EP4058783A1

EP4058783A1 - Open-loop/closed-loop process control on the basis of a spectroscopic determination of undetermined substance concentrations

Info

Publication number: EP4058783A1
Application number: EP21706559.8A
Authority: EP
Inventors: Marek HÖHSE; Alexander Graf; Christian Grimm
Original assignee: Sartorius Stedim Biotech GmbH
Current assignee: Sartorius Stedim Biotech GmbH
Priority date: 2020-04-09
Filing date: 2021-02-18
Publication date: 2022-09-21
Also published as: WO2021204450A1; DE102020002256A1; US20230102813A1

Abstract

The invention relates to a method for the open-loop or closed-loop control of a process (3), more particularly a downstream bioprocess, on the basis of the prediction of an unknown concentration of at least one substance in a sample (5a, 5b) by means of spectroscopy, more particularly UV-vis spectroscopy, comprising the steps of: capturing (101) spectra (1) of a plurality of concentration samples, at least two concentration samples having different concentrations of the substance; generating (102) a plurality of quantitative models (M) on the basis of the spectra of the concentration samples, the models (M) each comprising an association between at least one spectral measurement variable (2) of the spectra (1) and concentrations in concentration ranges, the concentration ranges of two models (M) being non-identical; capturing (103) at least one sample spectrum (1; 1a, 1b) of the sample (5a, 5b); assigning (104) the sample spectrum (1; 1a, 1b) to at least one quantitative model (M) of the generated quantitative models (M); applying (105) the at least one quantitative model (M) assigned to the sample spectrum (1; 1a, 1b) to the sample spectrum (1; 1a, 1b) in order to determine a prediction value for the undetermined concentration; and open-loop and/or closed-loop control (108) of the process (3) with respect to at least one parameter on the basis of the prediction value for the undetermined concentration.

Description

"Process control / regulation based on a spectroscopic determination of undetermined substance concentrations"

The present invention relates to a method, a device and a computer program product for determining an undetermined concentration of at least one substance, in particular a biological substance such as a protein or a DNA in a sample by means of spectroscopy, in particular by means of UV / vis spectroscopy or by means of infrared , Raman and / or fluorescence spectroscopy.

Preferably, the invention can be practiced in the broad bioprocess industries, including the environmental, agricultural and food industries, biofuels and pharmaceuticals.

In purification steps, in particular in downstream processes, which are generally used to separate and purify fermentation products from a fermentation broth of a biotechnological process, concentrations of substances, in particular protein and DNA concentrations, are important parameters or variables that are used throughout Process chain must be followed and determined again and again. This determination usually takes place offline, in particular as a random sample outside the container in which a fermentation broth is subjected to chemical and / or biochemical processes and not in real time. A fermentation broth can be understood, for example, as a biotechnological medium, in particular as a product that can be a medium that contains at least one protein and / or one vaccine and / or one hormone.

The determination or measurement of the concentrations of substances is often carried out by means of optical spectroscopy, in particular by means of UV / vis absorption spectroscopy. As a rule, for this purpose, the molecules of the substance of a sample of the fermentation broth with individual, in particular essentially predetermined discrete wavelengths, the excitation wavelengths or excitation frequencies i correspond, excited and the absorbance or extinction or alternatively the intensity is determined at these essentially discrete excitation wavelengths in order to determine the concentration. For example, narrow-band sensors are often used for this purpose. However, the sensors do not necessarily have to be narrow-band. In the case of sensors, the excitation is usually essentially narrow-band, whereas the wavelength range can be detected or recorded relatively broadly. In other words, on the one hand, only the narrow-band light is used for excitation, whereas the measured or measurable wavelength range can be relatively wide. In spectrometers with such sensors, a dispersive element such as a grating, a prism, etc. is often used.

In spectrometers that are designed to capture spectral information, however, broadband emitting light sources (also called broadband light sources) and correspondingly broadband detecting sensors (also called broadband sensors) are used. A broadband light source can be, for example, a deuterium light source or a halogen light source.

