Method for analyzing motion activity of particles

The time-dependent signals when particles come into contact with flexible support are analyzed through motion detectors and linking algorithms, and the problem of distinguishing particle motion activities in the prior art is solved, and more accurate analysis of particle motion activities and bacterial susceptibility recognition is achieved.

CN118984932APending Publication Date: 2024-11-19REICHSTEL AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202380028119.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2022-03-15
Filing Date
2023-03-06
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

Prior art When analyzing the movement activities of particles, especially under similar states or complex conditions, it is difficult to effectively distinguish the movement activities of particles, and a single statistical parameter is not sufficient to distinguish different states, such as susceptible strains and resistant strains of bacteria. .

Method used

Using a motion detector, the generated time-dependent signals are detected through the flexible support, and the time dependence of the signal is analyzed using mathematical methods, and a linking algorithm is performed to extract the particle's motion activity indicators, including the selection and evaluation of signal parameters. to improve the accuracy of the analysis.

Benefits of technology

The ability to analyze particle motion activities has been improved, and the ability to more accurately distinguish particle motion activities under similar states or complex conditions has been improved, and the ability to identify bacterial susceptibility and metabolic activities has been improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118984932A_ABST
    Figure CN118984932A_ABST
Patent Text Reader

Abstract

A method of analyzing the motion activity of particles with a motion detector (1) comprises the steps of: bringing at least one particle into contact with a flexible support (2); detecting at least one time-dependent signal indicative of the deflection of the flexible support (2) due to the movement of the at least one particle; and evaluating the detected time dependent signal. The evaluation includes analyzing a time dependency of the signal to derive from the signal a plurality of signal parameters characterizing a change in the signal according to time; and performing a linking algorithm having as an input variable an input vector comprising at least a selection of said signal parameter, and having as an output variable at least one activity indicator indicative of a motion activity of the particle.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to methods for analyzing the motion activity of particles using motion detectors. In particular, the present invention relates to the analysis of the motion activity of particles that exhibit motion at very low scales (e.g., having a size range from angstroms to micrometers). The present invention specifically, but not exclusively, relates to the analysis of the movement or internal dynamics of the particles. Prior art

[0002] Nanomechanical oscillators have been widely used as sensors for small changes in the mass of detected particles. In WO 2013 / 054311, micro-cantilevers are used to detect and measure the nano-level vibrations (or nano-motions) of particles attached to their surfaces. The particles can be chemical or biological properties, including proteins, lipids, glycosides, viruses, bacteria, yeasts or mammalian cells, etc. The system allows to distinguish between the states of the test particles in the presence or absence of stimulation, thereby allowing, for example, antibiotic susceptibility detection. For this reason, the variance of the nano-level vibrations of micro-cantilevers has been analyzed, wherein the increase in variance has been demonstrated to be induced by the particles Vibration or motion.

[0003] However, using only the variance, i.e. a single statistical parameter, is not sufficient in every case to distinguish between different states. For example, in the measurement records of susceptible and resistant strains of particles as bacteria, the populations often overlap. Therefore, a better way of separating the populations is needed. Furthermore, living and dead cells constitute an example of extreme differences in metabolic activity, where, however, there is also a need to distinguish between less extreme cases or to analyze other motor activities of the particles such as respiration levels, growth rates, etc. Summary of the invention

[0004] The object of the present invention is to provide a method capable of improving the analysis of the motion activity of particles. In particular, the object is to provide a method capable of improving the analysis of the motion activity of particles in similar states or under complex conditions.

[0005] This object is achieved with a method according to claim 1. That is, a method for analyzing the motion activity of particles with a motion detector is provided. The motion detector comprises: a flexible support configured to deflect (such as bend); a detection device for detecting a signal generated when the flexible support deflects; and an evaluation device for evaluating the detected signal. The method comprises the steps of: i) bringing at least one particle into contact with the flexible support, ii) detecting at least one time-dependent signal indicating a deflection of the flexible support due to the motion of the at least one particle using the detection device, and iii) evaluating the detected time-dependent signal with the evaluation device. Preferably, steps i) to iii) are performed continuously and in the order described. The evaluation of the detected time-dependent signal comprises: a) analyzing the time dependency of the signal to derive from the signal a plurality of signal parameters characterizing the variation of the signal according to time, and b) executing a linking algorithm having as input variable an input vector comprising at least a selection of said signal parameters, and having as output variable at least one activity indicator indicating the motion activity of the particle. Preferably, steps a) and b) are performed continuously and in the order described, i.e. step b) is preferably performed after step a).

[0006] The present invention is based on the recognition that particles in contact with a flexible support lead to random signals, i.e. signals that vary in time but do not follow a simple oscillation pattern, so that at first glance, they appear to be random, or in other words, the signals are buried in a noise background, measurement artifacts, etc., while comprising useful information about the motion activity of the particles, and wherein said information can be extracted using mathematical methods. In fact, the present method enables the extraction of said information by analyzing the temporal variation of the signal properties. That is, in the method according to the present invention, mathematical methods are used to analyze time-dependent signals. Mathematical methods can be applied to signals detected in the time domain and / or to transformed signals obtained by transforming the detected signals into the frequency domain. In any case, the signals are detected at multiple time points to obtain the temporal dependency of the signals, and the temporal dependency is analyzed to determine time-independent signal parameters. These signal parameters characterize the temporal variation of the signals. Then, at least a selection of these signal parameters is fed as input variables to a linking algorithm, and the linking algorithm provides at least one output variable, which is based on the input variables and is at least one activity indicator indicating the motion activity of the particles.

[0007] The flexible support may be a cantilever, preferably an AFM cantilever, a fiber such as a hollow fiber or a glass fiber, a membrane, a wire, a sponge, a flexible electrode, etc. In addition, it is preferred that the particles are attached to the surface of the flexible support before being analyzed. The attachment is preferably carried out according to methods known in the prior art, such as by functionalizing the surface of the flexible support, respectively, or by using attachment compounds known in the art. It is particularly preferred that the particles are attached to the surface of the flexible support as described in WO 2021 / 130339, which is incorporated herein by reference. That is, it is preferred that the particles are dispersed in a solution comprising at least one of a gelling agent, a gelling agent and a thickening agent, and wherein the dispersion is subsequently added to the surface of the flexible support.

[0008] Furthermore, the motion detector preferably corresponds to a nano-scale motion detector, as disclosed in WO 2013 / 054311, which is also incorporated herein by reference. Therefore, the deflection of the flexible support is preferably measured by an optical level detection method, wherein a change in the position of a light spot caused by laser light reflected from the surface of the flexible support is measured with an optical detector such as a photodetector, particularly preferably a position-sensitive photodetector. The change in the position of the light spot results in a change in the signal recorded by the detector. As will be explained in more detail further below, the signal is preferably processed before the signal parameters are derived from the signal. Therefore, the signal may also be referred to as a raw input signal (RIS).

[0009] The method preferably includes: in the step of analyzing the time dependency of the signal, defining a plurality of time intervals, wherein the time dependency of the signal is analyzed in each time interval respectively to obtain the value of each of a plurality of signal estimates for each time interval, and wherein the time series of each signal estimate is analyzed to obtain each of the signal parameters.

[0010] The signal estimate may be determined in the time domain or in the frequency domain. In the latter case, it is preferred that the power spectrum is estimated for each time interval of the signal.

[0011] Each signal estimate value is preferably determined by fitting a noise model to the power spectrum of the signal and / or by applying at least one statistical algorithm to the signal. The signal estimate is preferably an estimator of a moment of a probability density function and / or a geometric property of a power spectrum density function and / or a percentile of a probability density function and / or a correlation parameter and / or a partial correlation parameter and / or a parameter of an autoregressive moving average model and / or a parameter of a nonlinear autoregressive moving average model.

[0012] The noise model is preferably selected from a flicker noise model, a white noise model, a band noise model or a combination thereof. The flicker noise model is preferably expressed as 1 / f α The weighted sum of the functions, where f represents the frequency and alpha (α) is a free parameter. The white noise model is preferably represented as a constant or a constant modulated by the frequency response of the motion detector. The band noise model is preferably represented as a rational function of an order that is preferably optimized for a specific application. For example, in the susceptibility test of E. coli strains, a second-order band noise model can correspond to a model based on a damped resonator. Examples of geometric properties of the power spectral density function are extreme points, inflection points, areas under different parts of the curve, or horizontal and vertical distances between them.

[0013] About the estimator of the moment of probability density function, it should be noted that they can have moments of different orders, they can be standardized or not standardized, etc. For example, the estimator of the moment of probability density function can be variance, skewness, kurtosis, etc. In addition, it is preferred that they are optimized for specific applications, for example, variance can be used for Escherichia coli antibiotic susceptibility detection. About the percentile of probability density function, it should be noted that they can have different percentile rankings. For example, percentile ranking can be the 25th percentile, the 50th percentile (median), the 75th percentile, etc. In addition, it is preferred that percentile ranking is optimized for specific applications, for example, median can be used for metabolic activity classification.

[0014] As already outlined above, signal analysis can be based on signal parameters associated with different noise models, wherein, for example, signal estimates are determined by fitting the noise model to the power spectral density function of the signal. However, these theoretical models are only as useful as practice follows theory. In order to correctly analyze situations where the estimated spectrum deviates from the theoretical model, the geometric properties of the spectrum are helpful.

[0015] Therefore, analyzing the time dependence of a signal using the power spectral density function is a useful method for analyzing the properties of random signals in the frequency domain.

[0016] Thus, in the method according to the invention, each signal estimate may be a geometric property of a power spectral density function.

[0017] Various methods of determining the power spectral density function are conceivable, wherein these methods are well known in the art. For example, the power spectral density function may be determined by the Welch method.

[0018] The geometric attribute can be an extreme point of a power spectral density function, an inflection point of a power spectral density function, an area under a curve described by a power spectral density function, or a horizontal distance and / or a vertical distance between them. An example of an extreme point of a power spectral density function is a global maximum or a local maximum or a global minimum or a local minimum of a power spectral density function. An inflection point of a power spectral density function is understood as a point at which the power spectral density function changes the direction of its curvature. The area under a curve described by a power spectral density function is understood as an area under the power spectral density function between two points on the power spectral density function. The area is preferably determined by integrating the power spectral density function between the two points. For this purpose, any two points on the power spectral density function can be used. The horizontal distance and / or the vertical distance between them is, for example, a horizontal distance and / or a vertical distance between extreme points or between inflection points.

[0019] Signal parameters derived from such signal estimators can be considered as spectral signal parameters. Quantile signal parameters are generalizations of spectral signal parameters. As just outlined above, spectral signal parameters can be calculated based on the power spectral density function. The power spectral density function of a signal is estimated as the average of the signal periodogram, where the periodogram is the squared amplitude of the Fourier transform of the signal segment. The power spectral density function is statistical information. This means that the power spectral density function reveals the properties of a random signal (random process). If the average of the periodogram is used, some occasional but high-amplitude events degrade the statistical information. For this reason, for example, robust statistical information such as the median is used very frequently. However, in cases where these occasional events carry useful information, it is found that not only the median but also the probability distribution of the periodogram is useful to capture the information.

[0020] Therefore, in the method according to the invention, at least one signal estimate can be obtained by determining a plurality of periodograms for each time interval. For all periodograms, at least one quantile is determined for at least one frequency range. The quantile is preferably a percentile. These signal parameters can be referred to as quantile signal parameters.

[0021] Quantiles, in particular percentiles, may be determined for various frequency ranges. For example, the frequency range for which the quantiles or percentiles of the periodogram are estimated may be 0 Hz to 10 Hz, 10 Hz to 100 Hz, etc., although other frequency ranges are equally conceivable. In other words, for each periodogram, a quantile such as a percentile is determined for a specific frequency range. Thus, a set of time series of signal estimates is obtained, which is a set of time series of quantiles of the periodogram, in particular quantiles such as percentiles for a specified frequency range.

[0022] The set of time series may be extended, and wherein at least one signal parameter is obtained from the extended time series. In other words, the time series of signal estimates may be extended by combining the time series of signal estimates, and wherein the signal estimate is obtained from the combined time series.

[0023] In particular, the time series of at least one of the signal estimates may be extended by combining at least two signal estimates as a difference and / or a ratio thereof to form an extended time series.

