Method for determining a minimum inhibitory concentration, and analysis method
The method uses a machine learning model to analyze spectrometric data and shift signals for accurate MIC determination, addressing measurement inconsistencies and instrument calibration issues, enhancing reliability and efficiency in determining MICs.
Patent Information
- Application Number
- PCT/DE2026/100556
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-07
- Filing Date
- 2026-05-04
- Publication Date
- 2026-11-12
AI Technical Summary
Current methods for determining minimum inhibitory concentrations (MICs) are inaccurate due to biological variance and measurement inconsistencies, leading to unusable data and difficulties in assessing antibiotic resistance and selecting appropriate antimicrobial therapy.
A method involving spectrometric measurement data analysis using a machine learning model to determine similarity values between samples pretreated with different concentrations of an antimicrobial agent, incorporating signal shifting to compensate for measurement inaccuracies and inconsistencies, and using unified time-of-flight data for standardized comparison.
This approach provides a more reliable and accurate determination of MICs, even with poorly calibrated instruments, simplifying the process and reducing the need for internal standards, while improving accuracy and cost-effectiveness.
Smart Images

Figure DE2026100556_12112026_PF_FP_ABST
Abstract
Description
[0001] BRE0627PT25WO1 May 4, 2026
[0002] - 1 -
[0003] Methods for determining a minimum inhibitory concentration and analytical methods
[0004] The present invention relates to a method for determining a minimum inhibitory concentration. The present invention also relates to an analytical method using flight-time data.
[0005] The invention is defined in the attached claims. Preferred aspects of the present invention will also become apparent from the following description, including the examples.
[0006] Where certain embodiments are designated as preferred for an aspect of the invention, the corresponding descriptions also apply to the other aspects of the present invention, mutatis mutandis. Preferred individual features of aspects of the invention (as defined in the claims and / or disclosed in the description) can be combined with one another and are preferably combined with one another, unless otherwise apparent to a person skilled in the art from the present text in a particular case.
[0007] The term minimum inhibitory concentration (MIC) is generally understood, as in the context of the present invention, to be the lowest concentration of an antimicrobial agent (for example, an antibiotic) that prevents the growth of microorganisms under investigation—such as a bacterial strain. The MIC is an important measurement in microbiology and infection diagnostics for determining the efficacy of an antibiotic against pathogens and represents a key parameter for assessing antibiotic resistance and selecting appropriate antimicrobial therapy.
[0008] The growth of microorganisms can be determined using various evaluation algorithms. Typically, the growth of microorganisms pretreated with a known concentration of an antimicrobial agent – that is, exposed to a known concentration of an antimicrobial agent – is compared with the growth of microorganisms without the influence of the antimicrobial agent (growth control). Based on several such comparisons with varying concentrations of the antimicrobial agent, the MIC (minimum inhibitory concentration) is determined.
[0009] WO 2020 / 083486 discloses the determination of MICs using conventional methods in accordance with the CLSI and EUCAST guidelines for antibiotic susceptibility testing. BRE0627PT25WO1 4 May 2026
[0010] - 2 -
[0011] In EP 3548895 Bl, the so-called log-score algorithm is proposed as a basis for calculating the MIC. However, this approach has difficulties with poorly calibrated instruments. Furthermore, increasing the accuracy of MIC determination, for example in clinical applications, is desirable.
[0012] Current methods for determining the MIC are typically designed to assume a clear cut-off value (also called a threshold) at a specific concentration of antimicrobial agent used for pretreatment. This means that all samples pretreated with a lower concentration of the antimicrobial agent will show growth, while all samples pretreated with a higher concentration will not. However, due to biological variance and measurement inaccuracies, this is not always the case in practice. It can happen that growth is observed at a certain concentration of antimicrobial agent, while no growth is observed at a lower concentration. Data with such inconsistencies are often unusable with current methods.
[0013] The primary objective of the present invention was to provide a method that allows for a more reliable and accurate determination of minimum inhibitory concentrations (MICs). A further objective was to provide an analytical method that allows for simpler and more reliable sample characterization.
[0014] Further tasks arise from the following description and the patent claims.
[0015] The primary object of the present invention is achieved by a method for determining a minimum inhibitory concentration (MIC), comprising the following steps:
[0016] To prepare or provide a sample containing microorganisms;
[0017] Pretreating two or more subsets of the sample containing microorganisms with different concentrations of an antimicrobial agent;
[0018] Generating spectrometric measurement data from samples comprising subsets of microorganisms pretreated with different concentrations of an antimicrobial agent; BRE0627PT25WO1 May 4, 2026
[0019] - 3 -
[0020] Creating or providing spectrometric measurement data from a subset of the sample comprising microorganisms that was not pretreated with the antimicrobial agent, as a growth control;
[0021] Comparing the spectrometric measurement data of the sample comprising subsets of microorganisms pretreated with different concentrations of an antimicrobial agent with the spectrometric measurement data of the growth control and determining similarity values between the compared spectrometric measurement data;
[0022] Determining the minimum inhibitory concentration (MIC) using the determined similarity values;
[0023] where
[0024] The determination of the minimum inhibitory concentration (MIC) is carried out using a machine learning model, and the determined similarity values are used as input for the machine learning model.
[0025] and / or
[0026] The determination of the similarity values involves shifting one or more signals from at least one of the two spectrometric measurement data being compared.
[0027] The sample containing microorganisms can comprise exclusively identical microorganisms (for example, exclusively bacteria of a specific species) as well as microorganisms that differ (to a greater or lesser degree) from one another (for example, different bacteria of a specific genus or family, or microorganisms from different groups, such as bacteria and fungi). A preferred method according to the invention is one in which the sample containing microorganisms comprises bacteria and / or fungi, preferably bacteria, particularly preferably exclusively bacteria, and most preferably exclusively bacteria of a single family.