In different stages of the purification or in different process steps, the substance - in particular the protein concentration - can differ considerably. The values for substance concentrations can vary, for example, between approximately 0 and approximately 200 g / l. The hardware, for example the measuring cell or the sensor, is often not suitable for all sub-ranges of the concentrations achieved. For example, because of the limited dynamic range of the detectors, many available online sensors only allow substance concentrations of up to about 10 g / l to be determined. In addition, the optical path or path length must be adapted depending on the concentration range in order to be able to shift the detection range if necessary or to also be able to measure other substance concentrations. It is therefore possible that one and the same hardware, in particular not the same measuring cell, can be used for all applications, in particular for all cleaning steps.

In addition, UV sensors can often have significant undesirable cross-sensitivity between DNA and protein. This is the case, for example, when an increase in the DNA concentration leads to a supposedly higher one measured protein concentration leads. For example, it can happen that an absorption band for DNA overlaps with an absorption band for a protein. Since the bands are not isolated and cannot be recorded separately and independently using sensors, the problem often arises to clearly differentiate and identify which substances (here the protein or the DNA) have changed to what extent in terms of concentration.

There are online UV sensors that work with 1 or 2 individual wavelength ranges (Optek, Pendotech), but which are already saturated at relatively low protein concentrations. In contrast, broadband sensors are often used in spectrometers that are designed to capture relatively broad spectral information. In addition to the information from typical UV sensors, a UV-Vis spectrum can accordingly have a large number of other excitation wavelengths or frequencies. In other words, a spectrum that is emitted by a light source can by all means be wider or more extensive than the part that is measured or detected by the sensor. It depends on the choice of sensors and the light source (s). Such UV / vis spectrometers can accordingly comprise one or about two (possibly also several) light sources, which accordingly generate one or two excitation wavelengths. Corresponding to the fact that, for example, two narrow-band emitting light sources can be used, corresponding sensors can each be designed to detect electromagnetic radiation in the wavelength range that is of interest. For example, UV sensors can detect a few narrow-band excitation wavelengths, in particular one or two, each having a wavelength of around 10 nm. For example, the central wavelengths, that is to say the wavelengths at which the maximum lies, can be approximately 250 nm and / or approximately 280 nm.

Under certain circumstances it can happen that the useful signal saturates. One problem that arises from saturation is the behavior of the absorbance at a defined wavelength and a fixed optical path length. This relationship is based on the Lambert-Beer law: Where A corresponds to the absorbance, 1 _Q corresponds to the intensity of the incident light, I corresponds to the intensity of the light which passes through the sample, e _l corresponds to the decadic extinction coefficient at wavelength 1, c corresponds to the concentration of the substance and d to the layer thickness of the Sample or the path length of the light through the sample corresponds.

The problem of saturation occurs in particular with high absorbances A, since the intensity / at high concentrations c on the measuring channel becomes extremely small and thus possibly no longer distinguishable from the noise of the detector. Therefore, at high concentrations c, the intensity / approaches zero and thus A becomes large and the behavior of the useful signal A changes to a non-linear saturation curve because of the signal-to-noise ratio (SNR) of the detector.

In addition, different substances, in particular different proteins, can have absorption bands which each have a maximum essentially centered around different excitation wavelengths. In other words, the central wavelength of an absorption band or an absorption maximum can vary for different proteins. As a rule, the commercially available UV / vis spectrometers do not adapt the excitation and / or detection wavelength to the substances currently to be measured, in particular the protein currently to be measured. It is therefore a matter of chance whether a protein is measured by chance at the maximum or in the vicinity of the maximum of the absorption wavelength or whether the molecules of a protein are excited with the excitation wavelength around which the absorption band of the protein is centered or a maximum at this excitation wavelength having. With a high degree of probability, however, the previously determined excitation wavelength can lie beyond the wavelength range in which the absorption band has an absorption maximum. For example, the previously fixed excitation wavelength can overlap with a flank region of an absorption band.

The maximum is particularly important or relevant if there is a tendency towards lower concentrations (lower LOD). However, there is usually better dynamic behavior on the flank. A flank range, that is to say a wavelength range for which the absorption levels off to higher or lower wavelengths around the absorption maximum, usually has a low sensitivity to changes in the concentration. The degree of cross-sensitivity also depends in particular on the wavelength-dependent extinction coefficient e _{A of} the target protein and the impurities (HCP) and their concentration ratio. For this reason, absorption values of host cell proteins (HCPs) are often recorded together with those for target proteins (cross-sensitivity is close to 100%) and it is hardly possible to isolate the respective contributions of both absorption values to the analysis of the respective concentration of both substances.