[0024] Additionally or alternatively, the time series of at least one of the signal estimators can be expanded by transforming multiple quantiles, particularly percentiles, into the value of an empirical probability density function to form an expanded time series. In particular, all quantiles, such as all percentiles, can be transformed into the value of an empirical probability density function (EPDF), and wherein, two consecutive quantiles, such as percentiles, are used to calculate an EPDF value. In addition, at least one distance between the value of the empirical probability density function and the value of the theoretical probability density function can be determined. For example, the distance between all EPDF values ​​and the theoretical value of the chi-square probability density function can be measured. Examples of distances are Kullback-Leibler divergence and Jensen-Shannon distance, although other distances are also conceivable. Additionally or alternatively, at least one geometric property of the empirical probability density function and / or various moments (mean, variance, skewness, kurtosis, etc.) of the EPDF can be determined.

[0025] At least one signal parameter can be derived from the time series of at least one of the signal estimates and / or from the extended time series of at least one of the signal estimates by the following operations: i) defining a plurality of subsequences for the time series and / or for the extended time series and determining statistical properties for each subsequence and preferably also determining at least one of their ratios, differences and ratios of differences, and / or ii) fitting one or more parameterized curves comprising one or more parameters to the time series and / or the extended time series, and / or iii) defining a plurality of subsequences for the time series and / or for the extended time series and fitting one or more parameterized curves comprising one or more parameters for each subsequence to obtain fitting parameters and preferably also determining their ratios, differences and ratios of differences.

[0026] For example, conceivable statistical properties are percentiles and moments. Conceivable curves are exponential curves, polynomial curves, trigonometric curves and linear and non-linear mixtures thereof.

[0027] The power spectral density function of the signal of some particles reveals a strong 1 / f characteristic. Such a spectrum is often associated with the signal self-affinity. The self-affinity can be analyzed by detrended fluctuation analysis (DFA) and its generalization multifractal detrended fluctuation study (MF-DFA). In these methods, the dependence of the statistical properties of the signal on the time scale is studied.

[0028] Therefore, in the method according to the invention, the evaluation of the detected time-dependent signal with the evaluation device may comprise detecting the time-dependent signal during the time t, and wherein analyzing the time dependency of the signal preferably comprises the following steps:

[0029] - divide the signal at time t into time intervals of length T, and

[0030] - For each time interval j=1, ...t / T, perform the following steps:

[0031] ○Calculate the digital integral of the signal in each time interval;

[0032] ○ dividing the time interval into at least a number of sub-intervals;

[0033] o For each subinterval, detrending the digital integral of the signal by applying a detrending algorithm to the digital integral of the signal;

[0034] ○ For at least one number of subintervals, combine the subintervals into detrended intervals;

[0035] as well as

[0036] o Determining at least one generalized mean of the detrended signal of the detrended interval for at least one number of subintervals, thereby obtaining at least one signal parameter and / or initial signal parameter.

[0037] A digital integration is applied to the signal in each time interval. Therefore, the signal in each time interval becomes an integrated signal. Any digital signal x of N samples i The numerical integral of can be defined by the following equation,

[0038] For each n=1..N,

[0039] Preferably, the detrending algorithm fits a polynomial of any order (i.e., second order) to the integrated signal, and further subtracts the fit from the integrated signal. That is, the fitting residual becomes the detrended integrated signal. Preferably, detrending is applied to each subinterval separately. Therefore, all subintervals become detrended subintervals.

[0040] The time interval is divided into a certain number of subintervals, and the integrated and detrended signals of all subintervals are combined back to one signal of the entire time interval (e.g., a 20 minute interval of the signal is integrated, and thereafter it is divided into four 5 minute subintervals, each of the subintervals is detrended, and all subintervals are combined back to a 20 minute interval signal). The result of the detrending depends on the number of subintervals into which the signal time interval is divided. In other words, the result of the detrending depends on the length of each subinterval.

[0041] At least one generalized mean of the signal time interval is calculated (integrating at least one number of subintervals, detrending and combining at least one number of subintervals back together) (e.g., F2(128) represents a generalized mean with index 2 and 128 subintervals).

[0042] Any digital signal x of length N i The generalized mean of with non-zero exponent / power q is defined by the following equation:

[0043]

[0044] Any signal x of length N i The generalized mean of with zero exponent / power q is defined by the following equation:

[0045]

[0046] The generalized mean with exponent 2 is just the standard deviation, the generalized mean with exponent 0 is the geometric mean.

[0047] Signal parameters are preferably obtained from generalized means and / or ratios and / or differences thereof.

[0048] In particular, a signal parameter may be derived from other signal parameters, such as by combining two or more signal parameters as a ratio and / or difference thereof, wherein the resulting signal parameter becomes independent of certain factors, such as environmental variations.

[0049] The method preferably further comprises executing a feature selection algorithm to automatically select a subset of the signal parameters to obtain an input vector fed to the linking algorithm.The feature selection algorithm is preferably configured to select the subset of signal parameters in such a way that the subset is optimal for multiple objective criteria.

[0050] Preferred feature selection algorithms implement univariate selection such as F-test, mutual information, chi-square, penalty-based methods such as L1, L2 and elastic net, and forward or backward feature selection based on properties of specific linkage algorithms (such as coefficients in logistic regression or importance in tree-based algorithms) or based on various forms of cross-validation or bootstrapping with multi-objective criteria.

[0051] The multi-objective criterion is preferably at least one of accuracy, sensitivity, specificity, mean square error, mean absolute error, root mean square error, portion of unexplained variance, coefficient of determination, number of signal parameters or generalization gap.

[0052] For example, the input vector may include the optimal signal parameters in the Pareto sense. The optimal in the Pareto sense is understood as a non-dominated solution for which the improvement of all standards is impossible. In other words, improving one standard will degrade at least one other standard. As an example, when the feature selection algorithm corresponds to a forward selection with a cross-validation algorithm, the selection of signal parameters may occur when the cross-validation score does not improve in the Pareto sense. Additionally or alternatively, the input vector may include the optimal signal parameters relative to a single standard including, for example, a linear combination of two or more standards in all standards. The use of a linear combination of two or more standards may be understood as capturing several aspects of the evaluation of the link algorithm. Additionally or alternatively, however, under constraints, such as constraints defined by experts in the field or constraints on other standards such as specificity not less than 80%, the input vector may include the optimal signal parameters relative to one of the above-mentioned standards. These constraints may be defined after the analysis of the Pareto optimal solution.

[0053] Before deriving signal parameters from the signal, drift cancellation is preferably applied to the signal. That is, drift cancellation is preferably applied to the initially mentioned original input signal or RIS. In other words again, the signal estimate is preferably determined after drift cancellation.

[0054] Drift cancellation preferably includes a curve fitting process or high pass filtering to model the drift component of the signal. The drift cancellation is used to eliminate the drift of the motion detector that manifests itself in the detected signal. The drift may be caused by various factors such as temperature, humidity, state changes of the surrounding fluid, etc.

[0055] For example, drift cancellation can be performed by dividing the signal, in particular the RIS, into intervals and then applying a curve fit to the signal to model the drift component of the drift. Examples of curves to be fitted are polynomials, exponential functions, trigonometric functions or mixtures thereof. The fitted curve can be defined in linear and nonlinear form. Preferably, the type of curve, the number of fitting parameters and the length of the time interval are optimized for a specific application. The fitting error is preferably used in the next steps of the analysis, in particular in the calculation of the time series of the signal estimate.

[0056] However, another example of drift cancellation consists in applying the curve fitting just described, in a moving window regime, and with continuously adjusted fitting parameters.The fitting error is preferably used in the next steps of the analysis, in particular in the calculation of the time series of signal estimates.

[0057] In this example, the fit is preferably repeated for each single point of the signal, in particular the RIS, while using a buffer, such as a circular buffer, storing the latest samples of the signal. The fit is preferably modified to efficiently update the curve fit parameters, so that millions of signal points can be processed. The type of curve, the number of fit parameters and the length of the buffer are also preferably optimized for the specific application.

[0058] Another example of drift removal includes applying high pass filtering to the signal, in particular the RIS, and wherein the high pass filtering is preferably of a specific order, type and parameters including phase shift. Examples of this type are Butterworth, Chebyshev I and II, Kaiser, elliptic, etc. The filter output is preferably used in the next steps of the analysis, in particular in the calculation of the time series of signal estimates.

[0059] The drift removal algorithm is preferably a detrending algorithm. The detrending algorithm is preferably configured to detrend the signal.

[0060] In particular, in case the at least one signal parameter is a spectral signal parameter as described earlier, it is particularly preferred to apply a detrending algorithm to the signal before analyzing the time dependency of the signal within the time interval.

[0061] In the case of determining at least one signal parameter as a signal estimator based on a quantile, such as a percentile, of a probability density function as a periodogram, it is particularly preferred that a detrending algorithm is applied to the signal before determining the periodogram and wherein a plurality of periodograms are determined for the detrended signal for each time interval.

[0062] For example, the detrending algorithm may be a fitting algorithm that fits a k-th order polynomial to the signal and subtracts the fitted polynomial from the signal, thereby detrending the signal. By detrending the signal, the drift of the signal may be removed.

[0063] The linking algorithm is particularly preferably a regression algorithm or a classification algorithm.

[0064] In case the linked algorithm is a regression algorithm, the multi-objective criterion mentioned earlier is preferably at least one of the mean square error, mean absolute error, root mean square error, part of the unaccounted for variance, coefficient of determination etc., number of signal parameters or generalization gap.

[0065] In case the linking algorithm is a classification algorithm, the multi-objective criterion mentioned earlier is preferably at least one of accuracy, sensitivity, specificity, etc.

[0066] It is further preferred that the linking algorithm is a machine learning (ML) algorithm.

[0067] In some embodiments, the ML algorithm may be an ML regression algorithm, specifically, an ML algorithm implementing linear regression, decision tree regression, random forest regression, gradient boosted tree regression, kernel regression, multi-perceptron neural network, and RBF neural network regression. Such algorithms are well known in the art.

[0068] In other embodiments, the ML algorithm can be an ML classification algorithm, specifically, an algorithm that implements logistic regression, decision tree, random forest, gradient boosting tree, support vector machine, multi-perceptron neural network, and radial basis function (RBF) neural network. These types of algorithms are also well known in the art.

[0069] Many other suitable ML algorithms are known in the art and may be employed.

[0070] The particle is preferably a biological object and / or a part of a biological object and / or a non-biological object and / or a part of a non-biological object. Preferred biological objects are cells such as prokaryotic cells or eukaryotic cells, spores, viruses, bacteriophages, and substances of biological origin such as vesicles, peptides, proteins, polysaccharides, lipids, glycosides, nucleic acids and copolymers, protein-RNA copolymers, RNA-DNA copolymers, protein-DAN copolymers, RNA / DNA-protein copolymers, protein-protein copolymers or capsules. Preferred parts of biological objects are organelles. Preferred non-biological objects are proteins, lipids, nucleic acids such as DNA and nanodevices. Cells are preferably living cells and / or organelles are preferably living organelles. Cells can be prokaryotic cells and / or eukaryotic cells. Cells can also be motile cells and / or non-motile cells. Cells are preferably bacterial cells and / or fungal cells such as yeast cells, and / or mammalian cells and / or insect cells and / or plant cells. Organelles are preferably mitochondria and / or nuclei or any other subcellular structures.

[0071] The activity indicator preferably indicates at least one of the following: metabolic activity of the particle, physiological state of the particle, respiration level of the particle, ATP level of the particle, NADH / NAD + ratio of particles, membrane potential of particles, growth rate of particles, killing rate of particles, interaction of particles with other particles, and interaction of particles with environmental factors of particles, or any combination of these.

[0072] Environmental factors can be chemical in nature, for example, metabolites or drugs such as antibiotics, toxins, denaturants or any other chemical compounds that affect the activity of particles. However, environmental factors can also be physical in nature, for example, temperature, humidity, viscosity, oxygen level, redox potential, radiation, pressure or any other influence that affects the activity of particles. The interaction of particles with environmental factors can be understood as the reaction of particles to environmental factors. The metabolic activity of particles can be respiration, energy family (e.g., ATP, NADH, NAD+ levels), membrane potential, activity of ion channels, activity of cytoskeleton, etc. The physiological state of particles can be a response to one or more of the above environmental factors. The interaction of particles with other particles can be killing, proliferation, predation, mating, quorum sensing or conformational changes of membrane complexes and / or protein complexes and / or other cell complexes and / or macromolecular complexes.