[0028] For the purposes of the present invention, the term "antimicrobial agent" refers to a substance or combination of substances that can inhibit the growth of microorganisms—such as bacteria, fungi, and / or yeasts—reduce their ability to reproduce, or kill or inactivate them. A preferred method according to the invention is one in which the BRE0627PT25WO1 May 4, 2026
[0029] - 4 -
[0030] The antimicrobial agent (used in different concentrations for the pretreatment of two or more subsets of the sample containing microorganisms) is an antibiotic and / or an antifungal agent, preferably an antibiotic. In addition to the antimicrobial agent, which is used in different concentrations for the pretreatment of two or more subsets of the sample containing microorganisms, other agents may be used for pretreatment, such as one or more substances to inhibit resistance enzymes (e.g., a beta-lactamase inhibitor) or one or more (other) agents to enhance the effect of the antimicrobial agent. Such additional agents may either be kept constant in concentration or adjusted proportionally to the concentration of the antimicrobial agent.
[0031] For the purposes of the present invention, the term "spectrometric measurement data" refers to quantitative measurement results obtained during spectrometric analysis using a spectrometer. This data represents the intensity or frequency distribution of a measured signal as a function of a physical quantity, such as wavelength, frequency, energy, mass, or time. A preferred method according to the invention is one in which the generated and optionally provided spectrometric measurement data are preferably mass spectrometric measurement data, which are further preferably obtained using a time-of-flight mass spectrometer (TOF instrument). Preferably, the spectrometric measurement data are obtained using a sample-consuming spectrometry method.
[0032] A method according to the invention is also preferred, wherein the creation and / or provision of the spectrometric measurement data comprises the performance of TOF measurements, preferably MALDI-TOF measurements.
[0033] "TOF" (short for "Time of Flight") refers, within the meaning of the invention, to a specific form of mass spectrometry, namely time-of-flight mass spectrometry, which is based on measuring the flight time of ions accelerated in an electric field. The term "TOF measurements" therefore refers, within the meaning of the invention, to measurements of the flight time that are or have been performed with a time-of-flight mass spectrometer. Within the meaning of the invention, a time-of-flight mass spectrometer also includes measuring instruments that, in addition to measuring the flight time, offer other metrological possibilities, such as the possibility of performing an ion mobility measurement. For example, the combination of a time-of-flight mass spectrometer and a BRE0627PT25WO1 May 4, 2026
[0034] - 5 -
[0035] Ion mobility spectrometers, such as a TIMS-TOF instrument, also constitute a time-of-flight mass spectrometer within the meaning of the invention.
[0036] In the context of the invention, "MALDI" stands for "Matrix-Assisted Laser Desorption / Ionization," which refers to a specific method for the ionization of molecules. In MALDI-TOF, the principle of matrix-assisted laser desorption / ionization (MALDI) is combined with the principle of time-of-flight mass spectrometry (TOF), meaning that TOF measurements of ions previously generated using MALDI are performed.
[0037] The term "spectrometric measurement data" further encompasses, within the meaning of the invention, raw spectrometric data up to processed peak lists containing only the positions and intensities of signals. This means that the raw data from spectrometric measurements can already be used for comparison. However, before comparison, the spectrometric measurement data can also be subjected to signal processing, which includes, for example, baseline correction (subtraction), signal smoothing, noise reduction, selection of signals above a defined noise threshold, and / or conversion of the measured signals into another physical quantity (for example, converting time-of-flight data into m / z values).
[0038] Within the scope of the present invention, multiple spectrometric measurement data can be generated from one or more of the subsets of the sample containing microorganisms to be examined, and these data can be taken into account when determining the similarity values. For example, based on several spectrometric measurement data sets generated from a subset of the sample containing microorganisms that has been pretreated with a specific concentration of an antimicrobial agent, several similarity values for this concentration of antimicrobial agent can be determined.These multiple similarity values for a concentration of antimicrobial agent can then be used as input for the machine learning model either directly or an average value can first be calculated from these multiple similarity values for the corresponding concentration of antimicrobial agent and this average value can be used as input for the machine learning model.
[0039] It goes without saying that the subsets of the sample comprising microorganisms, which were pretreated with different concentrations of an antimicrobial agent, should first be left for a certain period of time under conditions under which a spectrometric measurement data is generally possible before the spectrometric measurement data is obtained. BRE0627PT25WO1 4 May 2026
[0040] - 6 -
[0041] Detectable growth of the microorganisms would be expected. The same applies to the subset of the sample containing microorganisms that serves as a growth control (that is, provided that the spectrometric measurement data for the growth control are also generated specifically for the execution of the procedure and are not merely obtained from, for example, a reference database or previous in-house measurements). In this context, the person skilled in the art knows from their general professional practice and experience what waiting period should be observed before generating spectrometric measurement data and how the conditions for the subsets of the sample containing microorganisms to be examined should be chosen during the waiting period.Preferably, all subsets of the sample containing microorganisms to be examined are subjected to the same waiting time and the same prevailing conditions, so that the handling of the subsets of the sample containing microorganisms to be examined before the creation of the spectrometric measurement data differs only by pretreatment with different concentrations of antimicrobial agent.
[0042] Preferably, the pretreatment of the subsets of the sample containing microorganisms to be examined, which is carried out before the generation of the spectrometric measurement data, additionally includes one or more of the following steps:
[0043] Introducing the sample, comprising two or more subsets of microorganisms, into a nutrient medium;
[0044] Transferring the sample, comprising two or more subsets of microorganisms, into an incubation chamber with a defined atmosphere for a predetermined period of time;
[0045] Removal of residual liquid from the nutrient medium after a predetermined time period;
[0046] Preparing the sample comprising two or more subsets of microorganisms for desorbing ionization;
[0047] Transferring the sample, comprising two or more subsets of microorganisms, into a desorption ion source of a mass spectrometer.