One approach, which leads to the fact that a wide range of different concentrations can be measured by means of UV / vis spectroscopy, comprises a reversible change in the optical path length through a measuring cell which contains the sample. When measuring the transmission of UV / vis spectra, a fiber is made movable so that the optical path length can be varied reproducibly (FlowVPE). Thus, an optical path length can also have much smaller values than the minimum path length of about 1 mm customary for UV sensors. In this way, substance concentrations, in particular protein concentrations above 100 g / l, can also be recorded. This approach has the consequence that there are moving parts in the device and the distance between the fibers must be adjustable with great precision and reproducibility. However, smaller path lengths have the disadvantage that high concentrations are usually associated with high dynamic viscosities, which can, under certain circumstances, represent a disadvantage for the fluidics (including pressure drop, shear flows, etc.).

According to one aspect, the object is to provide a method, a device and a computer program product for determining an undetermined concentration of at least one substance in a sample over a large concentration range.

The object is achieved by the subject matter of the respective main claims. Preferred embodiments are the subject of the dependent claims.

The present invention is based on the generation of quantitative models, in particular on multivariate data analysis. Essential concepts of multivariate data analysis, such as PLS, OPLS, multivariate calibration, multivariate process modeling, hierarchical modeling, etc., are in the book "Multi- and Megavariate Data Analysis - Basic Principles and Applications" ^(3rd Edition) by L. Eriksson et ai. described in detail.

The following aspects can be combined with one another in different embodiments, even if not explicitly mentioned, provided they are not logically mutually exclusive.

According to one aspect, a method for controlling or regulating a process, in particular a downstream bioprocess, based on the prediction of an unknown or undetermined concentration of at least one substance in a sample or in a medium by means of spectroscopy, in particular by means of UV / vis -Spectroscopy, the steps:

Acquisition of spectra, in particular optical spectra, preferably absorption spectra and / or excitation spectra of a plurality of concentration samples, at least two concentration samples having mutually different concentrations of the substance;

Generation of several quantitative models based on the spectra of the concentration samples, the models each comprising an assignment and / or a plot of at least one spectral measured variable of the spectra to concentrations in concentration ranges, the spectral measured variable in particular comprising an absorbance and / or an intensity of the spectra and wherein the concentration ranges of two models are not identical, in particular two concentration ranges overlap or are adjacent to one another;

Acquisition of at least one sample spectrum of the sample or of the medium;

Assignment of the sample spectrum to at least one quantitative model of the generated quantitative models; Applying the at least one quantitative model, which has been assigned to the sample spectrum, to the sample spectrum in order to determine a predictive value for the undetermined concentration; and

Controlling and / or regulating the process with respect to at least one parameter on the basis of the prediction value for the indefinite concentration.

In the present case, an undetermined concentration relates in particular to an unknown concentration or a concentration that has not yet been determined by means of reference analysis. The prediction by means of the method achieves in particular a determination of the concentration.

In contrast to conventional methods in which only absorption values for discrete wavelengths are recorded, a whole series or set of absorption values for a wavelength range can be recorded under this aspect. In particular, entire spectra, in particular absorption or intensity spectra, of dilution samples of a dilution series and a sample are recorded or recorded. In other words, spectra, for example an absorption spectrum, are recorded instead of individual values such as an absorbance for an excitation wavelength.

Another advantage is that a single measuring cell with a constant optical path length can be used for the measurements. A flow cell with a constant optical path length is particularly suitable. Particularly preferred is a flow cell which has a sample layer thickness or an optical path length or a width of the interior between approximately 10 μm and approximately 10 cm, in particular between approximately 15 μm and approximately 5 mm and particularly preferably between approximately 50 μm and approximately 3 mm having. A particularly preferred sample layer thickness is around 1 mm. This has the advantage that one or more samples can be examined several times or repeatedly with one and the same measuring cell.