[0073] The particles are preferably subjected to at least one chemical stimulus and / or at least one physical stimulus. The chemical stimulus is preferably the addition of a drug such as an antibiotic or a compound that affects the metabolism or viability of the particles (such as cells). The chemical stimulus may also be a change in environmental conditions such as a change in culture conditions, etc. The physical stimulus is preferably the application of stress.

[0074] The activity index is preferably determined before and / or during and / or after the particles are subjected to a chemical and / or physical stimulus. Furthermore, it is preferred that the activity indexes thus determined are compared with one another.

[0075] For example, if one is analyzing the effectiveness of a drug on a particle, such as the effectiveness of an antibiotic on bacteria, an activity indicator can be determined before the antibiotic is added to the bacteria and after the antibiotic is added to the bacteria. In the case where the antibiotic is effective, a difference in the activity indicator will be determined. For example, killed bacteria will no longer exhibit metabolic activity. In addition, dying or susceptible bacteria may change their metabolic activity after being exposed to the antibiotic. That is, the bacteria may not necessarily die immediately.

[0076] In another aspect, a method for generating a training data set for a machine learning algorithm is provided. The method comprises the following steps:

[0077] a) bringing at least one particle into contact with a flexible support,

[0078] b) detecting, using a detection device, at least one time-dependent signal indicative of a deflection of the flexible support due to the movement of at least one particle, and

[0079] c) evaluating the detected time-dependent signal with an evaluation device, wherein the evaluation of the detected time-dependent signal comprises analyzing the time dependency of the signal in order to derive from the signal a plurality of signal parameters characterizing the temporal variation of the signal,

[0080] d) performing independent measurements to derive at least one associated activity indicator indicative of the motion activity of the particles,

[0081] e) storing the plurality of signal parameters and at least one associated activity indicator in a training data set, and

[0082] f) Repeating steps a) to e) for a plurality of particles exhibiting different motor activities.

[0083] Therefore, the training data set includes multiple signal parameters and associated activity indicators of multiple particles.

[0084] In another aspect, a method for training a machine learning algorithm is provided, wherein the machine learning algorithm is trained using the training data set mentioned above.

[0085] That is, the machine learning algorithm is trained with the signal parameters and the associated activity indicators. Thus, the machine learning algorithm can make predictions about the activity indicators based on the signal parameters, and vice versa, without the need to understand the particles and their motion activities.

[0086] For example, training can be based on a training data set comprising experimental results detected by the detection means of the motion detector. This training data set is then preferably labeled by the results of a reference experiment that uses a reference method to measure, for example, a metabolic activity of interest, such as the susceptibility of a strain measured by a Kirby-Bauer test. The labeled data is then used to select signal parameters, optimize all stages of the linking algorithm, and evaluate performance using a cross-validation technique. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Preferred embodiments of the present invention are described below with reference to the accompanying drawings, which are for the purpose of illustrating the current preferred embodiments of the present invention, rather than for the purpose of limiting the present invention. In the accompanying drawings,

[0088] Figure 1 A schematic diagram showing preferred steps of the method according to the present invention;

[0089] Figure 2 shows a signal detected by a motion detector which has been subjected to drift cancellation during evaluation by the method according to the invention;

[0090] Figure 3 shows a signal detected by a motion detector which, during evaluation with the method according to the invention, undergoes fitting of a noise model when determining a signal estimate;

[0091] Figure 4shows a graph depicting the accuracy of various algorithms for various numbers of signal parameters to identify a Pareto optimal feature selection algorithm suitable for use in accordance with the method of the present invention;

[0092] Figure 5a shows a schematic diagram of a motion detector comprising a cantilever with attached live bacteria (upper part) and a time series of signal estimates determined during evaluation with the method according to the invention (lower panel);

[0093] Figure 5b shows a schematic diagram of a motion detector comprising a cantilever with attached dead bacteria (upper part) and a time series of signal estimates determined during evaluation with the method according to the invention (lower panel);

[0094] Figure 6a Shown is depicted in Figure 5a and Figure 5b A graph of signal estimates determined during the evaluation period depicted in;

[0095] Figure 6b Shown is a depiction of Figure 5a and Figure 5b Graph of fluorescence over time determined for live and dead bacteria used for analysis as depicted in;

[0096] Figure 7a The following diagram is shown, which depicts the signal detected with a motion detector for E. coli strain ATCC-25922 under different conditions, and the signal estimate over time determined for this signal during evaluation with the method according to the invention;

[0097] Figure 7b The following diagram is shown, which depicts the signal detected with a motion detector for the E. coli strain BAA-2452 under different conditions, and the signal estimate over time determined for this signal during the evaluation with the method according to the invention;

[0098] Figure 8a The following diagram is shown, which depicts the signal detected with a motion detector for E. coli strain ATCC-25922 and E. coli strain BAA-2452 under different conditions, and other signal parameters determined during the evaluation with the method according to the invention;

[0099] Figure 8b Shows a depiction Figure 8a Graphs of linear combinations of other signal parameters;

[0100] Fig. 9The following diagram is shown, which depicts the linear combination of the signal detected with the motion detector for E. coli strain ATCC-25922 and E. coli strain BAA-2452 under different conditions, other signal parameters determined during the evaluation with the method according to the invention;

[0101] Fig.10 The following diagram is shown, which depicts the signal detected with a motion detector for a NARA-5032-resistant E. coli strain under different conditions, the signal estimate determined during the evaluation with the method according to the invention and the signal parameter - the median value of the time series of the signal estimate;

[0102] Fig.11 Shown is a depiction of Fig.10 Graphs of signal parameters determined by linear and exponential fitting of the signal estimators depicted in;

[0103] Fig.12 shows a table depicting the performance of a feature selection algorithm suitable for evaluation with the method according to the invention for different input vectors;

[0104] Fig.13a shows a graph depicting the correlation between the signal parameters of the method according to the invention and the kill rate determined with a standard microbiological method;

[0105] Fig.13b shows a graph depicting the correlation between other signal parameters according to the method of the invention and the kill rate determined with a standard microbiological method;

[0106] Fig.13c Shows a depiction Fig.13a and Fig.13b A graph showing the correlation between the linear combination of the signal parameters and the kill rate determined by the standard microbiological method;

[0107] Fig.14a shows a graph depicting a time series of signal estimates determined with the method according to the invention for Escherichia coli ATCC-25922 under different conditions;

[0108] Fig.14b shows a graph depicting a time series of signal estimates determined with the method according to the invention for Escherichia coli ATCC-25922 under different conditions;

[0109] Fig.15a Shows the depiction according to Fig.14a and Fig.14b Graphs of time series of determined signal parameters;

[0110] Fig.15b Shows the depiction according to Fig.14a and Fig.14b A graph of a time series of a combination of signal parameters determined by the graph;

[0111] Fig.16a Shows the depiction according to Fig.14a and Fig.14b Graphs of time series of linear combinations of other signal parameters determined;

[0112] Fig.16b Shown is depicted in Fig.14a and Fig.14b A graph of ATP (adenosine triphosphate) concentration, a marker of metabolic activity, of Escherichia coli ATCC-25922 under culture conditions measured in FIG. ;

[0113] Fig.17 shows a schematic diagram of a motion detector;

[0114] Fig.18a The distribution of Klebsiella pneumoniae strains according to their reference MIC for ciprofloxacin (CIP) is shown;

[0115] Fig.18b The performance of the chaining algorithm as a classification algorithm depending on the number of signal parameters (SP) is shown;

[0116] Fig.19 The performance of the classification model as a linkage algorithm for Klebsiella pneumoniae and ciprofloxacin with different numbers of SPs is shown;

[0117] Fig.20a Shown for training Fig.19 The score for each experiment of Model 1 was plotted against the reference MIC for ciprofloxacin using one SP;

[0118] Fig.20b Shown for training Fig.19 The scores for each experiment of Model 4 were plotted against the reference MIC for ciprofloxacin using the four SPs;

[0119] Fig.21a The distribution of E. coli strains and K. pneumoniae strains according to their reference MICs for ceftriaxone (CRO) is shown;

[0120] Figure 21b Shown for training Fig.24 The score for each experiment of the CRO model plotted against the reference MIC for ceftriaxone;

[0121] Fig.22a The distribution of E. coli strains according to their reference MIC for ciprofloxacin (CIP) is shown;

[0122] Figure 22b Shown for training Fig.24 The score for each experiment of the CIP model in is plotted against the reference MIC for the CIP;

[0123] Fig.23a The distribution of E. coli strains according to their reference MIC for cefotaxime (CTX) is shown;

[0124] Figure 23b Shown for training Fig.24 The score for each experiment of the CTX model in is plotted against the reference MIC of CXT;

[0125] Fig.24 Shows Figure 21b (CRO), Figure 22b (CIP) and performance of the classification model for the data shown in Figure 22c (CTX);

[0126] Fig.25a b shows the following diagram, which depicts the signal detected with the motion detector for the cancer cell line SW480 under different conditions (a, in the case of doxorubicin and b, medium control), and the signal estimate over time determined for this signal during the evaluation with the method according to the invention;

[0127] Fig.26 A graphical representation of the geometric properties of the power spectral density function of the vibration is shown. The geometric properties of the power spectral density function of the vibration are used to calculate the integrals N1 to N5, distances and ratios described. A line is fitted to the logarithmic plot using the section from 20 Hz to 200 Hz to determine N0 and alpha (slope and intercept of the logarithmic line fit);

[0128] Fig. 27 A comparison between the quantile spectrum and the power spectral density is shown. The 50th percentile spectrum (P50) is close to the classical power spectral density function (PSD). The other percentiles introduce additional information about the probability distribution of the Fourier components;

[0129] Fig.28 The empirical probability density function (EPDF) calculated based on the percentiles of the spectrum of the three Fourier components 10 Hz, 100 Hz and 1000 Hz is shown. As long as the transformed noise is Gaussian, the EPDF is a chi-square distribution, but the estimation of this statistic for the cell vibration signal reveals its discriminative power;

[0130] Fig.29 MF-DFA signal parameters used in the embodiment are shown. The MF-DFA signal parameters are calculated as the F qThus, the signal parameter "F0" is the F value for the last 20 minutes of the drug phase. -10 (1024) divided by the F of the first 20 minutes of the drug phase -10 (1024);

[0131] Fig.30 The dependence of the performance of the chaining algorithm as a classification algorithm on the signal parameters selected by the method described in the present invention in the classification problem of metabolic reactions to two antibiotics: ceftriaxone and ciprofloxacin is shown;

[0132] Fig.31 The dependence of the performance of the chaining algorithm as a classification algorithm on the signal parameters selected by the method described in the present invention in the classification problem of the metabolic response to the physical stressor - temperature is shown;

[0133] Fig.32 The dependence of the performance of the chaining algorithm as a classification algorithm on the signal parameters selected by the method described in the present invention in the classification problem of colon cancer cell metabolism in the presence and absence of antibiotics is shown;

[0134] Fig.33 The results of the application of the linking algorithm, which is a linear regression algorithm with quantile signal parameters, for predicting metabolic responses to different drug concentrations are shown. High values ​​of the determined coefficients indicate a strong dependence between the linking algorithm output and the actual values ​​of the drug concentrations;

[0135] Fig.34 The dependence of the performance of the link algorithm as a regression algorithm on the number of signal parameters selected by the method described in the present invention in the problem of estimating the quantitative metabolic response to increasing concentrations of an antibiotic is shown. DETAILED DESCRIPTION

[0136] Various aspects of the invention will now be illustrated with reference to the drawings.

[0137] As mentioned earlier, the present invention enables an improved analysis of the motion activity of particles using a motion detector 1 comprising a flexible support 2 in contact with the particles to be analyzed and configured to be deflected by the motion of the particles, see Fig.17 . Further according to Fig.17, the motion detector 1 comprises a radiation source, such as a laser 5, which irradiates electromagnetic radiation onto the flexible support 2, and also comprises a detection device 3 which detects a signal in the form of reflected electromagnetic radiation and generated upon deflection of the flexible support 2, and an evaluation device 4 which evaluates the detected signal and finally outputs at least one activity indicator indicative of the motion activity of the particles. In the depicted example, the motion detector corresponds to a motion detector as described in WO 2013 / 054311.

[0138] For illustrative purposes only, the following and referenced Figure 1 Preferred steps and a preferred order of these steps which allow the analysis of the motor activity of particles according to the method of the invention are summarized.