[0048] In various embodiments, the temperature and humidity in the incubation chamber can be set to approximately 36 °C or close to saturation to create optimal or required growth conditions for the subsets of microorganisms in the sample under investigation. The general aim should be to create conditions in the incubation chamber that most clearly reveal differences in growth. The temperature of 36 °CBRE0627PT25WO1 May 4, 2026
[0049] - 7 -
[0050] This temperature corresponds approximately to human body temperature and is suitable for microorganisms that have specialized in using humans as hosts. However, in veterinary, food, or environmental diagnostics, for example, other temperatures may be deemed optimal, depending on the host or environmental environment preferred by the microorganism in question. The high humidity, close to 100%, serves primarily to prevent premature evaporation of the nutrient medium, ensuring that the volume of liquid available for microbial growth is maintained for the predetermined period, typically several hours, during which the sample containing the microorganisms to be examined remains in the incubation chamber (generally 1-18 hours).
[0051] Preferably, the quantity of microorganisms in the sample containing microorganisms is below the detection limit for the spectrometric measurement method used to generate the spectrometric measurement data before the pretreatment step; for example, at 10 5 CFU per milliliter or less (CFU = colony forming unit). This allows for a better assessment of microbial growth under the influence of different concentrations of antimicrobial agent.
[0052] According to the inventive method, the spectrometric measurement data for growth control can be provided, for example, from a reference database or previous in-house measurements. Preferably, however, the spectrometric measurement data are also generated (newly) in each case. Preferably, the generation of the spectrometric measurement data for growth control is carried out simultaneously with, or directly before, or after the generation of the spectrometric measurement data for the subsets of the microorganisms pretreated with different concentrations of an antimicrobial agent.
[0053] For the execution of the method according to the invention, all spectrometric measurement data are preferably generated using the same measuring instrument.
[0054] The determination of similarity values can be carried out in various ways, for example, using the log-score algorithm described in EP 3548895 Bl. Preferably, the similarity values are determined by Pearson correlation or by calculating the Euclidean distance between the spectrometric measurement data to be compared. The determination of similarity values can also be carried out using a machine learning model, which, for example, performs embedding to determine the similarity values. In a preferred embodiment, the determination of the BRE0627PT25WO1 May 4, 2026
[0055] - 8 -
[0056] Similarity values and the determination of the MHK are performed using the same machine learning model.
[0057] To determine the similarity values, either the entire spectrometric measurement data of the compared measurements or only one or more characteristic signals (also called peaks) from the compared spectra can be used.
[0058] Determining the similarity values between the compared spectrometric measurement data and using these similarity values as input for the machine learning model leads to the acquisition of minimum inhibitory concentrations (MICs) that exhibit better accuracy than those MICs previously obtainable by methods known from the prior art. Furthermore, this inventive approach makes it possible to determine a reliable MIC even when dealing with inconsistent datasets or series of similarity values (as explained above).
[0059] For the purposes of this invention, a "machine learning model" is generally understood to be a mathematical or programming construct that has been trained to recognize patterns in data and to make predictions or decisions based on these patterns. The machine learning model is created through a training process in which an algorithm learns to identify relationships and structures using example data (training data). After training, the model can be applied to new, unknown data. Examples of machine learning models usable within the scope of this invention include k-means algorithms, support vector machines, decision trees, random forests, neural networks, and generative adversarial networks. The log-score algorithm described in EP 3 548 895 Bl, however, does not constitute a machine learning model within the scope of this invention.
[0060] A preferred method according to the invention is one in which the machine learning model is a neural network.
[0061] A preferred method according to the invention comprises the training data with which the machine learning model was trained, spectrometric measurement data of a microorganism pretreated with different concentrations of an antimicrobial agent, in which no continuous attenuation of the growth of the microorganism (in the form of a continuously weaker expression of the measurement signals for the microorganisms) is observed with increasing concentration of the antimicrobial agent; that is, the concentration series exhibits one or more inconsistencies (so-called "skipped wells") in which, despite increasing concentration of the BRE0627PT25WO1 May 4, 2026
[0062] - 9 -
[0063] A stronger signal intensity for the microorganism is observed in the antimicrobial agent compared to the previous measurement (with a lower concentration of antimicrobial agent for pretreatment). Such "skipped wells" in measurement series are occasionally observed in practice, and the causes for their occurrence can be complex. Possible causes include, for example, the development of heteroresistance, an inhomogeneous inoculum, or inhomogeneous inoculation during pretreatment. Although the machine learning model is already able to output correct MIC values with a high degree of probability even when trained without such inconsistent measurement series, incorporating such inconsistent measurement series into the training data of the machine learning model further improves its predictive power for determining the MIC values.
[0064] Shifting one or more signals from at least one of the two compared spectrometric measurements during the determination of similarity values also leads to a more reliable and accurate determination of MIC. This allows, for example, the compensation of measurement inaccuracies due to so-called "jitter." This contributes significantly to the fact that the method according to the invention enables the determination of reliable values regardless of the instrument-specific calibration and can, for example, also lead to reliable results with poorly adjusted or incorrectly calibrated measuring instruments. The method according to the invention thus also enables a simpler and more cost-effective determination of MIC, since, for example, the use of an internal standard can be dispensed with.
[0065] The term "jitter" refers to temporal fluctuations (in either a positive or negative direction) or irregularities in the occurrence of signals, particularly in the measurement of flight times or pulses. Generally, jitter is understood as "clock jitter"—that is, deviations from the ideal timing at which a signal is expected. Jitter can occur, for example, if the times at which ions reach a detector vary slightly from measurement to measurement. These temporal fluctuations lead to a deterioration of the temporal resolution and thus to reduced mass accuracy. The causes of jitter are manifold, including, among other things, a technically inherent measurement inaccuracy of the instrument used or inadequate calibration.
[0066] Jitter can negatively affect the results when determining similarity scores. It has been shown that the negative effect caused by jitter can be mitigated by shifting the values. (BRE0627PT25WO1 May 4, 2026)
[0067] - 10-
[0068] The similarity values can be compensated for by one or more signals from at least one of the two compared spectrometric measurement data sets. For this purpose, at least one signal from one of the compared spectrometric measurement data sets is shifted at least once, and a comparison including artificially shifted signals is also taken into account when determining the similarity values.
[0069] Within the scope of the invention, one or more signals from one or both of the spectrometric measurement data to be compared can be shifted. A preferred method according to the invention is one in which one or more signals from the spectrometric measurement data of the growth control are shifted. The shifted signals are preferably shifted multiple times, particularly preferably at least once to positive values and at least once to negative values.