Depending on which process is to be monitored, the method can be adapted comprehensively to the substance in question for the respective medium to be examined. The method can be used in particular for monitoring a downstream process. Alternatively or additionally, the method also find application for monitoring a fermentation and / or an upstream process.

In the case of fermentation, which can be part of an upstream process, it is advisable, for example, to monitor the fermentation broth within a bioreactor. In a downstream process, for example, the medium to be separated and / or the medium to be purified, for example the fermented broth with the target protein and / or the target DNA, can be used in a purification step, for example before, in and / or after a Filtration system and / or a dialysis unit to be monitored.

For example, the method can be carried out particularly advantageously for analyzing a medium, in particular a fluid medium, in an SU bioreactor if a process in the bioreactor is to be monitored. Such an SU bioreactor, which for example can have an SU bag and / or another SU container and / or a hose with a flow measuring cell, preferably an SU flow cell, contains in particular a fermentation broth that goes through chemical and / or biological and / or biochemical processes . The fermentation broth is then monitored downstream, preferably during purification. In order to monitor the operations or processes within the container, the content is repeatedly analyzed over time and / or analyzed permanently at times.

For this purpose, a container can be equipped with a sensor, for example. Alternatively or additionally, a container can be equipped with a bypass line or sampling line and a flow cell. Alternatively or additionally, a wall projection with a measuring area can be arranged or provided on the container. A spectrometer can be arranged directly or indirectly on the flow cell and / or on the wall projection, which spectrometer is designed to repeatedly or permanently or permanently record data, in particular measured absorption values. In this way, spectra can be recorded over a longer period of time, for example over many hours, in order to follow the processes over this period. For example, the concentration of a protein and / or a DNA can thus be followed and / or monitored. In a system for product purification, in particular for protein purification, the following downstream steps can occur, for example:

1. Cell separation (clarification)

2. Product capture (bind & elute)

3. Polishing steps (cation and / or anion exchange chromatography and / or size exclusion chromatography)

4. Concentration and / or buffer exchange, such as ultrafiltration and / or diafiltration in the crossflow setup.

5. Depletion of viruses, through a filtration step and / or inactivation of viruses, e.g. through pH shift or UV-C treatment, ...

6. Sterile filtration

7. Filling processes (including freeze and thaw)

In addition, an advantage of this aspect is that different substances or substances can be analyzed in isolation from one another, provided that their bands do not have their local absorption or intensity maximum at identical wavelengths. If, for example, the absorption maximum of a “host cell protein” (HCPs) differs in its excitation wavelength and / or in its shape from that of the target protein, an analytical differentiation can be made possible by recording an entire spectrum. When making a distinction, however, the absorption spectrum in comparison to the spectrum of the target analyte is also important. In particular, the width of the peak, the maximum and the shape of the peak are important. A superposition of several protein peaks of the HCPs can then result in a structure in the region of the protein absorption which does not necessarily correspond to a Gaussian-like absorption band. In this way, so-called cross-sensitivities of the measurement method can be suppressed and / or bypassed. An effective quality control of the medium can thus take place, so that it can be determined as precisely as possible how high the respective proportion of, for example, HCP and / or target protein is present in the medium. Based on this, it can be estimated whether further cleaning steps are necessary.

In other words, the aspect mentioned allows in particular the measurement of essentially continuous absorption spectra or absorption spectra comprising discrete measurement points, instead of measuring just individual absorption values at essentially singular wavelengths or measuring individual absorption values at relatively narrow wavelength ranges. For example, a continuous absorption spectrum between a wavelength range of approximately 280 nm and approximately 750 nm can be recorded, whereas a customary known method allows, for example, a measurement between approximately 310 nm and 320 nm. The full width at half maximum (FWHM) can be relatively narrow-band in the case of LEDs and can be around 10 nm, for example. The absorption spectroscopy, in particular the UV absorption spectroscopy and the corresponding sensors, is preferably based on a comparatively broadband excitation or on a broadband emitted and correspondingly recorded spectrum. Light sources can represent, for example, a deuterium lamp and / or a xenon lamp and / or one or more LEDs.