[0139] S0: The starting point of the analysis is at least one time-dependent signal indicating the deflection of the flexible support of the motion detector. Said signal is here called raw input signal RIS.

[0140] In order to eliminate any drift in the RIS, the method preferably subjects the RIS to drift elimination.

[0141] S1: From the signal, or from the RIS, but preferably after its drift has been eliminated, the method computes a time series of signal estimates. The computation of the time series of signal estimates can be done in the time domain or in the frequency domain.

[0142] In case a signal estimate in the time domain is to be calculated, the signal is split into time intervals, wherein the time dependency of the signal is analyzed in each time interval respectively to obtain a signal estimate, and wherein a time series of the signal estimates is analyzed to obtain signal parameters. For example, the signal estimate may be determined by using at least one of the following:

[0143] - estimators of moments of probability density functions;

[0144] - percentiles of probability density functions;

[0145] - Correlations, partial correlations, or results from Autoregressive Moving Average (ARMA) and Nonlinear Autoregressive Moving Average (NARMA) modeling.

[0146] The values ​​of signal estimates, e.g., a specific statistical algorithm or noise model (such as time-varying noise model parameters), preferably become samples of new time series. These new time series of signal estimates preferably consist of fewer points, reveal characteristic patterns, and can be used as input to determine or extract signal parameters, see further below.

[0147] In the case where the signal estimator in the frequency domain is to be calculated, the signal is preferably split into intervals of the estimated power spectrum. The estimation and windowing methods are preferably optimized for specific applications. After this, various noise models are preferably fitted to a given spectrum, and the signal estimator (here, the fitting parameters) is preferably used as a sample of a new time series. In addition to the signal estimator as a noise model parameter, the geometric properties of the estimated spectrum (e.g., extreme points, inflection points, areas under different parts of the curve, horizontal distances and vertical distances between them) can also be measured.

[0148] The result is a set of time series, each time series of the signal estimator being a specific fitted parameter or property. Furthermore, in this way the number of points can be significantly reduced, allowing the identification of patterns or shapes in the time series properties of a given metabolic activity by extracting the signal parameters, see further below.

[0149] S2: After the calculation of the time series of signal estimates, transform the time series of signal estimates in the time domain or in the frequency domain into a vector of signal parameters. The transformation is preferably done by one of the following three methods.

[0150] - The estimated time series is split into intervals. Statistical properties such as percentiles or distances are calculated for each interval. The statistical properties form a vector of basic signal parameters. After this, the vector of extended signal parameters can be calculated using ratios, differences and ratios of differences of the basic signal parameters. In this way, the signal parameters become more robust to environmental conditions.

[0151] - Fitting the estimated time series using a curve of a type and parameterization preferably optimized for a specific application. The type of curve can be exponential, polynomial, trigonometric and their linear and non-linear mixtures.

[0152] - A third method to extract signal parameters is a combination of the first two methods. To this end, it is preferred that the fitting is not performed on the entire signal but only within intervals, and preferably a mixture of using ratios and differences is performed.

[0153] The result of the extraction of the signal parameters is a large vector of real numbers (typically over 200). This vector is preferably shortened by selecting a subset of the signal parameters by performing a feature selection algorithm.

[0154] S3: Therefore, after the extraction of the signal parameters, the selection of the signal parameters is performed by a feature selection algorithm. In this way, the robustness and generalization level of the classification, or in a sense, the prediction algorithm, is improved. The feature selection algorithm for selecting the signal parameters is preferably optimized for a specific application and can be one of the following:

[0155] - univariate selection (such as F-test, mutual information, chi-square),

[0156] - Penalty-based methods (L1, L2, Elastic Net),

[0157] - Forward or backward selection of signal parameters based on properties of a specific model or based on cross-validation or bootstrapping.

[0158] The model Pareto optimality concept can be used to select the optimal signal parameter vector for multiple objective criteria in forward or backward selection. The number of signal parameters should be a minimum, while accuracy is a maximum. Usually, it is not possible to reduce the number of signal parameters used without reducing accuracy. Therefore, for example, according to these Pareto optimal algorithms, with the help of human experts, the final linking algorithm including the optimal input vector of signal parameters is preferably selected based on specific domain criteria. The additional criteria may be the performance of a specific test data set or additional requirements defined by experts.

[0159] S4: The input vector obtained after the selection of the signal parameters is fed to the linking algorithm. The linking algorithm uses the input vector of signal parameters to output an activity indicator of the particle. The activity indicator can be discrete, where the input vector of signal parameters extracted and selected in the previous step is used as input to the classification algorithm. That is, in this case, the linking algorithm corresponds to the classification algorithm. The classification algorithm can be one of the following:

[0160] - Logistic regression,

[0161] - Decision trees,

[0162] - Random Forest,

[0163] - Gradient Boosting Trees,

[0164] - Support Vector Machines,

[0165] -Multi-perceptron neural networks,

[0166] -Radial Basis Function (RBF) neural network.

[0167] The type of linking algorithm and the values ​​of its hyperparameters are preferably optimized for the specific application.

[0168] The activity indicator can also be a real number, wherein the input vector of signal parameters is also used as input for the regression algorithm. That is, in this case, the linking algorithm corresponds to the regression algorithm. The regression algorithm is preferably one of the following:

[0169] - linear regression,

[0170] - Decision Tree Regression,

[0171] - Random Forest Regression,

[0172] - Gradient Boosted Tree Regression

[0173] - kernel regression,

[0174] -Multi-perceptron neural networks,

[0175] -RBF neural network regression.

[0176] The type of linking algorithm and the values ​​of its hyperparameters are preferably optimized for the specific application.

[0177] More practically speaking, particles in contact with the flexible support cause random signals to be buried in a significant noise background in which useful information is encoded in both the time and frequency domains. Particles such as organisms evolve in time under different stimuli (e.g. nutrients, antibiotics, chemicals) that affect the properties of the random signal. For example, the method according to the invention can be used to extract signal properties by observing the temporal changes of signal properties that provide, for example, diagnostic information under appropriately optimized measurement conditions. Due to the biological variability between the tested organisms and between different conditions of measurement, the method proposed in this article has been proven to be a powerful tool that allows the separation of diagnostic information from measurement artifacts and the noise background. Comparing this example with Figure 1 In association with the analytical overview depicted in , the detected signal is transformed into a time series of signal estimates that encode, for example, the following: i) changes in the life state of the organism and ii) changes in the state of the measurement environment itself. The time series resemble different shapes and patterns that need to be quantified and are related to the phenomenon of interest. In order to detect a given phenomenon or to measure its intensity, these time series are then subjected to the extraction of signal parameters, where the signal parameters define, for example, a single experiment and are stable for a given application field. In order to optimize the set of signal parameters in order to improve the strength of the relationship between the signal parameters and the phenomenon of interest in the next stage of the analytical overview, a selection of the signal parameters is performed. Finally, an indicator indicative of the phenomenon of interest is output.

[0178] Figure 2 A signal detected with a motion detector, which has undergone drift elimination during evaluation with the method according to the invention, is depicted. In the depicted example, the drift is eliminated by fitting a linear curve (actually a first order polynomial) with 5 minute duration intervals.

[0179] Figure 3 Depicted is a signal detected with a motion detector, which undergoes fitting of a noise model when determining a signal estimate. The noise model here corresponds to a mixture of white noise and band noise and comprises four fitting parameters: amplitude of the band noise, frequency of the noisy poles, bandwidth and level of the white noise.

[0180] Figure 4The identification of a Pareto optimal feature selection algorithm that can be used in the method according to the invention is shown.The Pareto optimal feature selection algorithm is based on accuracy and the number of signal parameters.

[0181] Figure 5a and Figure 5b It is shown that the method according to the invention can be used for the differentiation of living and dead cells or, more generally, cells that show extreme differences in metabolic activity. This example can be considered as an introduction to the method and as a very basic and less complex case. In particular, in each case, the figures depict in the upper part a flexible support in the form of a microcantilever of a motion detector, comprising living bacteria ( Figure 5a ) and dead bacteria ( Figure 5b ).

[0182] In this experiment, the reference strain of the Gram-negative bacterium Escherichia coli ATCC-25922 was grown to the late logarithmic growth phase (OD 600 =1=10 9 Cells / mL) were added and centrifugation (2000 g, 3') was used to collect the reference strain ATCC-25922 of the Gram-negative bacterium Escherichia coli. The pellet was separated and divided into two groups for subsequent analysis. The first group of bacteria remained viable at room temperature, while the second group was exposed to 60°C for 20 minutes to produce bacteria without viability due to denaturation of enzyme complexes and other essential cell processes. As expected and confirmed by independent measurements, in this latter group, we were unable to detect metabolic activity using the redox-related fluorescent dye Resazurin, which changes its emission / excitation spectrum due to respiratory activity, a major indicator of the metabolic state of the cell.

[0183] Before the experiment begins, the microcantilever of the motion detector is functionalized with a linker required for achieving reliable and continuous sustainable cell attachment throughout the nanomotion recording. Attachment is performed as described in WO 2021 / 130339A1. In fact, in this experimental setup, positively charged poly-D-lysine (PDL) is used because it can promote the attachment of E. coli cells that exhibit a negatively charged lipopolysaccharide surface. In fact, see also WO 2021 / 130339A1, most cells exhibit a negative net charge on their surface, and if this is not the case, the functionalizing agent can be adjusted. Here, a solution of 0.1 mg / ml poly-D-lysine under ultrapure water is applied to the microcantilever for 20 minutes, and the microcantilever is rinsed and dried with ultrapure water again. Using a motion detector, the deflection of the functionalized bare microcantilever is recorded in 1 / 2 concentrated LB, and the deflection of the bare microcantilever is used as a blank for subsequent measurements of two groups of E. coli cells. Groups of live and heat-killed E. coli cells were then attached to the microcantilevers in parallel recordings. That is, two phases of signal recording were performed: a Blank phase, in which the deflection of an empty cantilever was measured, followed by a Bac phase, in which the deflection of the cantilever with attached bacteria was measured.

[0184] From the cantilever deflection and after drift removal using a linear detrending (first order polynomial) with 10 s long intervals, the time series of the signal estimate, in this case the variance (second order moment of the probability density function), was calculated and plotted for 20 min, first for the bare cantilever (Blank phase) and secondly for the cantilever with attached ATCC-25922 for each group of live and non-live cells (Bac phase). Figure 5a In FIG. 5 , the variance of those deflections of a microcantilever holding immobilized E. coli ATCC-25922 in 1 / 2 LB after incubation in LB at 37° C. is shown. As just mentioned, these are live bacteria. Figure 5b Shown is the variance of a microcantilever holding bacteria of the same strain after exposure to 60°C for 20 minutes, where the bacteria die.

[0185] Signal parameters were calculated from the variance, i.e., the signal estimate: the median of the variance for the 20 min recorded for the blank and Bac phases. In the blank phase, the mean μ was 1.6 × 10 -6 V 2 / V 2 After attachment of living cells, the variance increased by an order of magnitude to 1.5 × 10 -5 V 2 / V 2, while the median variance of heat-killed non-viable bacteria in the Bac stage remained at the level of the blank stage (μ 活 =1.5×10 -5 V 2 / V 2 With μ 热灭活 =2.5×10 -6 V 2 / V 2 , p=0.0013, t-test, μ of the variance within 20 min was calculated), which is consistent with other metabolic activity assessments such as those mentioned earlier using the fluorescent dye resazurin.

[0186] Figure 6a The median of the variance in the Bac phase of five experiments is depicted (bars represent median values, t-test p<0.005). Figure 6a , in this extreme case of viable and nonviable bacteria, the use of a single signal parameter was sufficient to distinguish between the two metabolically distinct groups.

[0187] Referring to the other figures, more complex scenarios with less extreme differences between groups (in terms of biological quality or phenotype) were analyzed by the method according to the invention.

[0188] to this end, Figures 7a to 12 The study of the reaction of cells to a changing environment which can be used for classification is shown. Here, the changing environment corresponds to a change from a favorable culture environment to a stressful environment. In particular, the figures show an antibiotic susceptibility test (AST) performed with the method according to the invention, in which susceptible and resistant E. coli strains were attached to the microcantilever of a motion detector and exposed to the antibiotic ceftriaxone.