[0070] Preferably, when comparing (a pair of) spectrometric measurement data, both a first comparison between the spectrometric measurement data without signal shift and at least one further comparison between the corresponding spectrometric measurement data after shifting one or more signals are performed. According to a preferred embodiment, the similarity value used to determine the minimum inhibitory concentration (MIC) can subsequently be derived from the comparison with the best agreement (i.e., the maximum value obtained).In an alternative preferred embodiment, multiple similarity values for a pair of compared spectrometric measurements can be determined from the multiple comparisons (with and without consideration of shifted signals), and these multiple similarity values (for a pair of compared spectrometric measurements) can be used as input for the machine learning model to determine the MIC. For use in the machine learning model, the multiple similarity values (for a pair of compared spectrometric measurements) can each be pooled (i.e., combined into a multidimensional vector).
[0071] A preferred method according to the invention is one in which time-of-flight data, preferably time-of-flight data from MALDI-TOF measurements, are used for the comparison of the spectrometric measurement data.
[0072] US20180053645A1 discloses a method comprising: Receiving spectrometer test data of a sample, which expresses time-of-flight data in units of time-of-flight and intensity of ionized particles passed through a BRE0627PT25WO1 May 4, 2026
[0073] - 11 -
[0074] flight tube migrate, include; matching the spectrometer test data with a reference library to determine characteristic information of the sample, wherein the reference library includes spectrometer pattern data in units of flight time and intensity of ionized particles from pre-stored reference samples acquired by at least one spectrometer, and wherein the spectrometer reference data have known properties that are associated with the matching with the received spectrometer test data.
[0075] Typically, the evaluation of (MALDI-)TOF measurements first involves converting the time-of-flight data obtained from the measurements into mass spectra or mass-to-charge ratios (m / z). This is because the time-of-flight data (also referred to as "TOF data") are often difficult to interpret directly and, above all, are instrument-specific, meaning they can vary considerably between different (MALDI-)TOF instruments. However, using mass spectrometric data calculated from time-of-flight data has the disadvantage of reducing the accuracy of identification. This is partly due to certain properties of mass spectra—such as their nonlinearity—which are detrimental to signal processing. To compensate for the nonlinearity of mass spectrometric data, methods such as binning (see [reference]) can be used.such as US2014358449A1) or curve fitting, which, however, have the disadvantages of leading to a reduction in resolution, potentially introducing additional errors into the measurement data and causing additional computational effort.
[0076] Malyarenko et al. have published several articles describing the preprocessing of flight-time data prior to its conversion into mass spectrometric data in order to compensate for the problem of nonlinearity. The following publications by Malyarenko et al. are referenced:
[0077] Malyarenko, DI, Cooke, WE, Adam, B.-L, Malik, G., Chen, H., Tracy, ER, Trosset, MW, Sasinowski, M., Semmes, OJ, & Manos, DM (2005), "Enhancement of Sensitivity and Resolution of Surface-Enhanced Laser Desorption / ionization Time-of-Flight Mass Spectrometric Records for Serum Peptides Using Time-Series Analysis Techniques," Clinical Chemistry, 51(1), 65-74;
[0078] Malyarenko, D. I., Cooke, W. E., Tracy, E. R., Drake, R. R., Shin, S., Semmes, O. J., Sasinowski, M., & Manos, D. M. (2006), "Resampling and deconvolution of linear time-of-flight records for enhanced protein profiling", Rapid Communications in Mass Spectrometry, 20(11), 1670-1678;BRE0627PT25WO1 4. Mai 2026
[0079] - 12 -
[0080] Malyarenko, D. I., Cooke, W. E., Tracy, E. R., Trosset, M. W, Semmes, O. J., Sasinowski, M., & Manos, D. M. (2006) "Deconvolution filters to enhance resolution of dense time-of-flight survey spectra in the time-lag optimization range", Rapid Communications in Mass Spectrometry, 20(11), 1661-1669.
[0081] DE102004051043A1 also discloses a method for making two time-of-flight mass spectra comparable, characterized in that the mass values of one of the two mass spectra are adapted to the mass values of the other mass spectrum by transformation using the intensity patterns of characteristic ion signals of both mass spectra.
[0082] By directly using time-of-flight data, preferably from MALDI-TOF measurements, for the comparison of spectrometric measurement data, the disadvantages described above, which are associated with the conversion of time-of-flight data into mass spectrometric data, are eliminated, and the computational effort required for the conversion is also eliminated. Since, according to the inventive method, the analysis is based on a comparison of spectrometric measurement data that either all originate from the same sample (which has only been pretreated differently) or, conversely (i.e., in the case of providing spectrometric measurement data for growth control), are at least known, the aforementioned problem of the difficult direct interpretation of time-of-flight data is also irrelevant.Preferably, all spectrometric measurement data are generated using the same instrument, so the aforementioned instrument variance in time-of-flight data does not pose a problem. The use of time-of-flight data thus enables a rapid comparison of generated (and, if applicable, additionally provided) spectrometric measurement data, which are characterized by high resolution.
[0083] The direct use of flight time data for evaluation and analysis also makes it possible to apply further methods of signal processing to this data within the framework of the evaluation, in particular methods of time series analysis such as wavelet transformation and / or Fourier transformation.
[0084] A further preferred method according to the invention is one in which the flight-time data to be compared are first calibrated using identical (calibration) flight-time data, which were acquired from a reference sample of known composition using a time-of-flight mass spectrometer (serving as a reference device), preferably a MALDI-TOF instrument. (soBRE0627PT25WO1 May 4, 2026)
[0085] - 13 -
[0086] Calibrated time-of-flight data are also referred to as "unified time-of-flight data" or "uTOF" within the meaning of the invention.
[0087] For the purposes of this invention, a "MALDI-TOF device" is any device capable of performing MALDI-TOF measurements. Examples include the neofleX™, rapifleX®, ultrafleXtreme, autoflex® maX, microflex®, smartfleX and MALDI Biotyper® systems from Bruker.
