatographic system: 1290 Infinity II (Agilent Technologies Germany GmbH & Co. KG)

[0206] -Detector: TripleTOF (Sciex)

[0207] The following calculation parameters were selected:

[0208] -Distance metric: Euclidean distance

[0209] -LC / MS peak table calculation (2602 individual signals)

[0210] - Preprocessing: latent variable calculation, variance coverage > 95%

[0211] For kinetic calculations ( Figure 19 ), always use the equation: distance = (distance max *t) / (k+t), and calculate and determine the distance max and k.

[0212] The results of the study are presented in Figure 19 It is clear that under the more severe climate conditions ("AC"), the changes occur more quickly, as higher distances are reached earlier. However, under both climate conditions, the samples reach almost the same distance plateau at the end of the study.

[0213] Irradiate the sample in the Suntester (see Figure 18 ) reached the same level of change within approximately 20 hours, which was otherwise only achieved after approximately 3.5 weeks of storage under climate zone II conditions. No significant differences were observed between the brown and white glass bottles.

[0214] By using a representative marker substance (m / z 329.17; retention time = 8.16 min, Figure 20 ) demonstrates the predictable degradation of compounds over time, consistent with the NIR assessment results, reaching an endpoint significantly faster under Condition AC than under Climate Zone II. Unlike processing datasets containing extensive compound information (such as the NIR spectra presented here), this analytical approach using only a single marker substance provides only a very limited picture of the product and is not truly representative of the entire multi-component mixture. Furthermore, evaluating datasets for individual compounds is significantly more complex and time-consuming.

[0215] Application Example 7: Comparison of packaging materials by NIR under high temperature and strong light irradiation

[0216] The subjects of the study are:

[0217] - A medicinal plant powder mixture consisting of rosemary, thyme and chamomile in a 1:1:1 ratio

[0218] - Storage conditions: sealed in paper bag (= "P1") or open storage (= "P2")

[0219] Light irradiation was performed under the following conditions:

[0220] -Equipment: Suntest CPS+ (Atlas MaterialTesting Technology GmbH)

[0221] - Wavelength: 320nm to 800nm

[0222] - Radiation dose: 3000klx*h

[0223] As a measuring instrument, a near-infrared spectrometer MPAII (Bruker Optik GmbH) was used:

[0224] -Diffuse reflectance measurement

[0225] As calculation parameters, select:

[0226] -Distance metric: Euclidean distance

[0227] - Calculates SNV-corrected NIR spectra without any restrictions on the wavenumber range

[0228] - Distance calculations were performed on latent variables covering >95% of the total variance

[0229] As a result of the study, the following was obtained:

[0230] The results are Figure 21The results are presented in the form of a box plot. At time 0 (before the start of irradiation in the Suntester tester), the distance values ​​are identical in both samples, whereas the values ​​at time 20 hours not only differ from the starting values, but also differ significantly between P1 (powder mixture in the paper bag) and P2 (powder mixture stored in the open air). This is due, on the one hand, to the intense light effect (and therefore also a significant increase in sample P2), but also to the temperature increase reached in the Suntester tester, which also affects sample P1 (in the paper bag). These results are also confirmed by classical analytical methods. For this purpose, total polyphenols are determined photometrically according to entry 2.8.14 of the European Pharmacopoeia. This is a very complex wet-chemical analytical method. Here, sample P2 shows a significant deviation from the initial value (difference from the initial value = 26%), while sample P1 (difference from the initial value = 1%) remains almost unchanged in the paper bag (see Figure 22 ).

[0231] Thus, the present method clearly demonstrates that temperature and light together have a measurable impact relative to heat alone or the absence of either factor. Time-consuming wet chemical analysis of polyphenols confirms changes in the sample but only captures a partial picture of these changes, as polyphenols represent only a fraction of the overall chemical profile. In contrast, the present method examines the full range of properties, thus providing a more comprehensive picture.

[0232] Application Example 8: Comparison of different climate zones by NIR

[0233] The subjects of the study are:

[0234] - A medicinal plant powder mixture consisting of rosemary, thyme and chamomile in a 1:1:1 ratio

[0235] Storage conditions:

[0236] AC (accelerated conditions): temperature = 40°C, relative humidity = 75%

[0237] KII (climate zone II): temperature = 25°C, relative humidity = 60%

[0238] Stored for 24 weeks

[0239] As a measuring instrument, a near-infrared spectrometer MPAII (Bruker Optik GmbH) was used:

[0240] -Diffuse reflectance measurement

[0241] As calculation parameters, select:

[0242] -Distance metric: Euclidean distance

[0243] - Calculates SNV-corrected NIR spectra without any restrictions on the wavenumber range

[0244] - Distance calculations were performed on latent variables covering >95% of the total variance

[0245] As measuring instruments in parallel with the classical analytical method, the following equipment was used:

[0246] -Chromatographic system: 1290 Infinity II (Agilent Technologies Germany GmbH & Co. KG)

[0247] -Detector: TripleTOF (Sciex)

[0248] As a result of the study, the following was obtained:

[0249] from Figure 23 It can be seen immediately that the more severe climate zone AC leads to significantly higher distances in the first week compared to climate zone II. This is of course also in line with the expectations of the parallel profiling by the classical analytical method of HPLC / MS using a single representative compound. Figure 24 Figure 1 shows the signal intensity curve for the main compound (m / z = 329.17; retention time = 8.16 minutes) from the powder mixture. Here, degradation is much faster in climate AC than in climate zone II. However, this traditional approach, which considers only a single substance, fails to capture the complexity of plant products. The new approach considers a large dataset of the entire complex product. Similar to Example 1, the kinetics in two different climate zones were calculated here using distances determined from the NIR dataset.