[0189] In fact, for Figure 7a The analysis presented in Figure 7b The analyses presented in , used ceftriaxone-susceptible E. coli strain ATCC-25922 and ceftriaxone-resistant E. coli strain BAA-2452, respectively. Figure 7a and Figure 7b Plots are depicted for the following phase signal estimates, i.e., the variance of the cantilever deflection over time, here, for (i), bare functionalized cantilever (blank), (ii), cantilever with attached bacteria under culture conditions (Bac), and (iii), attached bacteria exposed to ceftriaxone (drug). In this example, 1 / 2 LB was used as culture medium in the blank phase and the Bac phase, and the 1 / 2 LB was supplemented with 32 μg / mL ceftriaxone in the drug phase.

[0190] Figure 7a and Figure 7bThe two patterns that can be identified in the data are characteristics of susceptible and resistant strains, but when the analysis is not done visually by the human eye but by a machine, the assessment is not always so clear and easy to distinguish.

[0191] For this reason, the signal parameter is calculated and the signal parameter is thresholded to determine whether a given strain is susceptible or resistant - that is, in the case where the value of the signal parameter is greater than the threshold, the strain is determined to be resistant, and in the case where the value is not greater than the threshold, the strain is determined to be susceptible. In order to determine whether cells with different responses to antibiotic stress can be classified, the signals of two other susceptible clinical isolates RN-26 and RN-49 from Swiss hospitals, and the signals of two other resistant clinical isolates B1 and B15 were detected with a motion detector. A total of 67 experiments with these strains exposed to ceftriaxone were recorded. The signal parameter is extracted from a single signal estimator, which gives one of the best separations between two different groups (such as, for example, the median of the first 30 minutes of the drug phase normalized to the median of the entire drug phase, or the median of the last 30 minutes of the drug phase normalized to the median of the entire drug phase).

[0192] Figure 8a Two signal parameters are depicted, namely the median from the first 30 minutes of the drug phase normalized to the median of the entire drug phase (lower figure) and the median from the last 30 minutes of the drug phase normalized to the median of the entire drug phase (upper figure). From these figures, it is apparent that the two signal parameters (SP) have different values ​​for susceptible and resistant strains. The two examples of signal parameters show significantly different populations from a large number of experiments conducted using the same conditions (t-test, p<0.00005) for many additional susceptible and resistant strains. However, in each case, the use of a single signal parameter is insufficient to distinguish a single record for a susceptible strain from a record for a resistant strain because there is overlap between the two populations. Therefore, a better way to separate the two populations is needed. When using a signal parameter from Figure 8a When two signal parameters are used to enhance the separation of susceptible and resistant populations, each of which includes different strains, a better way to separate the two populations is Figure 8b is shown in .

[0193] according to Figure 8b , it can be seen that the combination of signal parameters better separates two metabolic states, such as susceptible and resistant strains. That is, by using Figure 8aThe method is to enhance the separation of susceptible and resistant populations, each of which includes different strains, by using two signal parameters. In the depicted example, the two signal parameters are represented as a linear combination, wherein the separation line is neither vertical nor horizontal, but diagonal - the diagonal is mathematically represented as a linear combination of the two signal parameters. That is, the two signal parameters are weighted by two constants and added together to form a classification indicator that can be thresholded. The method can be generalized, and a large number of signal parameters can be extracted from a time series of signal estimates weighted by corresponding constants and thresholded to classify susceptible and resistant strains, for which the optimal number of signal parameters will achieve the highest accuracy. The optimal weights and thresholds can be estimated using a logistic regression algorithm. That is, here, the link algorithm is a logistic regression algorithm.

[0194] exist Fig. 9 In , it is shown that linear combinations of more than two signal parameters (in this case four signal parameters) can further improve the separation. This linear combination of the signal parameters becomes the classification indicator. That is, Fig. 9 How to add other signal parameters to improve classification accuracy (percentage of correctly classified experiments) is shown. The weighted sum of signal parameters is thresholded in the following manner, which makes negative values ​​indicate resistant strains, and positive values ​​indicate susceptible strains. In this group of 67 experiments, only one signal parameter (the ratio of the median of the variance of the last 30 minutes of the drug phase to the median of the variance of the entire drug phase) reaches 85.1% accuracy. The weighted sum of four signal parameters (the median of the variance of the last 30 minutes of the drug phase and the median of the variance of the entire drug phase, the median of the variance of the first 30 minutes of the drug phase and the median of the variance of the entire drug phase, the median of the variance from 60 minutes to 90 minutes of the drug phase and the median of the variance from 30 minutes to 60 minutes of the drug phase, the median of the variance from 60 minutes to 90 minutes of the drug phase and the median of the variance of the entire drug phase) improves the accuracy of separation of two categories to 89.6%. For this limited dataset, the weights and thresholds were optimized using a logistic regression algorithm to illustrate the potential of this solution.

[0195] However, in practical applications, the extraction and selection of signal parameters need to be carefully optimized. A large number of signal parameters lead to so-called algorithm overfitting. In this case, when the algorithm is used in a real situation, the high performance of the data set for fitting cannot be maintained. For this reason, a cross-validation process is applied, in which the data set is split into k subsets. The algorithm is fitted on k-1 subsets and verified on the last subset that is not used for fitting. The process is repeated k times each time, wherein the new subset is excluded. In this way, the average accuracy of the estimate reveals the problem of overfitting (the overfitting algorithm has a worse accuracy than the non-overfitting algorithm).

[0196] Cross validation helps to select only those signal parameters that process useful information. The selection starts with one signal parameter and successively increases the number of parameters one by one (forward selection) and after that starts removing signal parameters (backward selection) until only one signal parameter is left. Cross validation helps to evaluate how the addition or removal of signal parameters affects the performance of the indicator measured by a specified single or multiple objective criterion. The process allows to extract many signal parameters from many time series of signal estimators and select only the most important ones.

[0197] In the examples just presented, a fairly small data set was used. Figures 10 to 12 , the method of the present invention was applied to a much larger data set counting 1102 experiments with 84 clinical strains of E. coli isolated from patient specimens tested with ceftriaxone sodium. 561 of those experiments were performed on susceptible strains, and 541 of those experiments were performed on resistant strains.

[0198] In a first step, drift removal of the detected RIS is performed. For this purpose, the detected RIS is split into 10 s intervals. Thereafter, a linear fit of a first order polynomial is applied to the RIS of 10 s intervals to calculate the fitting error, which is the input for the calculation of the signal estimate, hereinafter in this article the variance (an estimate of the second order moment of the probability density function). In this way, a signal estimate varying with time is obtained, i.e. a variance signal s(t) with a sampling frequency of 0.1 Hz.

[0199] In the frequency domain, the 85-second intervals are used for power spectrum estimation and fitting by a mixture of a white noise model and a second-order band noise model using the common equation (1),

[0200]

[0201] Where f represents the frequency, A, f0 and Q are the parameters of the noise, and B is the parameter of the white noise (a constant). This produces four additional signals (A(t), f0(t), Q(t) and B(t)) with a sampling frequency of about 0.01 Hz (1 / 85). These (time series of statistical estimators) together with the variance signal s(t) are the inputs to determine the signal parameters.

[0202] In said determination of the signal parameters, the signal parameters are calculated. The signals s(t), A(t), f0(t), Q(t), B(t) are divided into intervals of 20 minutes in length for the Bac phase and 30 minutes in length for the drug phase, see Fig.10 , which depicts m2 to m7 and m9 to m 12 Intervals. For each interval, its median (50th percentile) and all possible ratios thereof are calculated. The definition of the median ratio is presented in equation (2),

[0203]

[0204] Among them, mi, m j Yes Fig.10 The median value between the i-th and j-th intervals defined in .

[0205] For the variance signal in the drug phase, linear and exponential fits were also applied, and the fitting parameters were the additional four signal parameters (slope a, intersection b, growth / decay rate T, and size c). The exact mathematical formula is defined in equation (3),

[0206] x(t)=at+b,x(t)=cexp(τt) (3)

[0207] The coefficients determining both the fit and the ratios C / T and b / a are in the set of extracted signal parameters. The linear and exponential fits are described in Fig.11 Shown in.

[0208] Hence, a set of 192 signal parameters have been determined or extracted and have gone through a signal parameter selection phase, ie have been fed to a feature selection algorithm.

[0209] exist Fig.12In the table depicted in, the performance of the feature selection algorithm based on Pareto optimality is presented. In addition to accuracy (percentage of correctly classified experiments), sensitivity (percentage of experiments with susceptible strains correctly classified) and specificity (percentage of experiments with resistant strains correctly classified) are also estimated. N represents the number of signal parameters used by the algorithm. In the example presented, the feature selection algorithm is a classification algorithm that estimates the susceptibility of cells to antibiotics by a 3-fold stratified cross-validation method repeated 300 times. Results were obtained for a data set that counted 1102 experiments with clinical strains of antibiotics-cefotaxime and Escherichia coli. 561 experiments were performed on susceptible strains, and 541 experiments were performed on resistant strains. Each column of the table corresponds to a different vector of signal parameters, and the different vectors of signal parameters become the input of the Pareto optimal classification algorithm with maximum accuracy and minimum number of signal parameters. The range of accuracy ranges from 84.8% for algorithms based on a single signal parameter to 86.7% for algorithms based on five signal parameters. The addition of subsequent signal parameters increases sensitivity from 84.8% to 86.7%.

[0210] Correlation with the kill rate of organisms

[0211] Reference Figures 13a to 13c , the method according to the present invention was shown to correlate with the kill rate of the organism.

[0212] That is to say, susceptible cells exposed to toxins or antibiotics show reduced viability that eventually leads to death. In the case of bacterial exposure to bactericidal antibiotics, the kill rate can be measured by determining the change in the number of colony forming units (CFU) at different time points. Thus, CFU is defined as bacteria (=unit), which forms visible colonies on a culture medium that allows growth through multiple replication cycles. Therefore, the number of these CFU is a measure of the concentration of live bacteria in the bacterial suspension of interest. Multiple samplings that vary over time evaluate the change in the number of CFUs of surviving bacteria, and are therefore used to calculate the kill rate. For more than a century, this method has been a very reliable standard technique, and for example, Robert Koch has applied this method.

[0213] However, one of the main disadvantages is its dependence on growth, and therefore the bacteria need time to form visible colonies for analysis. Fast growing bacteria such as Klebsiella pneumoniae form visible colonies within ten hours, however, slow growing pathogens such as Mycobacterium tuberculosis take several weeks to form visible colonies. In any case, this is a very reliable but slow method.

[0214] exist Figures 13a to 13cIn the present example depicted in , different susceptible strains of the bacterium Klebsiella pneumoniae exhibited different kill rates when exposed to the antibiotic ceftriaxone, depending on a range of causes and biological components. In our test set of six Klebsiella pneumoniae isolates, the kill rates ranged from 0.05 CFU*min-1 to 0.25 CFU*min-1. To determine the kill rate, the strains were cultured similarly to the strains described previously and exposed to the same concentrations of ceftriaxone. The population was sampled every 20 minutes and the number of viable bacteria was determined by the number of CFU after plating of colonies and overnight incubation of culture plates. At least three experiments were performed for each strain. As Figure 7a to Figure 7b As shown in , the same strains were used in parallel in the same order of recording phases for detecting the deflection of the flexible support with the motion detector. The same two signal parameters SP1: median VAR-DRUG 0-30min / median VAR-DRUG 0-120min and SP2: median VAR-DRUG 90-120min / median VAR-DRUG 0-120min were used to test the correlation between the detected signal and the killing rate, respectively. In doing so, at least three signal recordings with the motion detector in the case of a complete blank phase, a Bac phase and a drug phase were used to calculate the mean and SEM of the two signal parameters derived from the drug phase. A positive correlation with RSP1=0.81 and RSP2=0.5 was observed, which indicates a correlation between the deflection of the flexible support of the motion detector and the killing rate, see Fig.13a and Fig.13b .

[0215] However, and as follows Fig.13c , the correlation was significantly improved by increasing the number of signal parameters (in this case, four signal parameters: median VAR-DRUG 90-120min / median VAR-DRUG 0-120min, median VAR-DRUG 0-30min / median VAR-DRUG 0-120min, median VAR-DRUG 60-90min / median VAR-DRUG 30-60min, median VAR-DRUG60-90min / median VAR-DRUG 0-120min) to calculate the classification index. Killing rate (CFU*min -1) with the classification index from the signal increased to R = 0.98. Therefore, the linear combination of more than two signal parameters for the calculation of the classification index helps to more accurately describe the biological phenomenon (antibiotic susceptibility) that responds to the stress environment. Therefore, in summary, it can be said that the signal of the motion detector reflects the killing rate of the organism, and wherein the linear combination of the signal parameters improves the correlation with the killing rate determined with the standard microbiological method. The high correlation of the classification index means that the killing rate can be predicted and is an example of a linear regression algorithm.