[0088] The "use of identical (calibration) time-of-flight data" means that all time-of-flight data to be compared are calibrated beforehand with one and the same time-of-flight data acquired on a time-of-flight mass spectrometer serving as a reference instrument, preferably a MALDI-TOF instrument. This allows all time-of-flight data to be calibrated against the same standard or norm-serving time-of-flight values for known signals. For calibration purposes, the same reference sample of known composition is typically measured on the same time-of-flight mass spectrometer, preferably a MALDI-TOF instrument, with which the time-of-flight data to be compared for analysis were measured. Based on the signal values of the reference sample on the reference instrument and the instrument(s) used for the actual analysis, the time-of-flight data used for comparison can then be aligned and thus unified or standardized.The reference sample can either be added to the sample to be analyzed as an internal standard or measured separately on the respective instrument used for the analysis.
[0089] Creating uTOF improves the signal quality of the data recorded by a time-of-flight mass spectrometer, preferably a MALDI-TOF device.
[0090] In contrast to TOF data, uTOF is device-independent and the signals are comparable for devices with different designs and / or parameters (such as tube length and laser energy).
[0091] The advantage of generating and using uTOF data is that it allows for the standardization of time-of-flight data. Due to differing characteristics of time-of-flight mass spectrometers, such as MALDI-TOF instruments (e.g., varying tube lengths or laser power), the time-of-flight data for a sample can differ between instruments. This can significantly complicate the direct use of TOF data for analysis, especially when comparing time-of-flight data from different instruments. BRE0627PT25WO1 May 4, 2026
[0092] - 14 -
[0093] In other words, converting time-of-flight (TOF) data to mass spectrometric data offers a way to standardize the time-of-flight data from different instruments (by using specific masses for calibration). However, mass spectrometric data has disadvantageous signal characteristics (such as nonlinearity). To preserve the advantageous properties of the TOF signal, the idea is to standardize the TOF data so that the TOF data are more comparable between instruments and these standardized uTOF data can be used instead of mass spectrometric data. This has the advantage of obtaining standardized data (as with mass spectra) while retaining the advantageous signal characteristics of the TOF signal (such as better resolution and signal linearity), thus yielding overall better data for further analysis and evaluation.
[0094] To calculate the uTOF, a time-of-flight mass spectrometer, such as a MALDI-TOF instrument, is required as a reference device. The TOF data from all other measuring instruments are calibrated to this reference device, ensuring that the resulting uTOF values can be easily and reliably compared across different instruments.
[0095] For this calibration, a reference sample (e.g., Bruker BTS) is measured on the reference instrument. For this reference instrument, the TOF data are equal to the uTOF (which means, for example, that for all other calibration constants Ci=l and c) x =0 can apply).
[0096] To calculate the standardized uTOF, the other devices are calibrated to the reference device. A polynomial can be used for calibration (the polynomial can be of any degree, but a first- or second-degree polynomial is recommended).
[0097] For a first-degree polynomial, the relationship between TOF (t) can be expressed as follows: x ) and uTOF (u x ) can be described by the following formula:
[0098]
[0099] C o + Q • U x
[0100] The following formula can be used to calculate uTOF:
[0101]
[0102] For a second-degree polynomial, the relationship between TOF (t) can be expressed as follows: x ) and uTOF (u x ) can be described by the following formula:
[0103] c0+ c ± ■ u x + r?2 * u x BRE0627PT25WO1 May 4, 2026
[0104] - 15 -
[0105] The following formula can be used to calculate uTOF:
[0106] V c i “ 4C 2(CO - - Q
[0107]
[0108] 2C2
[0109] Various methods can be used to calculate the calibration constants. For example, a similarity metric can be defined, and then an optimization procedure can be used to determine the calibration constants. Preferably, the least squares method is used to determine the calibration constants.
[0110] A further preferred method according to the invention is one in which the reference sample of known composition is preferably a bacterial test standard (BTS). An example of this is the bacterial test standard from Bruker. This contains an extract of Escherichia coli DH5 alpha, which has a characteristic and known peptide and protein profile mass spectrum. The Bruker BTS is supplemented with two additional proteins that extend the upper limit of the mass range of the BTS; the total mass range covered by the BTS is therefore between 3,600 and 17,000 mcg.
[0111] A preferred method according to the invention is one in which four, preferably eight, subsets of the sample containing microorganisms are pretreated with different concentrations of an antimicrobial agent. The examination of four to eight subsets of the microorganisms pretreated with an antimicrobial agent increases, on the one hand, the accuracy for determining the MIC and, on the other hand, represents an acceptable number of samples to be examined in terms of effort.
[0112] The creation and use of uTOF has not only proven advantageous in the context of determining a MIC, but also offers the aforementioned advantages for other analyses for which flight time data are available, especially if this flight time data was created using various measuring instruments.
[0113] Part of the invention therefore also includes an analysis method comprising the following steps:
[0114] Preparing or providing a sample to be analyzed;
[0115] Receiving flight time data from the sample to be analyzed
[0116] or BRE0627PT25WO1 May 4, 2026
[0117] - 16 -
[0118] Performing one or more TOF measurements, preferably MALDI-TOF measurements, of the sample to be analyzed (preferably after pretreating the sample to be analyzed under different conditions, for example, after pretreating the sample to be analyzed with different concentrations of an antimicrobial agent), so that time-of-flight data are obtained from the sample to be analyzed;
[0119] Using the time-of-flight data to characterize the sample to be analyzed (that is, without converting the time-of-flight data into mass spectrometric data),
[0120] preferably i) by comparing the obtained time-of-flight data from at least two TOF measurements, preferably MALDI-TOF measurements, with each other and / or ii) by comparing the obtained time-of-flight data with provided reference time-of-flight data,
[0121] wherein the flight time data to be used for characterization are each calibrated beforehand using identical (calibration) flight time data which were recorded with a TOF device (serving as a reference device), preferably a MALDI-TOF device, from a reference sample of known composition.