[0250] After 24 weeks the changes are also noticeable in appearance ( Figure 25 ), where the powder in the sample bottles from climate zones AC was noticeably darker than its counterpart from climate zone II. However, this purely visual assessment is highly subjective and generally difficult to detect. However, our calculation method makes this change not only qualitatively but also, and especially, quantitatively detectable.

[0251] Application Example 9 - Comparison of packaging materials for solid pharmaceutical forms by NIR

[0252] The subjects of the study are:

[0253] -Hawthorn film tablets:

[0254] Each film tablet contains 450 mg of dried extract of flowering hawthorn leaves

[0255] The film sheets are stored in blister packs in folding boxes on the one hand and in polyethylene bags on the other hand.

[0256] Storage conditions:

[0257] AC (accelerated conditions): temperature = 40°C, relative humidity = 75%

[0258] KII (climate zone II): temperature = 25°C, relative humidity = 60%

[0259] Stored for 24 weeks

[0260] As a measuring instrument, a near-infrared spectrometer MPA II (Bruker Optik GmbH) was used:

[0261] -Diffuse reflectance measurement

[0262] As calculation parameters, select:

[0263] -Distance metric: Euclidean distance

[0264] - Calculates SNV-corrected NIR spectra without any restrictions on the wavenumber range

[0265] As measuring instruments in parallel with the classical analytical method, the following equipment was used:

[0266] -Chromatographic system: 1290 Infinity II (Agilent Technologies Germany GmbH & Co. KG)

[0267] -Detector: TripleTOF (Sciex)

[0268] The evaluation of the NIR dataset shows the expected results. On the one hand, the distance increases faster under AC climate conditions and reaches a larger maximum value over the 24-week period. It is also clear that the higher-grade packaging material (in the form of blister packs combined with folding cartons, marked as "PP" in the legend) has a stabilizing effect compared to polyethylene bags (marked as "FT" in the legend) (see Figure 26 The same results can be obtained by using a representative marker compound (dimeric proanthocyanidin, m / z 577.14; retention time = 2.69 minutes) in the classical evaluation method (see Figure 27 Here, too, it can be seen that the samples from climate zone II are largely consistent, while the samples from climate zones AC degrade significantly more rapidly. However, classic methods are associated with significantly higher measurement and evaluation costs and do not represent the full material complexity of the respective products. The new method overcomes both of these limitations, as the user does not need to specifically select and process individual raw data points. Instead, the complete overall composition spectrum is directly detected via NIR spectroscopy, ensuring that the properties of the sample under investigation are reflected.

[0269] Obviously, aspects described in the context of apparatus in the embodiments described herein also constitute descriptions of corresponding methods. Some or all of the method steps can be performed by (or using) hardware devices, such as processors, microprocessors, programmable computers, or electronic circuits. In certain embodiments, one or more of the most important method steps can be performed by such devices.

[0270] Embodiments of the present invention may be implemented in hardware and / or software. Implementation may be performed using a non-volatile storage medium, such as a digital storage medium, such as a floppy disk, DVD, Blu-ray, CD, ROM, PROM, and EPROM, EEPROM, or FLASH memory, in which electronically readable control signals are stored, which (can) cooperate with a programmable computer system to perform the corresponding method. Thus, the digital storage medium may be computer-readable. Some embodiments of the present invention include a data carrier having electronically readable control signals, which can cooperate with a programmable computer system to perform one of the methods described herein.

[0271] Generally speaking, embodiments of the present invention can be implemented as a computer program product having a program code, wherein the program code is effective to implement one of the methods when the computer program product is run on a computer. For example, the program code can be stored on a machine-readable carrier. Further embodiments include a computer program for performing one of the methods described herein, the program being stored on a machine-readable carrier.

[0272] Another embodiment of the invention is a storage medium (or data carrier or computer-readable medium) comprising a computer program stored thereon for performing one of the methods described herein when executed by a processor. The data carrier, digital storage medium or recorded medium is typically tangible and / or non-continuous. Another embodiment of the invention is an apparatus as described herein, comprising a processor and a storage medium.

[0273] Another embodiment of the invention is a data stream or a sequence of signals representing the computer program for performing one of the methods described herein.For example, the data stream or the sequence of signals can be configured to be transmitted via a data communication connection (for example via the Internet).