[0216] Classification of metabolic activities

[0217] Reference Figures 14a to 16b , showing that the method according to the present invention can be used to classify metabolic activities.

[0218] exist Fig.14a and Fig.14b In the example depicted in , the time series of signal estimates, the resonant frequency, are analyzed and signal parameters are extracted similarly to the previous examples. Fig.14a Depicted is the estimated amount of signal for E. coli ATCC-25922 attached to a microcantilever when exposed to 1 / 2 LB and then to 1 / 8 LB, i.e., the E. coli cells experienced different nutrient availability and therefore exhibited different metabolic activities. Fig.14b In the control experiment depicted in , the same strain was exposed to 1 / 2LB instead of 1 / 2LB. Fig.14a and Fig.14b , it becomes apparent that the increase in resonant frequency is less pronounced for the change from 1 / 2 LB to 1 / 2 LB (the "unchanged" condition) than for the exchange of medium from 1 / 2 LB to 1 / 8 LB. Thus, in principle, signal parameters can be extracted and can be used to distinguish between metabolic states of the same organism.

[0219] Fig.15a and Fig.15b The extracted signal parameters determined from the resonant frequency domain are depicted. Fig.15a None of them can adequately separate or discriminate between the two populations to an acceptable degree (t-test, p>0.05). Fig.15b In the graph depicted in FIG. 1 , there is an overlap in the records for E. coli in 1 / 2 LB and E. coli in 1 / 8 LB.

[0220] As mentioned earlier, and as Fig.16aAs shown in , the line separation of the populations exposed to 1 / 2LB and 1 / 8LB can be represented by a linear combination of these two signal parameters. In fact, using a linear combination of more than two signal parameters (three signal parameters in this case) can further improve the separation. This linear combination of signal parameters becomes the classification indicator. Fig.16a , the accuracy of separating the two groups increased from 70.6% using only one signal parameter to 88.2% using three signal parameters. The weights and thresholds were estimated using a logistic regression algorithm, that is, here, the link algorithm was a logistic regression algorithm.

[0221] To structure the description of signal parameters, Polish notation is used as well as some formal structure of signal parameter names.

[0222] Signal parameter names have the following structure

[0223] TIME_SERIES_NAME_PHASE_NAME_START_TIME-END_TIME_STATISTICS_USED

[0224] (eg, f0_Bac_80-90_p50 - represents the f0 signal estimate for which the 50th percentile was calculated for the bac stage in the time interval from 80 min to 90 min.) The statistics used were percentile, mean, standard deviation (std) and slope.

[0225] Polish notation is a way of encoding arithmetic expressions without using parentheses, for example,

[0226] / +a ba b is equivalent to (a+b) / (ab). Polish notation is used to automatically encode signal parameters that are ratios and differences of other signal parameters.

[0227] The feature selection algorithm selects the following noise signal parameters as Pareto optimal based on accuracy and number of signal parameters:

[0228]

[0229] Fig.16b Independent experiments involved measuring ATP concentration using a chemiluminescent method. Figures 15a to 16a Similar to the measurements presented, the energy status of cells can be assessed by ATP concentration in different media (1 / 2LB and 1 / 8LB) with different nutrient availability and by using a luciferase-activated protein kinase (Bacteriochrome Glo TM ) was assessed by chemiluminescent enzyme assays producing ATP-dependent enzyme activity.

[0230] As follows Fig.16b , a metabolic marker of the metabolic activity of the cell also leads to two partially overlapping populations, where each symbol in the figure represents a measurement value. Therefore, similar to the case of using one signal parameter to analyze the signal of the flexible support in the motion detector, using only one metabolic marker is not sufficient to separate two metabolic states of the same type of cells in a narrowly related environment (in this case 1 / 2LB and 1 / 8LB). Therefore, in this case, the use of a combination of signal parameters, i.e., a classification index, can also be understood as capturing several aspects of the metabolic state of the cell. Increasing the number of signal parameters can be understood similarly to increasing the number of metabolic markers.

[0231] The following embodiments explore three aspects of the time dependence of the signal evaluated by the evaluation device: complex aspects of the power spectral density not covered by theoretical noise models, rare events in the signal related to particle vibrations that may disappear during the power spectral density estimation, and the self-affine nature of cell vibrations.

[0232] In particular, Fig.26 A typical spectrum and signal estimates that can be derived from a signal detected with a motion detector and analyzed with the method according to the invention are presented in Figure 1. For such a spectrum, a list of 60 time series of the following signal estimates is defined. The time series are estimated every 2 minutes.

[0233] The local minimum point of the spectrum (f min and p min )

[0234] The local maximum point of the spectrum (f max , p max )

[0235] The point p that is halfway between the local minimum and maximum vertical distances left =p right =(p max +p min ) / 2 and the corresponding frequency (f left , f right )

[0236] The parameters of the flicker noise frequency range from 20Hz to 200Hz (N0, α)

[0237]

[0238] The area under the curves for the five frequencies range from N1 to N5 (20 Hz, f min )、(f min , f left )、(f left , f max )、(fmax , f right )、(f right , 20kHz).

[0239] The areas under the curves for 20 frequencies range equally spaced on a logarithmic scale from NE1 to NE20 (20 Hz to 28 Hz, 28 Hz to 39 Hz, ..., 224 Hz to 316 Hz, 316 Hz to 447 Hz, ..., 10023 Hz to 14158 Hz, 14158 Hz to 20000 Hz).

[0240] The normalized power normN for each i i and normNE i N i / (N1+N2+N3+N4+N5) and NE i / (N1+N2+N3+N4+N5).

[0241] These time series of signal estimates can be used to calculate signal parameters and / or ratios and differences thereof.

[0242] According to the following Fig. 27 and Fig.28 , quantiles, especially percentiles, of the periodogram (see Fig. 27 ) can be used to consider the shape of the probability distribution of the periodogram (see Fig.28 ). In the described solution, 9 percentiles from the 10th, 20th, ..., 90th are used, and 9 percentiles from the 10th, 20th, ..., 90th are calculated for a predefined frequency range. In the presented embodiment, the following frequency ranges are used: 0 Hz to 10 Hz, 10 Hz to 100 Hz, 100 Hz to 200 Hz, 200 Hz to 400 Hz, 400 Hz to 1000 Hz, 1000 Hz to 2000 Hz, 2000 Hz to 4000 Hz, 4000 Hz to 5000 Hz, 5000 Hz to 6000 Hz, 6000 Hz to 7000 Hz, 7000 Hz to 8000 Hz, 8000 Hz to 10000 Hz (p10_1000-2000 represents the 10th percentile of the frequency range 1000 Hz to 2000 Hz). In addition, the following extended time series is created.

[0243] The KLPQ-Kullback-Leibler divergence between the empirical probability density function and the chi-squared probability density function based on percentile calculations,

[0244] KLQP-Kullback-Leibler divergence between the chi-squared probability density function and the empirical probability density function calculated based on percentiles,

[0245] Kurtosis - is the kurtosis (moment) of the empirical probability density function,

[0246] JS——is the Jensen-Shannon distance between the chi-square probability density function and the empirical probability density function,

[0247] MO——(p 60 -p 40 ) / (p 90 -p 10 )(ratios and differences),

[0248] EEPDF - is the mean (moment) of the empirical probability density function.

[0249] They provide samples of 112 time series of signal estimates every 5 minutes, from which signal parameters and their ratios and differences are calculated for different percentiles, frequency ranges and time intervals. The signal parameters and their ratios and differences are the object of the feature selection algorithm.

[0250] Current findings on cell vibrations reveal that the power spectral density function of cell vibrations has a strong 1 / f characteristic. This spectrum is generally related to signal self-affine. Self-affine can be analyzed by detrended fluctuation analysis (DFA) and its generalized multifractal detrended fluctuation analysis (MF-DFA). In these methods, the dependence of signal statistical properties on the time scale is studied. In the presented embodiment, signal parameters based on MF-DFA are used. In particular, the following 21 generalized mean exponents / powers are used: -10.0, -8.0, -6.0, -4.0, -2.0, -1.0, -0.8, -0.6, -0.4, -0.2, 0.0, 0.2, 0.4, 0.6, 0.8, 1.0, 2.0, 4.0, 6.0, 8.0, 10.0. Together with 9 numbers of sub-intervals (4, 8, 16, 32, 64, 128, 256, 512, 1024) this results in 189 signal parameters from which ratios and or differences are further calculated.

[0251] exist Fig.29 , a table of signal parameters used in the presented embodiment is shown. The signal parameters are calculated as F for the last 20 minutes of the drug phase and the first 20 minutes of the drug phase. q Thus, the signal parameter "F0" is the F value for the last 20 minutes of the drug phase. -10 (1024) divided by the F of the first 20 minutes of the drug phase -10 (1024) These signal parameters are mixed again by calculating their ratio (eg, F146_F129 means F146 / F129).

[0252] and Figures 7a to 12 As well as Figures 14 to Fig.16a Similar to the example consisting of a relatively small data set depicted in , the combination of signal parameters (SP) significantly increases the performance of the linkage algorithm, which is a classification algorithm on a large data set consisting of extremely diverse samples. Again, we used the response of cells to a change from a favorable culture environment to a stressful environment, this time with the antibiotic ciprofloxacin (CIP). We used 83 different clinical isolates of K. pneumoniae strains whose minimum inhibitory concentrations (MICs) for ciprofloxacin varied and ranged over four orders of magnitude and were considered clinically susceptible or resistant, as shown in Figure 2. Fig.18a MICs were determined by standard diagnostic reference methods (E-test or culture microdilution). The level of resistance to ciprofloxacin is due to different resistance mechanisms and other genetic determinants in different strains. Although strain diversity is mentioned, in the Figure 7a and Figure 7b In the previously described experimental setup for nanomotility recordings, only one concentration, 8 μg / ml CIP ( Fig.18a Nanomotility of 83 strains was measured under "Conc." in Figure 1. Briefly, in each experiment, bacterial samples attached to micromechanical sensors were exposed to culture medium (1 / 2 concentrated LB) for 120 min and then exposed to CIP for 120 min.

[0253] exist Fig.18b In the example, the performance improvement as the number of SPs increases is visualized, and Fig.19 All metrics evaluated for the linkage algorithm as a classification algorithm for this dataset are listed. Accuracy, sensitivity, and specificity were calculated as follows:

[0254]

[0255] In this regard, true positives are considered to be experiments with correctly classified susceptible strains, true negatives are experiments with correctly classified resistant strains, false positives are experiments with incorrectly classified resistant strains, and false negatives are experiments with incorrectly classified susceptible strains.

[0256] Classification model No. 1 (a classification model is an example of a chaining algorithm as a classification algorithm) is based on a single SP (F128) that may have the greatest impact, resulting in a classification accuracy of 85.8%. Model No. 2 with two SPs (F128, F61) reaches 89.7% accuracy, model No. 3 with three SPs (F128, F61, F146_F129) reaches 91.4% accuracy, and finally, model No. 4 with a combination of four SPs (F128, F61, F146_F129, F7_F0) reaches 93.1% accuracy - in Fig.19 In summary. Fig.18b Increases in sensitivity and specificity were also shown.

[0257] The score, i.e., the classification metric for each of the 233 experiments, is Fig.20a For Model 1 based on one SP and Fig.20b The reference MIC that is considered susceptible (S reference) or resistant (R reference) for No. 4 models based on four SPs is used as the benchmark. The score assumes that the predicted susceptibility is a positive value (S invention), and the predicted resistance is a negative value (R invention). The experiments correctly classified for susceptible strains are described in light gray (true positives), and the experiments correctly classified for resistant strains are described in dark gray (true negatives). Misclassified experiments, false positives and false negatives are shown by white circles. The number of misclassified experiments drops from the initial 33 of No. 1 model to 16 of No. 4 models. All models are chained algorithms as classification algorithms, which are logistic regression algorithms using MF-DFA signal parameters.

[0258] In conclusion, one SP, which can be considered as a separate aspect of the nanomotility signature, is less suitable than the 4 SPs to describe the diverse responses to drugs found in such a diverse dataset of 83 different strains. The addition of SPs and their combination significantly improved the performance of the classification algorithm.