The invention is based on the Lambert-Beer law. In particular, it is avoided that the path length has to be varied in the case of fixed waves (or in the case of a few wavelengths that can be selected beforehand via the light source or detector / filter). The invention, on the other hand, allows the entire spectrum to be recorded and thus allows the possibility of choosing the effective e with a fixed or unchanged path length d. This is done via the selection of the wavelengths l to be used explicitly or using a multivariate algorithm.

In particular, said aspect allows the system to be monitored using a single flow cell with a constant optical path length. Depending on the concentration range to be measured, various multivariate models (e.g. PLS with substance concentration, in particular protein concentration as target variable and the absorption values of the respective spectral range as input variables) are created, which are based on the evaluation of different wavelength ranges. These can either be selected yourself or selected using an algorithm and / or weighted differently (e.g. using PLS / Partial Least Square or MLR / Multiple Linear Regression), the selected wavelength ranges being determined by the absorbance of the target analyte (e.g. DNA or protein) or change the matrix can be influenced. In particular, be at least two non-identical sub-ranges, in particular excitation wavelengths of a range, are selected according to the criterion whether the maximum intensity of the absorption band is still within the usable dynamic range of the detector.

In particular, the individual wavelength ranges have different sensitivities with regard to the detection of the substance concentration for the absorption values selected for the models. Thus, the dynamic range differs depending on the model. The same applies analogously to the achievable analytical accuracy of the measurement.

The spectra preferably include absorption spectra and / or intensity spectra, the absorption spectra and / or intensity spectra having a one-to-one assignment, in particular a one-to-one plot of the spectral measured variable, for example the absorption value and / or the intensity value, at several wavelengths, with preferably at least a subset of the absorption values or Intensity values that are assigned to the corresponding wavelengths have no saturation on the detector side. The spectral measured variable can in particular be the absorption value or the intensity value.

A saturation on the detector side occurs when there is no longer a concentration-dependent change in the detector signal at a specific wavelength or a wavelength range.

In particular, a separate weighting or weighting of each wavelength can also take place. In this case it is usually not formally necessary to select wavelengths. A weighting close to 0 does essentially the same thing.

A detector signal can enter a saturation range if the electromagnetic radiation has too high an intensity or too high an absorbance at a certain wavelength or in a wavelength range, so that this corresponding value is outside the dynamic range of the detector. In other words, saturation can take place in both directions of the “detector filling”, ie with regard to an “overflow” or an “underflow”. In UV spectroscopy, if the sample is saturated, it absorbs too much light, so that the light arriving at the detector can no longer be distinguished from the dark noise of the detector.

If saturation is reached, the detector cannot provide a resolution of the intensity or the absorbance within the saturated area. For example, this can be the case if there is a very high concentration in a sample and / or a long optical path length is due to a very thick measuring cell, so that the absolute number of molecules that interact with the light or the electromagnetic radiation is particularly important is high. To determine the dependency of an absorbance and / or intensity, it is advantageous if the wavelength range to be investigated has essentially no saturation for all concentrations to be investigated, so that the differences in absorbance and / or intensity are resolved and / or the dependence on the concentration a concentration series can be displayed if necessary. If a wavelength range experiences saturation, the detector is blind to the actual spectral measured variable, that is to say the absorbance and / or the intensity.

The quantitative models preferably include at least one global model comprising a plurality of sub-models that do not overlap one another.

For example, a global model can have values from about 0 g to about 150 g. Possible local models have values from about 2 g to about 10 g, from about 10 g to about 50 g and from about 30 g to about 80 g. A sub-model does not necessarily have to have a common limit value with the global model. Limit values can be identical, but this case is an exception.

A global model allows a process unit to estimate a rough dependency between the spectral measurand of a spectrum and different concentration ranges. The sub-models can essentially comprise relatively roughly graded concentration ranges that are limited to ever smaller ranges. The sub-models of a global model can be graded, for example, into a first range from about 0 g / l to about 5 g / l, in a second range from about 0 g / l to about 25 g / l and in a third range from about 0 g / l to about 75 g / l.

However, sub-models do not necessarily have to start at 0. This can possibly even be unfavorable, since in this case the model area has to be kept unnecessarily wide. For example, if the process step contains concentrations from about 50 to about 75 g, an optimal model will contain a slightly larger range (e.g. 45-80 g / l).