[0274] A further embodiment comprises a processing means, for example a computer or a programmable logic device, configured or adapted to perform one of the methods described herein.

[0275] A further embodiment comprises a computer comprising the computer program installed thereon for performing one of the methods described herein.

[0276] Another embodiment of the present invention includes a device or system configured to transmit a computer program for performing one of the methods described herein to a receiver. For example, the receiver may be a computer, a mobile device, a storage device, or the like. For example, the device or system may include a file server for transmitting the computer program to the receiver.

Claims

1. A computer-implemented method for determining the stability of a substance or a mixture of substances, wherein the method comprises the following steps: - a data detection step (102), in which (raw) data and / or metadata are received with a starting value measurement data set and at least one further measurement data set, wherein the starting value measurement data set and each further measurement data set respectively represent chemical properties, in particular phytochemical properties, of a respective substance or substance mixture; - a data evaluation step (106) comprising, for each further measurement data set: - determining (112) a starting value measurement data set for the respective substance or substance mixture based on the (raw) data and / or metadata of the starting value measurement data set; as well as - quantifying (114) the change of the respective substance or substance mixture over time relative to the starting value measurement data set and the at least one other measurement data set by means of a mathematical distance metric; and - a data output step (108), in which for each further measurement data set the variation of the respective substance or substance mixture is presented in a graphical manner.

2. The method according to claim 1, wherein the data evaluation step (106) comprises using a machine learning model, which has preferably been generated and / or trained by unsupervised and / or supervised machine learning.

3. The method according to claim 1 or 2, wherein the at least one measurement data set comprises data obtained by near infrared spectroscopy (NIR).

4. The method according to any one of the preceding claims, wherein the at least one measurement data set comprises: - Data obtained by UV / VIS spectroscopy; - data obtained by Raman spectroscopy; -(U)HPLC fingerprint; -GC fingerprint; - a peak table from chromatography; and / or - at least one physical, biological or chemical parameter, in particular sugar content, disintegration rate, color, breaking strength, disintegration time, friability, density, viscosity, refractive index and / or optical rotation angle. 5 . The method according to claim 1 , wherein the quantification of the change in the respective substance or substance mixture is performed based on the totality of the components in the respective substance or substance mixture.

6. The method according to any of the preceding claims, wherein the mathematical distance metric is selected from the group consisting of: Euclidean distance, Mahalanobis distance, Manhattan distance, Pearson distance and / or Gower distance.

7. A method according to any preceding claim, wherein the mathematical distance metric is selectable by a user.

8. The method according to any one of the preceding claims, further comprising: The data preprocessing step (104) includes: - performing scattered light correction on at least one measurement data set, in particular when said data set comprises near-infrared measurement data; - centering, normalizing and / or scaling at least one measurement data set; and / or -Perform principal component analysis.

9. The method according to any one of the preceding claims, wherein the data output step (108) comprises: - present the changes in the corresponding substance or mixture of substances in the form of a box plot; and / or - present the mean, median, 0.25 / 0.75 quantiles, and dispersion indicators to highlight possible outlier candidates; and / or - Perform at least one statistical test, specifically t-test, Wilcoxon rank-sum test, one-way ANOVA, and / or Kruskal-Wallis test.

10. The method according to any one of the preceding claims, wherein the data evaluation step (106) further comprises: - performing an additional principal component analysis (110) on the measurement data set, said principal component analysis not incorporating a quantification of the variation by a mathematical distance metric; - wherein in the data output step (108), the results of the additional principal component analysis are presented.

11. The method according to any one of the preceding claims, wherein the substance or substance mixture comprises solid and / or liquid and / or gaseous substances.

12. The method according to any one of the preceding claims, wherein the substance or substance mixture comprises a biological, chemical, plant, animal, human substance or substance mixture, a pharmaceutical composition, a botanical, a chemical and / or biological drug, a cell, a cell therapy product (e.g., a gene therapy product, such as CAR T cells (chimeric antigen receptor T cells), NK cells (natural killer cells), a somatic cell therapy product, a biotechnologically processed tissue product / tissue engineering product, a tissue, a stem cell, a stem cell product or preparation, such as CD34+ cells, CD19+ cells, CD20+ cells, HEK295 cells, TCR alpha / beta cells, TCRgamma / delta cells, CD3+, CD4+, CD8+, CD133+ cells), blood, blood products, organs, medicinal teas, extracts, in particular verbena extracts, medicinal plant mixtures of thyme, rosemary and chamomile, for example in ethanolic extract or powdered form, drops, tablets, coated pills, capsules, powders, granules, solutions, suspensions, slurries, foods, in particular meat or minced meat, fruit juices, in particular orange juice, nutritional supplements, cosmetics, lotions, ointments and / or creams, as well as packaging, packaging materials, films, in particular polyethylene, polyvinyl chloride.

13. A computer program comprising instructions which, when said program is executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 12.

14. A device, in particular a measuring instrument or a server computer, comprising means for carrying out the method according to any one of claims 1 to 12.

Images