[0259] In another example, an even larger and more diverse dataset of 487 samples—including 160 clinical isolates of two different bacterial species (Escherichia coli and Klebsiella pneumoniae)—was exposed to one concentration of ceftriaxone (indicated as “Conc.”, Fig.12 a). In response to a 120-min medium phase (1 / 2 concentrated LB) followed by a 120-min drug exposure phase (cefotaxime, CRO) Figure 7a and Figure 7b Nanomotility of bacterial cells was measured in the same experimental setup described, and 160 strains again spanned a wide range of MICs over several orders of magnitude ( Fig.21a, using the standard reference method of MIC). Signal parameters derived from signal estimators of geometric properties as a function of power spectral density are used. The feature selection algorithm selects the following signal parameters based on accuracy, number of signal parameters and Pareto optimality concepts:

[0260] / NE0_Drug_90-120_p50NE0_Drug_0-30_p50

[0261] / normNE19_Drug_90-120_p50normNE19_Drug_60-90_p50

[0262] / normNE13_Bac_90-120_p50normNE13_Bac_90-120_p50

[0263] / normNE1_Drug_90-1200_p50 normNE1_Drug_0-1200_p50

[0264] The linking algorithm, which is a classification algorithm with high performance for separating experiments with resistant strains from experiments with susceptible strains, is based on a combination of the above four signal parameters, thereby generating a score, i.e., a classification index, which assumes that experiments with predicted susceptibility are positive (S invention) and experiments with predicted resistance are negative (R invention). The accuracy reached 91.6%, the sensitivity reached 89.0%, and the specificity reached 94.6% ( Fig.24 ).according to Figure 21b The score of each sample is plotted against the reference MIC using the method shown in . Experiments correctly classified for susceptible strains are depicted in light grey (true positives), and experiments correctly classified for resistant strains are depicted in dark grey (true negatives). Misclassified experiments (false positives and false negatives) are shown by white circles.

[0265] In targeting Figure 7a and Figure 7b The nanomotor device described herein was used to measure the effects of exposure to 8 μg / ml ciprofloxacin (CIP, Fig.22a and Figure 22b A similar analysis was performed on 210 samples of 127 E. coli strains (indicated as "Con.c" in Figure 1). After processing of the nanomotion signals, in this embodiment, the quantile and MF-DFA signal parameters were used with a linkage algorithm as a classification algorithm (the classification algorithm was used as a logistic regression). The signal parameters selected by the feature selection algorithm are as follows:

[0266] / / / -p90_0-10_Drug_90-120_mean p50_0-10_Drug_90-120_mean-p50__0-10_Drug_90-120_mean p10_0-10_Drug_90-120_mean / -p90_5000-6000_Drug_90-120_meanp50_5000-6000_Drug_90-120_mean-p50_5000-6000_Drug_90-120_mean p10_5000-6000_Drug_90-120_mean / / -p90_0-10_Drug__0-30_mean p50_0-10_Drug__0-30_mean-p50_0-10_Drug__0-30_mean p10_0-10_Drug_0-30_mean / -p90_5000-6000_Drug_0-30_mean p50_5000-6000_Drug_0-3_mean-p50_5000-6000_Drug_0-30_mean p10_5000-6000_Drug_0-30_mean

[0267] / / p20_200-400_Drug_90-120_mean p20_10-100_Drug_90-120_mean / p20_200-400_Drug_30-60_meanp20_10-100_Drug_30-60_mean

[0268] / p30_10-100_Bac_60-90_meanp30_0-10_Bac_60-90_mean

[0269] F75

[0270] The final linkage algorithm as a classification algorithm was based on four SPs, and the final linkage algorithm correctly classified 190 out of 210 samples. Therefore, the final linkage algorithm achieved 90.5% accuracy, 89.8% sensitivity, and 91.2% specificity ( Fig.24 ). Twenty misclassified samples (such as Figure 22b The majority of these isolates (shown by white circles in Figure 3) were found around the clinical breakpoint that indicated the boundary between susceptible (≤S) and resistant (R>) strains defined by the reference AST method.

[0271] However, in another analysis, 155 samples of 125 diverse E. coli strains from a different strain collection were exposed to a third antibiotic, cefotaxime (CTX). In the same nanomotility measurement setup as described in 7a and 7b, we used 32 μg / ml cefotaxime (in Fig.23a and Figure 23b In the present embodiment, quantile and MF-DFA signal parameters are used together with a link algorithm as a classification algorithm (which is a logistic regression). The signal parameters selected by the feature selection algorithm based on accuracy and the number of signal parameters and based on the concept of Pareto optimality are as follows:

[0272] p50_400-1000_Drug_90-120_meanp50_100-200_Drug_90-120_mean /

[0273] p50_400-1000_Drug_0-30_meanp50_100-200_Drug_0-30_mean

[0274] / / / -p50_200-400_Drug_30-6_mean p10_200-400_Drug_30-60_mean-p90_200-400_Drug_30-60_mean p50_200-400_Drug_30-6_mean / -p50_100-200_Drug_30-60_meanp10_100-200_Drug_30-60_mean-p90_100-200_Drug_30-60_mean p50_100-200_Drug_30-60_mean / / -p50_200-400_Drug_0-30_mean p10_200-400_Drug_0-30_mean-p90_200-400_Drug_0-30_mean p50_200-400_Drug_0-30_mean / -p50_100-200_Drug_0-30_meanp10_100-200_Drug_0-30_mean-p90_100-200_Drug_0-30_mean p50_100-200_Drug_0-30_mean

[0275] F186_F64

[0276] F45_F38

[0277] / p70_1000-2000_Drug_90-120_mean p70_100-200_Drug_90-120_mean / p70_1000-2000_Drug_00-30_mean p70_100-200_Drug_0-30_mean

[0278] The linkage algorithm was a classification algorithm based on the five SPs and resulted in only 11 misclassifications, yielding an accuracy of 92.9%, a sensitivity of 91.7%, and a specificity of 94% ( Fig.24 ). The score of each sample, i.e. the classification index, is again calculated as Figure 22b The reference AST method in is used as a benchmark.

[0279] In short, for Figures 18a to 24 For each of the four bacterial-drug combinations in the study, different high-performance linking algorithms were developed as classification algorithms based on a small number of SPs calculated from the time series of the signal estimator. In each case, the group of resistant cells could be almost perfectly separated from the susceptible cells based on the different cellular and metabolic responses to the three different drugs, ceftriaxone, ciprofloxacin, and cefotaxime, within 120 minutes of exposure.

[0280] The two aforementioned drugs, cefotaxime and ciprofloxacin, hinder different cellular processes of bacterial cells. Cefotaxime belongs to the family of beta-lactam antibiotics that interfere with cell wall metabolism, while ciprofloxacin, as a quinolone, binds to topoisomerases involved in DNA folding, a process that affects replication and, effectively, all processes involving DNA. The proper functioning of both drug targets is essential. While for both drugs their obstruction is detrimental to the cell in the long run, the cellular stress responses triggered by the two drugs are different.

[0281] On a total dataset size of 404 experiments, 202 experiments out of the 404 were performed with a susceptible E. coli strain and 32 μg / ml cefotaxime, and the same number of experiments were performed with a susceptible E. coli strain and 8 μg / ml ciprofloxacin, a combination of nanomotility recordings and machine learning was applied to develop a linking algorithm as a classification algorithm to separate the desired information in the nanomotility response to cefotaxime from the desired information in the nanomotility response to ciprofloxacin. All 404 experiments were used simultaneously to develop the classification model. In the case where an experiment was performed with cefotaxime and then therefore correctly predicted by the method, the experiment was considered correctly classified. The same was true for ciprofloxacin. The experimental setup was again the same as for cefotaxime. Figure 7a and Figure 7bThe experimental setup described was identical with a 120 min culture phase followed by a 120 min exposure to either drug. For the benchmark test, the MIC was determined by the reference method.

[0282] Quantile signal parameters were used and the linkage algorithm as the classification algorithm was the support vector machine algorithm with radial basis function. The feature selection algorithm selected the following Pareto optimal signal parameters based on accuracy and number of signal parameters:

[0283] -KLQP_200-400_Drug_90-120_p95 KLQP_200-400_Drug_0-30_p95

[0284] / EEPDF_0-10_Drug_090-120_p95EEPDF_0-10_Drug_0-30_p95

[0285] / Kurtosis_0-10_Drug_90-120_p10 JS_0-10_Drug_90-120_p10

[0286] -EEPDF_Drug_90-120_mean JS_2000-4000_Drug_0-30_mean

[0287] MO_0-10_Drug_90-120_p75

[0288] MO_100-200_Drug_30-60_std

[0289] KLQP_0-10_Drug_0-30_std

[0290] In the chaining algorithm as a classification algorithm, predicted susceptible responses to ciprofloxacin were assigned negative score values, while predicted responses to cefotaxime were assigned positive score values ​​(similar to the classification algorithm for R phenotype and S phenotype). Accordingly, the misclassified experimental hypotheses were positive for ciprofloxacin and negative for cefotaxime. On the high-performance classification algorithm based on seven SPs, the accuracy for both drugs reached 85.4%. 87.0% of the ciprofloxacin experiments were correctly classified, and 83.7% of the cefotaxime experiments were correctly classified. The impact of each additional signal parameter was Fig.30 Therefore, the required information in the nanomotor signals caused by different stressors can be extracted and described by the classification algorithm.

[0291] In addition to chemical stressors, such as different drugs ( Figures 7a to 12 as well as Figures 18a to 24 ) and different levels of nutrient supply (Figures 14 to Fig.16a), physical stressors may also affect cellular and metabolic activity, and therefore the nano-movement of cells. To this end, the nano-movement of several different E. coli strains with different genetic composition was recorded in culture medium (1 / 2 concentrated LB) for 30 minutes. Half of these experiments were performed at room temperature, while the other half were exposed to 37°C (76 experiments for 37°C and 72 experiments for RT). Since the bacterial E. coli has adapted to the lifestyle within the human intestinal microbiome, the metabolic rate at 37°C is expected to be higher. Spectral signal parameters and logistic regression classification algorithms are used. The feature selection algorithm selects the following Pareto optimal signal parameters based on accuracy and the number of signal parameters:

[0292] / NE17_Bac_0-15_p25 NE12_Bac_0-15_p25

[0293] / NE17_Bac_15-30_p10NE13_Bac_15-30_p10

[0294] / NE16_Bac_15-30_p75 NE15_Bac_15-30_p75

[0295] -NE19_Bac_15-30_p10 NE19_Bac_0-15_p10

[0296] N1_Bac_0-15_std

[0297] Comparing nanomotion at two different temperatures, and thus anticipating rather global changes in cellular metabolic activity, the accuracy of the classification algorithm in separating the two conditions ranged from 97% for a single SP up to 99% for five SPs. Fig.31 In , the dependence of the algorithm performance on the number of signal parameters chosen based on accuracy and the number of signal parameters and the concept of Pareto optimality is presented.

[0298] Drug sensitivity testing of cancer cells

[0299] Fig.25a and Fig.25b It is shown that the method according to the invention can be used to perform a drug sensitivity test (DST) on cancer cells attached to a microcantilever. In particular, these figures show the variance ( ) of the deflection of a microcantilever with attached human colon cancer cell line SW480 and subsequently exposed to an inhibitory concentration of the drug doxorubicin measured in cell culture medium. Fig.25a ) and the variance of cells maintained in cell culture medium ( Fig.25b ).

[0300] Doxorubicin-susceptible SW480 were cultured in cell culture flasks under standard laboratory conditions (i.e., cell culture medium (Dulbecco's Modified Eagle Medium (DMEM) containing 10% heat-inactivated fetal calf serum (FCS) at 37°C and 5% CO2). In preparation for nanomotility experiments, cells were detached, collected in cell culture medium, and washed in DMEM containing 10% FCS. The cell suspension was used for attachment to the cantilever, and all nanomotility measurements performed with the device were performed in DMEM containing 10% FCS. The cantilever was attached to the colon cancer cells. Three phases of signal recording were performed: (i) a blank phase, in which the deflection of the bare cantilever was measured; (ii) a culture phase, in which the deflection of the cantilever with attached SW480 was measured, followed by (iii) a drug phase or a second culture phase without doxorubicin, in which the deflection of the cantilever with attached SW480 was recorded after exposure to the drug. The recordings presented below were performed in a motion detector installed in a CO2-supplied incubator to allow optimal culture conditions for cancer cells.