[0122] The definitions and advantageous effects explained for the features of the inventive and preferably inventive method for determining a minimum inhibitory concentration apply mutatis mutandis to the respective same features of the inventive analytical method.
[0123] The invention is explained in more detail below with reference to exemplary embodiments and the accompanying figures. The exemplary embodiments given below are intended to describe and explain the invention in more detail without limiting its scope.
[0124] Example of the determination of a minimum inhibitory concentration using the method according to the invention
[0125] To determine the minimum inhibitory concentration (MIC), four subsets (S0, S1, S2, and S4) of a provided sample containing microorganisms were pretreated with different concentrations of the antibiotic levofloxacin. A further subset of the provided sample containing microorganisms (SGC) served as a growth control. The following concentrations of levofloxacin were used for pretreatment: BRE0627PT25WO1 May 4, 2026
[0126] - 17 -
[0127] Table 1: Concentration of levofloxacin used for pretreatment
[0128] Sample name Concentration for pretreatment
[0129] used levofloxacin
[0130] SGC 0 mg / L
[0131] S0.5 0.5 mg / L
[0132] Si 1 mg / L
[0133] S22 mg / L
[0134] S44 mg / L
[0135]
[0136] Apart from the different levofloxacin concentrations, all subsets of the provided sample containing microorganisms were treated identically: each was stored in a liquid medium (containing the different antibiotic concentrations) for 8 hours at a temperature between 35 and 37 °C. After storage, the liquid was removed, and mass spectrometric data were acquired from each of the five subsets of the provided sample containing microorganisms using MALDI-TOF measurements. A Bruker MALDI Biotyper® was used for the mass spectrometry. 240 laser pulses were used to acquire each mass spectrum. The acquisition range for the mass spectra was set to 2,000–20,000 m / z. The resulting mass spectra are shown in Figures 1 to 5.
[0137] In the next step, the mass spectrum of the growth control (SGC) was compared with the mass spectra of the samples containing different concentrations of levofloxacin (S0.5, Si, S2, and S4) of the provided microorganisms, and similarity values were determined for the compared mass spectra. For the comparisons, the signals (also referred to as peaks) in the range of 4,000 to 12,000 m / z of the mass spectra were considered. Before performing the comparisons, the mass spectra were also subjected to baseline correction and smoothing.
[0138] The similarity values were determined using Pearson correlation with the following formula: BRE0627PT25WO1 May 4, 2026
[0139] - 18 -
[0140] E(%i - %)(yt -y)
[0141] corr(X, Y)
[0142]
[0143] VSCxi - xP SCyi -y) 2
[0144] where
[0145] X and Y represent the two mass spectra being compared.
[0146] Xi and yi represent the values of the respective mass spectra at point i, and
[0147] where x and y represent the respective mean values of all the considered values of the respective mass spectra.
[0148] Using Pearson correlation, the following similarity values were determined for the compared mass spectra:
[0149] corr(So.5,SGc) = 0.882
[0150] corr(Si,Scc) = 0.625
[0151] corr(S2,Scc) = 0.260
[0152] corr(S4,Scc) = 0.086
[0153] Based on the determined similarity values, the minimum inhibitory concentration was then determined using a machine learning model. The input vector for the machine learning model was as follows:
[0154] Input = [corr(So.5,SGc), corr(Si,Scc), corr(S2,Scc), corr(S4,Scc)] = [0.882, 0.625, 0.260, 0.086]
[0155] A neural network was used as the machine learning model, trained to predict the logarithmic MIC value based on similarity scores. The neural network had 4 neurons in the input layer, 20 neurons in the first hidden layer, 10 neurons in the second hidden layer, and 1 neuron in the output layer. Figure 6 shows a schematic representation of the neural network. The ReLU function was used as the activation function.
[0156] f(x) = max(0,x)
[0157] used.
[0158] The LBFGS algorithm (Limited-memory Broyden-Fletcher-Goldfarb-Shanno) was used for weight optimization, and a value of BRE0627PT25WO1 was obtained on May 4, 2026.
[0159] - 19 -
[0160] a = 0.1 with I?-norm.
[0161] After training, the neural network had the following weights:
[0162] Weights of the input layer
[0163] Weights from input neuron 1 to the first hidden layer:
[0164] [-0.47773198 -0.05018906 -0.67626002 -0.014914080.06196212
[0165] 0.39282893 -0.018132150.01269077 -1.93151918 -0.08305675 -0.01696294 0.24694372 -0.281807110.65016068 -0.04063939
[0166] 0.73951461 -1.25392446 -0.09347303 -0.07216211 -0.07963627]
[0167] Weights from input neuron 2 to the first hidden layer:
[0168] [-0.21888013 -0.02833945 -1.04127891 -0.03871664 -0.019501150.2252032 1.160691320.036904790.057837560.071242060.02534972
[0169] 1.40285107 -0.53596816 -1.20327852 -0.06321619 -0.08104873 -0.0173391 -0.1038 1477 -0.01801883 -0.04903402]
[0170] Weights from input neuron 3 to the first hidden layer:
[0171] [0.278796370.02122988 -1.379379920.05642066 -0.066956760.04185015 0.904781190.03399985 -0.62560687 -0.02427449 -0.00781835 1.73315874 1.606635141.23365085 -0.04480957 -0.9442924 -0.85893164 -1.12295707 0.03032549 -0.0803454 ]
[0172] Weights from input neuron 4 to the first hidden layer:
[0173] [-0.08170436 -0.06820398
[0174] 0.31699842 -0.0036324 -0.05759175 -0.09785439 -0.85300936 -0.02113118 -0.4457 227 0.048509790.04304712 -0.29182797 -0.648911790.46140722 0.03490493 -0.20575495 -0.110816420.62408522 -0.01032585 -0.05592846]
[0175] Weights of the first hidden layer
[0176] Weights of neuron 1 in the first hidden layer to the second hidden layer: BRE0627PT25WO1 May 4, 2026 - 20 -