The quantitative models preferably each include at least one individually selected sub-model.

A sub-model can include any concentration range. In particular, several sub-models can be created, each with different concentration ranges. In this case, at least two concentration ranges can essentially overlap, essentially adjoin one another and / or be spaced apart from one another. A sub-model can in particular be a real subset (e.g. with a restricted y-range) of the superordinate model spectrum set.

The quantitative models preferably include hierarchically organized models corresponding to at least a first hierarchical level and a second hierarchical level, the first hierarchical level in particular comprising a global model and the second hierarchical level comprising a local model.

In a first hierarchy level, a global model of the process unit and / or the user allows a rough dependency between the spectral measured variable of a spectrum and different concentration ranges to be recognized and / or assessed, for example automatically or automatically. The sub-models can essentially comprise relatively roughly graded concentration ranges that are limited to ever smaller ranges. The sub-models of a global model can, for example, be graded into a first range between approximately 0 g / l and 5 g / l, a second range between approximately 0 g / l and 25 g / l and a third range between approximately 0 g / l l and 75 g / l. In a second hierarchical level, a local model of the process unit and / or the user allows a more finely resolved dependency between the spectral measurement of a spectrum and different concentration ranges to be recognized and / or assessed, in particular recognized automatically or automatically. The sub-models can essentially comprise relatively finely graduated concentration ranges. The sub-models of a local model can, for example, be graded into a first range between approximately 0 g / l and 5 g / l, a second range between approximately 5 g / l and 25 g / l and a third range between approximately 25 g / l l and 75 g / l. In this respect, the concentration areas of a local model are essentially adjacent to one another.

The assignment of the sample spectrum to at least one quantitative sub-model preferably includes an estimate based on determined and in particular based on displayed data, in particular based on displayed spectra.

A quick, uncomplicated estimation, for example “by hand” and / or on the basis of a rough approximation, can provide a first approximate result, which can then be verified and / or specified more precisely using a fit, for example.

The assessment by the user occurs in particular when selecting the sub-models. In one example, the user can choose the 0-50 g / l model because, for example, a content of around 40 g / l is expected at the time of measurement. Alternatively, a classification can be carried out via PLS-DA or SIMCA or Euclidean distances or Mahalanobis distance between model and current spectra can be determined. This shows which of the models - based on different concentration ranges - the current spectrum best fits.

The sample spectrum is preferably assigned to at least one quantitative model (M) by means of Euclidean distance and / or Mahalanobis and / or PLS-DA and / or SIMCA.

Assigning the sample spectrum to at least one quantitative model by means of Euclidean distance and / or Mahalanobis and / or PLS-DA and / or SIMCA has the advantage that precise predictions about the concentration of a substance in a medium or in a sample can be taken. Accuracy of the concentration prediction achieved. When used, some processes can have a very wide range of protein concentrations that should preferably be covered. It may be that the model areas are too small and, as a result, it may be necessary to switch between the models within a process

The accuracy of the prediction depends in particular on the area of the model. It may be that this range is too small in relation to the error of the reference and as a result an algorithm cannot determine which wavelengths are required to determine the analyte concentration. It may also be that the range is too large and, as a result, it will not always be the highest, which may cause a jump in the forecast.

The control and / or regulation is preferably carried out with regard to the at least one parameter and the parameter comprises at least one of the following parameters: a valve circuit, a pump control, a media flow, a media pressure, a gas pressure, a pH value, a filtration step, a fractionation step, a time control, a process time setting, a temperature, an ion concentration, a chromatography step.

The process control to which the forecast value is passed on can influence the process to the desired extent on the basis of the forecast value determined. A process can thus be controlled and / or regulated efficiently based on a specific prediction value. The feedback can in particular serve to make the process more efficient and / or to obtain the product or the substance particularly gently and / or to protect it from possible denaturation and / or damage due to undesired values of the process parameters. The process parameters that can be influenced, for example, include the following: a valve circuit, a flow pressure, a gas pressure, a pH value, a filtration step, a fractionation step, a time control, a process time setting, a temperature, an ion concentration or a salt concentration or a titer, a gradient, a chromatography step. The at least one sample spectrum of the sample is preferably recorded online or in real time, in particular during or during the process, and the control and / or regulation also preferably takes place during the process as a direct consequence of the determined prediction value for the undetermined concentration.