[0301] Fig.25a Depicted is the variance of the nanomotor signal over time with attached SW480 colon cancer cells in culture conditions followed by medium supplemented with 32 μM doxorubicin during the drug phase. Fig.25b In the experiment depicted in Figure 2 without the addition of doxorubicin, cells were maintained in DMEM and the second culture phase was recorded. Fig.25a and Fig.25b The reduction in variance after exposure to the drug was significant compared to the unmodified condition. Based on 61 experiments with SW480 (of which 32 experiments were exposed to doxorubicin and 29 experiments were without doxorubicin), spectral signal parameters were calculated and a link algorithm as a logistic regression classification algorithm was used to distinguish between the two conditions.

[0302] The signal parameters selected by the feature selection algorithm based on Pareto optimality, accuracy and the number of signal parameters are as follows:

[0303] / NE8_Drug_0-300_std NE2_Drug_0-30_std

[0304] -NE0_Drug_90-120_p75 NE0_Drug_0-30_p75

[0305] / NE10_Drug_90-120_slope NE3_Drug_90-120_slope

[0306] -NE19_Drug_90-120_std NE13_Drug_30-60_std

[0307] -NE9_Drug_60-90_std NE5_Drug_0-30_std

[0308] -NE7_Drug_0-30_std NE1_Drug_0-30_std

[0309] NE4_Drug_0-30_p95

[0310] In fact, the accuracy of the Pareto-optimal linking algorithm ranges from 83.2% for one SP to 91.6% for seven SPs. Fig.32 In , the dependence of the algorithm performance on the number of signal parameters selected based on accuracy, number of signal parameters and the concept of Pareto optimality is presented. The method is a forward selection method based on 300 repeated 3-fold cross validation.

[0311] Quantification of metabolic activity

[0312] In addition to the qualitative assessment of the cellular metabolic state by using classification models, the impact of drug interference on cellular metabolic activity can also be quantified using regression models. We used a clinical Escherichia coli isolate IMHA-2155385 susceptible to the drug combination ceftazidime-avibactam. This cephalosporin / β-lactamase inhibitor combination attacks bacterial cell wall synthesis and simultaneously blocks the resistance mechanism of β-lactamase media. Fig.33 In the present paper, the results of metabolic state regression induced by different inhibitory concentrations of antibiotics (32 μg / ml to 512 μg / ml) are presented. The figure shows the relationship between the predicted results of drug concentration by the model based on five signal parameters and the actual drug concentration. The determined coefficients of the presented model range from 0.86 for a single signal parameter to 0.993 for five signal parameters. Fig.34 In the present paper, the dependence of the algorithm performance on the number of signal parameters selected based on the mean square error and the number of signal parameters and the concept of Pareto optimality is presented. The method is a forward selection method based on 10-fold cross validation. The link algorithm as a regression algorithm (which is a linear regression algorithm) is used with the following selected signal parameters:

[0313] / / -p50_8000-10000_Drug_90-120_mean p10_8000-10000_Drug_90-120_mean-p50_1000-2000_Drug_90-120_mean p10_1000-2000_Drug_90-120_mean / -p50_8000-10000_Drug_0-30_mean p10_8000-10000_Drug_0-30mean-p50_1000-2000_Drug_0-30_mean p10_1000-2000_Drug_0-30_mean

[0314] / / p10_6000-7000_Bac_90-120_mean p10_0-10_Bac_90-120_mean / _p10_6000-7000_Bac_60-90_mean p10_0-10_Bac_60-90_mean

[0315] / / p20_6000-7000_Drug_90-120_mean p20_5000-6000_Drug_90-120_mean / p20_6000-7000_Drug_60-90_meanp20_5000-6000_Drug_60-90mean

[0316] / / / -p90_7000-8000_Drug_90-120_mean p10_7000-8000_Drug_90-120_mean-p60_7000-8000_Drug_90-120_mean p40_7000-8000_Drug_90-120_mean / -p90_2000-4000_Drug_90-120_mean p10_2000-4000_Drug_90-120_mean-p60_2000-4000_Drug_90-120_mean p40_2000-4000_Drug_90-120_mean / / -p90_7000-8000_Drug_0-30_mean p10_7000-8000_Drug_0-30_mean-p60_7000-8000_Drug_0-30_mean p40_7000-8000_Drug_0-30mean / -p90_2000-4000_Drug_0-30_mean p10_2000-4000_Drug_0-30_mean-p60_2000-4000_Drug_0-30_mean p40_2000-4000_Drug_00-30_mean

[0317] / / p30_6000-7000_Drug_60-90_mean p30_5000-6000_Drug_60-90mean / p30_6000-7000_Bac_60-90_mean p30_5000-6000_Bac_60-90_mean

[0318] The results presented demonstrate that the present invention allows for the measurement of the effects of increased metabolic activity associated with increasing drug concentrations.

[0319] Reference numerals list

[0320] 1 Motion Detector

[0321] 2 Flexible support

[0322] 3. Detection device

[0323] 4 Evaluation device

[0324] 5 Radiation Sources

Claims

1. A method for analyzing the motion activity of particles using a motion detector (1), wherein: The motion detector (1) comprises: - a flexible support (2) configured to deflect, - a detection device (3) for detecting a signal generated when the flexible support (2) is deflected, and - an evaluation device (4) for evaluating the detected signal, Wherein, the method comprises the following steps: - bringing at least one particle into contact with said flexible support (2), - detecting, using said detection means (3), at least one time-dependent signal indicative of a deflection of said flexible support (2) due to the movement of said at least one particle, and - evaluating the detected time-dependent signal with the evaluation device (4), Characterized in that the evaluation of the detected time-dependent signal comprises: - analyzing the time dependency of the signal in order to derive from the signal a plurality of signal parameters characterizing the variation of the signal over time, and - executing a linking algorithm having as input variable an input vector comprising at least a selection of said signal parameters and having as output variable at least one activity indicator indicative of a movement activity of said particles.

2. The method according to claim 1, wherein: In the step of analyzing the time dependency of the signal, a plurality of time intervals are defined, wherein the time dependency of the signal is analyzed in each time interval respectively to obtain the value of each signal estimate in a plurality of signal estimates for each time interval, and wherein the time series of each signal estimate is analyzed to obtain each of the signal parameters.

3. The method according to claim 2, wherein: each signal estimate value is determined by fitting a noise model to the signal and / or by applying at least one statistical algorithm to the signal, and / or Wherein each signal estimator is selected from the group consisting of: estimators of moments of a probability density function, geometric properties of a power spectral density function, percentiles of a probability density function, correlation parameters, partial correlation parameters, parameters of an autoregressive moving average model, and parameters of a nonlinear autoregressive moving average model.

4. The method according to claim 2 or 3, wherein: Each signal estimator is a geometric property of the power spectral density function, and The geometric attribute is preferably an extreme point of the power spectral density function, an inflection point of the power spectral density function, an area under a curve described by the power spectral density function, or a horizontal and / or vertical distance therebetween.

5. The method according to claim 2 or 3, - determining a plurality of periodograms for each time interval, and - for all periodograms, determining for at least one frequency range at least one quantile, preferably at least one percentile, thereby obtaining said signal estimate.

6. The method according to claim 5, wherein: The time series of at least one of the signal estimates is expanded to form an expanded time series by at least one of the following operations: - combining at least two signal estimates as their difference and / or ratio; - transforming the plurality of quantiles into values ​​of an empirical probability density function, and preferably also: o determining at least one distance between the value of the empirical probability density function and the value of the theoretical probability density function; and / or o determining at least one geometric property and / or various moments of the empirical probability density function, and Therein, the at least one signal parameter is obtained from the extended time series.

7. The method according to any one of claims 2 to 6, wherein: At least one signal parameter is derived from the time series of at least one of the signal estimates and / or from an extended time series of at least one of the signal estimates by at least one of the following operations: - defining a plurality of subsequences for the time series and / or for the extended time series and determining for each subsequence a statistical property and preferably also determining at least one of a ratio, a difference and a ratio of differences of said statistical property, - fitting one or more parameterized curves comprising one or more parameters to the time series and / or the extended time series, and - defining a plurality of subsequences for the time series and / or for the extended time series, and fitting one or more parameterized curves comprising one or more parameters for each subsequence to obtain fitting parameters, and preferably also determining ratios, differences and ratios of differences of the fitting parameters.

8. The method according to claim 1, wherein: Evaluating the detected time-dependent signal with the evaluation device (4) comprises detecting the time-dependent signal during a time t, and wherein analyzing the time dependency of the signal comprises the following steps: - dividing said signal at time t into time intervals of length T, - For each time interval j=1, ...t / T, perform the following steps: ○ Calculate the digital integral of the signal in each time interval; ○ dividing the time interval into at least a number of sub-intervals; o for each subinterval, detrending the digital integral of the signal by applying a detrending algorithm to the digital integral of the signal; o for the at least one number of subintervals, combining the subintervals into detrended intervals; and o Determining at least one generalized mean of the detrended signal of the detrended interval for the at least one number of subintervals, thereby obtaining at least one signal parameter.

9. The method according to claim 8, wherein: At least one signal parameter is derived from a ratio and / or a difference between at least two signal parameters.

10. The method according to any one of the preceding claims, further comprising executing a feature selection algorithm to automatically select a subset of signal parameters to obtain the input vector fed to the linking algorithm, in, The feature selection algorithm is preferably configured to select a subset of the signal parameters in such a way that the subset is optimal with respect to multiple objective criteria.

11. A method according to any one of the preceding claims, wherein: applying a drift cancellation algorithm to said signal before deriving said signal parameter from said signal, and Wherein, the drift elimination algorithm preferably includes a curve fitting process or high pass filtering to model the drift component of the signal, and / or Wherein, the drift elimination algorithm is preferably a detrending algorithm.

12. A method according to any one of the preceding claims, wherein: The linking algorithm is a regression algorithm or a classification algorithm.

13. A method according to any one of the preceding claims, wherein: The linking algorithm is a machine learning algorithm.

14. A method according to any one of the preceding claims, wherein: The particle is at least one of the following: a cell such as a prokaryotic cell or a eukaryotic cell, a spore, a virus, a bacteriophage, a vesicle, a peptide, a protein, a polysaccharide, a lipid, a glycoside, a nucleic acid or a copolymer of nucleic acid, a protein-RNA copolymer, an RNA-DNA copolymer, a protein-DNA copolymer, an RNA / DNA-protein copolymer, a protein-protein copolymer, a capsule, an organelle, or a nanodevice.

15. A method according to any one of the preceding claims, wherein: The activity indicator indicates at least one of: the metabolic activity of the particle, the physiological state of the particle, the respiration level of the particle, the ATP level of the particle, the NADH / NAD+ ratio of the particle, the membrane potential of the particle, the growth rate of the particle, the killing rate of the particle, the interaction of the particle with other particles, and the interaction of the particle with environmental factors of the particle, or a combination of any one of the above.

16. A method according to any one of the preceding claims, wherein: The particles are subjected to at least one chemical stimulus and / or at least one physical stimulus, and wherein the particles are preferably cells, and wherein the chemical stimulus is preferably the addition of a drug such as an antibiotic or a compound that affects the metabolism or viability of the cells, or a change in environmental conditions such as a change in culture conditions, and / or Wherein, the physical stimulation is preferably the application of stress.

17. The method according to claim 16, wherein: The activity indicators are determined before and / or during and / or after the particles are subjected to the chemical stimulus and / or the physical stimulus, and wherein the activity indicators thus determined are compared with one another.

18. A method for generating a training data set for a machine learning algorithm, comprising: a) bringing at least one particle into contact with a flexible support (2), b) detecting, using a detection device (3), at least one time-dependent signal indicative of a deflection of the flexible support (2) due to the movement of the at least one particle, and c) evaluating the detected time-dependent signal with an evaluation device (4), wherein the evaluation of the detected time-dependent signal comprises analyzing the time dependency of the signal in order to derive from the signal a plurality of signal parameters characterizing the change of the signal over time, d) performing independent measurements to derive at least one associated activity indicator indicative of the motion activity of said particles, e) storing the plurality of signal parameters and the at least one associated activity indicator in the training data set, and f) Repeating steps a) to e) for a plurality of particles exhibiting different motor activities.

19. A method for training a machine learning algorithm, wherein: The machine learning algorithm is trained using a training data set generated by the method according to claim 18.