[0177] [ 0.044227420.06292371 -0.59881102 -0.03906604 -0.02578889 0.32727348 -0.039861340.036200940.36506189 0.07255671]
[0178] Weights of neuron 2 in the first hidden layer to the second hidden layer:
[0179] [ 0.066256820.051545970.07398316 -0.066713210.0721988 -0.05088941 0.045365370.007861490.051530640.03159963]
[0180] Weights of neuron 3 in the first hidden layer to the second hidden layer:
[0181] [ 0.167003190.35626596 -0.79922529 -0.007091720.03708306 1.43635824 -0.06593015 -0.60436251 -0.133454020.05122351]
[0182] Weights of neuron 4 in the first hidden layer to the second hidden layer:
[0183] [ 0.01879819 -0.047941380.00784031 -0.06891221 -0.01565116 -0.03737086 0.002510160.01790656 -0.03753954 -0.05111637]
[0184] Weights of neuron 5 in the first hidden layer to the second hidden layer:
[0185] [ 0.03865282 -0.00773497 -0.004773890.02683823 -0.00253659 0.04191734 -0.040493 0.011643980.02808164 0.02340144]
[0186] Weights of neuron 6 in the first hidden layer to the second hidden layer:
[0187] [ 0.07198728 -0.1379997 0.177049610.06161195 -0.05911077 0.01665186 -0.05839159 -0.02409531 -0.395492380.03262203]
[0188] Weights of neuron 7 in the first hidden layer to the second hidden layer:
[0189] [ 0.31665509 0.03367325 -0.89537078 0.02025147 -0.01416891 1.02828792 -0.02066665 0.60084195 0.31605301 0.02226608]
[0190] Weights of neuron 8 in the first hidden layer to the second hidden layer:
[0191] [ 0.02182855 -0.03014361 -0.03241931 -0.0358013
[0192] 0.06458324 -0.01564164 -0.04798881 -0.01174262 -0.000762320.00324101]
[0193] Weights of neuron 9 in the first hidden layer to the second hidden layer: BRE0627PT25WO1 May 4, 2026 - 21 -
[0194] [ 0.025238440.00592176
[0195] 0.45219954 -0.02093533 -0.07246169 -1.25038307 -0.07192419 -1.33070837 -0.644 00269 -0.16441998]
[0196] Weights of neuron 10 in the first hidden layer to the second hidden layer:
[0197] [ 0.05010319 -0.018698170.01679948
[0198] 0.02083927 -0.07398798 -0.06818286 -0.01540851 -0.065851270.01193305 0.06440702]
[0199] Weights of neuron 11 in the first hidden layer to the second hidden layer:
[0200] [-0.0050439 0.05731856 -0.02235925 0.01146723 -0.04058439
[0201] 0.04160239 -0.06690342 -0.02262739 -0.05283536 0.03861159]
[0202] Weights of neuron 12 in the first hidden layer to the second hidden layer:
[0203] [ 0.04938222 -0.3381731 0.57278294 0.00786923 -0.04159644 -1.58355723 0.01953901 0.1097866 1.17545997 0.06283598]
[0204] Weights of neuron 13 in the first hidden layer to the second hidden layer:
[0205] [-0.06242463 -0.19591368 0.40567214 0.05922391 -0.0280703 -0.8632999 0.05481847 -1.47544023 0.89244878 0.01527253]
[0206] Weights of neuron 14 in the first hidden layer to the second hidden layer:
[0207] [ 0.24881212 -0.0070519 -0.80911807 0.11064268 0.06185366
[0208] 1.3940526 -0.07202201 -0.84772712 0.13576782 0.11711913]
[0209] Weights of neuron 15 in the first hidden layer to the second hidden layer:
[0210] [ 0.02705446 -0.06255003 -0.07219998 -0.04303825 -0.00507495 -0.05060111 0.02950832 -0.01609288 0.04055721 -0.00125561]
[0211] Weights of neuron 16 in the first hidden layer to the second hidden layer:
[0212] [-0.01025338 -0.10521135 0.22307914 0.02148611 -0.06632842
[0213] 0.64493386 -0.03773846 0.92284169 -0.19119532 0.003309 ]BRE0627PT25WO1 May 4, 2026
[0214] - 22 -
[0215] Weights of neuron 17 in the first hidden layer to the second hidden layer:
[0216] [ 0.17117525 0.17859743 0.27882908
[0217] 0.12902673 -0.06005163 -1.05364227 -0.0555476 -0.5855268 -0.95790486 -0.1249 4064]
[0218] Weights of neuron 18 in the first hidden layer to the second hidden layer:
[0219] [ 0.07180012 0.19716315 -1.0553019 -0.00506253 -0.04642589 -0.06709972 0.0214276 -0.03006463 0.03180223 0.00339743]
[0220] Weights of neuron 19 in the first hidden layer to the second hidden layer:
[0221] [ 1.46645038e-02 -6.46348807e-02 -9.08979438e-05
[0222] 6.55390159e-02 -6.80276210e-02 5.75328760e-02 4.68242797e-02 7.55011081e-03 -6.96959408e-02 5.73716723e-02]
[0223] Weights of neuron 20 in the first hidden layer to the second hidden layer:
[0224] [-0.06670055 -0.0447673 -0.04840024 0.0191997 0.03971345 -0.00651285 0.0060355 0.0327909 0.05376227 0.03932645]
[0225] Weights of the second hidden layer
[0226] Weights of the second hidden layer to the output neuron:
[0227] [0.32722869, 0.34827927, -1.8392078, 0.03289302, 0.05299151,
[0228] 3.26127429, -0.04362894, 2.2677370, 2.1366512, 0.01883907]
[0229] With the input
[0230] Input = [corr(So.5,SGc), corr(Si,Scc), corr(S2,Scc), corr(S4,Scc)] = [0.882, 0.625, 0.260, 0.086]
[0231] This trained neural network produced an output of
[0232] Output_NN = [-0.173]
[0233] Since it is common practice for the clinical assessment of antibiotic resistance to test antibiotic concentrations in exponential steps to base 2 (i.e., 0.5, 1, 2, 4, 8, ...), it is recommended to use the neuronal BRE0627PT25WO1 May 4, 2026
[0234] - 23 -
[0235] To train a network from the logarithmic MHK values (to base 2), round the output value and take it as the exponent to base 2:
[0236] |y| |_| — 2 roun (Output_NN)
[0237] This results in a minimum carbide (MHK) of
[0238] M | C = 2 round(-0.173) = 2 0 = l mg / L
[0239] Investigation into the reliability and accuracy of the inventive method for determining a minimum inhibitory concentration
[0240] In a dataset of 672 samples for which the MIC (minimum inhibitory concentration) of the antibiotic levofloxacin was determined, an essential agreement (EIA) of 99.5% was achieved. The MIC value was determined correctly in 95.4% of the samples. A maximum deviation of one log unit was observed in 4.2% of the samples. A deviation of one log unit is the usual tolerance range within which MIC determination is permitted in a clinical setting and is therefore considered acceptable. Only in 0.45% of the samples was the MIC value outside the clinically acceptable range.