Real-time monitoring allows the process unit and / or the user and / or an automated apparatus to intervene immediately in the process parameters on the basis of the determined prediction value or on the basis of the determined concentration of the substance in question. Immediate intervention in the process allows, in particular, the regulation and / or control of the process. In other words, feedback can take place in such a way that at least one process parameter is monitored and / or controlled and / or regulated on the basis of the determined concentration value of at least one substance. This also avoids having to take a random sample of the medium and measure it ex situ. The sample or the medium can take place in situ, in particular during a process within a system, for example within a fermentation vessel and / or within a downstream system, for example upstream and / or downstream of a chromatography unit.

Quantitative models are preferably generated by means of multivariate model creation, in particular by means of a multivariate regression method, such as preferably by means of multiple linear regression (MLR), and / or partial least squares (PLS) and / or orthogonal partial least squares (OPLS).

A multivariate regression method allows efficient automated determination or creation of models with a high degree of precision. Parameters which determine the desired conditions in the automation, for example maximum and / or minimum values for process parameters, can be specified by the process unit and / or the user. The process can thus be automated or automatically monitored in real time even in the absence of a user. In particular, at least a partial range, in particular individual wavelengths of the spectra of the majority of the concentration samples, is weakened or excluded by weighting.

For example, at least one sub-range, in particular individual wavelengths of the spectra of the majority of the concentration samples and preferably sub-ranges for which the spectra are saturated, are weakened and in particular excluded for use on the sample spectrum.

By weighting, for example, by a factor (factors normalized to 1 in this example) that is very much smaller than 1, for example for values that are smaller than about 0.1, in particular smaller than about 0.05 and preferably smaller than about 0 , 01, it can be ruled out that undesired data, for example absorption values for which saturation occurs, automatically but undesirably enter into the creation of a model and / or the determination of the concentration of a substance. This can also be the case for concentration ranges for which an absorbance at a wavelength shows a dependence on the concentration which deviates greatly from a linear concentration. Concentration ranges and / or wavelengths in which the absorbance has a particularly low sensitivity to the change in concentration can also be affected. Weighting can for example take place automatically on the basis of predetermined criteria and / or can be predetermined by the process unit and / or the user. For the creation of models, it is particularly preferred to take into account or to prefer a linear dependence of the absorbance on the concentration of a concentration series, since the prediction or determination or estimation of the concentration can thus be carried out particularly precisely.

Preferably, at least one sub-range, in particular individual wavelengths of the spectra of the majority of the concentration samples, preferably sub-ranges for which there is a linear dependence between wavelengths and measured values, are weighted more heavily for application to the sample spectrum, the weighting being manual or by means of an algorithm, in particular Factor analysis, factor loadings or RLS loadings, regressions in RLS and / or OPLS takes place.

For the generation of models, it is particularly advantageous to weight areas for which there is at least an approximate linear dependence of the absorbance on the concentration of a concentration series. The prediction based on a linear dependency of the concentration can be made particularly precisely.

The acquisition of a large number of sample measured values is preferably carried out with a single, essentially identical gap width of one or more measuring cells.

Maintaining a single measuring cell enables the process unit and / or the user to determine the concentration particularly efficiently, since the replacement of hardware, in particular measuring cells, can be reduced and possibly completely avoided. The circumstances under which measurements are carried out are thus constant and variations in the measurement configuration and the resulting incorrect settings are avoided. For example, the reversible setting of a measuring cell to a certain layer thickness can be flawed. In addition, an additional method step or several additional method steps, which include setting the layer thickness, are avoided, which makes the method particularly efficient.