[0241] Example of shifting signals in mass spectra when determining similarity values
[0242] If S x ,o represents the original spectrum, then S x,i is the spectrum in which their signals (i.e., in the case of mass spectra, their m / z values) are shifted by i. To compensate for jitter, it is recommended to shift the signals of the growth control spectrum multiple times. When calculating the Pearson correlation, this yields several values for a given concentration. These values can be used individually as input for the machine learning model, or the maximum of these values, called the jitter-corrected correlation (JCcorr) here, can be calculated in a further step and used as input for the machine learning model. For S0.5, the JCcorr can be calculated as follows, for example:
[0243] JCcorr (So.5) = max(corr(So.5,SGc,-2), corr(S0.5,SGc,-i), corr(S 05 ,SGc,o), corr(S 05 ,SGc,i), corr(S 05 ,SGc,2))
[0244] In the formula above, the signals of the growth control spectrum were shifted in the range of -2 to 2, thus compensating for a maximum jitter of ±2 m / z. For higher jitter, it is recommended to increase the value and use signals with a greater shift. BRE0627PT25WO1 May 4, 2026
[0245] - 24 -
[0246] Referring to the above example for determining a minimum inhibitory concentration using the method according to the invention, the input for the machine learning model would thus be:
[0247] Input = [JCcorr (So.5), JCcorr (Si), JCcorr (S2), JCcorr (S4)]
[0248] Comparison of signal quality and resolution of spectrometric measurement data as time-of-flight data and as converted mass spectrometric data
[0249] Figure 7 shows a comparison of spectrometric data from a sample containing microorganisms, with the spectrometric data plotted as time-of-flight data (Figure 7B) and as data converted to m / z values (Figure 7A). Vertical dashed lines in the two spectra indicate the position of identical signals.
[0250] While the general pattern of the signals is similar in both spectra, it is evident that the time-of-flight data offers better signal quality and resolution. The range of the spectra containing the important signals (2000–10000 m / z; 25000–50000 TOF [ns]) is wider and exhibits better resolution in the plotted time-of-flight data. Furthermore, the signals are more pronounced in the plotted time-of-flight data, and closely spaced signals are easier to distinguish.
Claims
BRE0627PT25WO1 May 4, 2026 -25- Patent claims:
1. Method for determining a minimum inhibitory concentration, comprising the following steps: To prepare or provide a sample containing microorganisms; Pretreating two or more subsets of the sample containing microorganisms with different concentrations of an antimicrobial agent; Generating spectrometric measurement data from the sample comprising subsets of microorganisms pretreated with different concentrations of an antimicrobial agent; Creating or providing spectrometric measurement data from a subset of the sample comprising microorganisms that was not pretreated with the antimicrobial agent, as a growth control; Comparing the spectrometric measurement data of the sample comprising subsets of microorganisms pretreated with different concentrations of an antimicrobial agent with the spectrometric measurement data of the growth control and determining similarity values between the compared spectrometric measurement data; Determining the minimum inhibitory concentration using the determined similarity values; where The determination of the minimum inhibitory concentration is carried out using a machine learning model, and the determined similarity values are used as input for the machine learning model. and / or Determining the similarity values involves shifting one or more signals from at least one of the two spectrometric measurements being compared. BRE0627PT25WO1 May 4, 2026 - 26 - 2. The method of claim 1, wherein the sample comprising microorganisms comprises bacteria and / or fungi as microorganisms, preferably bacteria, particularly preferably exclusively bacteria, most particularly preferably exclusively bacteria of one family.
3. Method according to any of the preceding claims, wherein the antimicrobial agent is an antibiotic and / or an antifungal agent, preferably an antibiotic.
4. Method according to any of the preceding claims, wherein the machine learning model is a neural network.
5. Method according to any of the preceding claims, wherein the creation and / or provision of the spectrometric measurement data comprises the performance of TOF measurements, preferably MALDI-TOF measurements.
6. Method according to one of the preceding claims, wherein time-of-flight data, preferably time-of-flight data from MALDI-TOF measurements, are used for the comparison of the spectrometric measurement data.
7. Method according to claim 6, wherein the time-of-flight data to be compared are each calibrated beforehand using identical time-of-flight data which were recorded with a time-of-flight mass spectrometer, preferably with a MALDI-TOF instrument, from a reference sample of known composition.
8. Method according to claim 7, wherein the reference sample of known composition is preferably a bacterial test standard.
9. Method according to one of the preceding claims, wherein four, preferably eight, subsets of the sample comprising microorganisms are pretreated with different concentrations of an antimicrobial agent.
10. Analysis procedure, comprising the following steps: Preparing or providing a sample to be analyzed; Receiving flight time data from the sample to be analyzed or BRE0627PT25WO1 May 4, 2026 - TI - Performing one or more TOF measurements of the sample to be analyzed, so that time-of-flight data of the sample to be analyzed are obtained; Using the flight time data to characterize the sample being analyzed, wherein the time-of-flight data to be used for characterization are each calibrated beforehand using identical time-of-flight data which were recorded with a TOF device from a reference sample of known composition.