According to one aspect, a device for controlling and / or regulating a process, in particular a downstream biopnocess, based on the prediction of an undetermined concentration of at least one substance in a sample or a medium by means of spectroscopy, in particular UV / vis spectroscopy, comprises:

At least one interface for obtaining spectra of a plurality of concentration samples of the at least one substance and of at least one sample spectrum of the sample or of the medium;

At least one unit for

Generation of quantitative models based on the spectra of the concentration samples, Assignment of the sample spectrum to at least one quantitative model of the generated quantitative models; and

Applying the at least one quantitative model, which has been assigned to the sample spectrum, to the sample spectrum in order to determine a predictive value for the undetermined concentration;

At least one process unit for controlling and / or regulating the process with respect to at least one parameter on the basis of the prediction value for the undetermined concentration.

The device preferably comprises a spectrometer for recording the spectra of the plurality of concentration samples of the at least one substance and / or of the at least one sample spectrum of the sample.

According to one aspect, a computer program product comprises computer-readable instructions which, when the program is executed by a suitable computer-aided system, cause the program to carry out the steps of the method according to the above aspect or one of the embodiments.

All optional and preferred features of the aspects, examples and embodiments of the method can correspondingly be combined with the device and / or the computer program product. As a result, there are the corresponding advantages and technical effects which are mentioned herein.

In particular, the invention can be implemented with single-use (SU) - or disposable systems, comprising, for example, SU downstream elements, SU fermentation containers, SU bags, SU containers each with a wall projection which comprises a gap and window through which the The contents of the container can be analyzed spectroscopically in situ, i.e. in the container in real time. Tube lines with SU flow lines can also be used with preference, which can also be combined or directly connected to SU containers.

Some exemplary embodiments are described in more detail below, the invention not being restricted to the exemplary embodiments described. Individual features that are described in a specific embodiment can be combined as desired, provided that they are not mutually exclusive. Furthermore, various features that are provided together in the exemplary embodiments are not to be regarded as limiting the invention.

Brief description of the drawings

Add 1 is a schematic illustration of exemplary process monitoring for a downstream process;

2 is an illustration of an exemplary spectrum of a protein band with marked wavelength ranges for various models;

FIG. 3 is an illustration of an exemplary plot of absorption values of the exemplary protein band shown in FIG. 2 corresponding to different concentration samples against the corresponding concentration at three different excitation wavelengths; FIG.

4 is an illustration of an exemplary plot of absorption values of different concentration samples against the corresponding concentration at two different excitation wavelengths;

5 is a schematic illustration of an exemplary method for controlling a process; and

6 shows the exemplary prediction compared to the reference value for DNA.

1 is a schematic representation of an exemplary process monitoring for a downstream process 3. The process monitoring is implemented in this example by means of a UV / vis spectrometer 4. The two representations of UV / vis spectrometers 4 shown in FIG. 1 can also be only a single UV / vis spectrometer 4. The at least one UV / vis spectrometer 4 is connected or coupled to an evaluation unit. The evaluation unit 9 in the present case comprises a computer with a desktop b or an output unit or a screen 9 _a on which the measured spectra 1 _a , 1 b can be displayed or represented. The evaluation unit 9 can comprise an interface for receiving spectra, such as an interface for connecting a data cable. Additionally or alternatively, the UV / vis spectrometer 4 can also have a Include an interface for obtaining spectra, so that the data can be tapped, for example, by a device and / or an evaluation unit 9 and / or a data carrier.

A aufzureinigendes medium 5 _a 7a can pass through a first measuring cell before it passes through the purification unit 6 and passes through. The first measurement line 7a allows the process unit and / or the user to record a first spectrum 5a of the medium. The first measuring cell 7a comprises in particular a flow cell. In this step, the medium 5a can in particular comprise one or more substances to be monitored. In addition, the medium 5 _{a can} in particular comprise impurities or substances that are to be separated from the medium 5 a. For example, the medium 5 _a can comprise HCPs which are to be separated or separated or isolated _{from the medium 5 a.} The screen 9 _a in FIG. 1 exemplifies a spectrum 1 _a, in particular an absorption spectrum of the medium 5 _{a to be} cleaned, which has two bands a, b. The first band a can for example be caused by the contamination and the second band b can for example be caused by the target substance, for example the target protein. As an alternative to the first measuring cell 7a, the medium 5a to be purified can also be examined inside a container. For example, a range of 1 _a of the medium to be purified 5a can be at least partially received or measured or detected within a fermentation vessel.

The medium 5 passes through _a downstream of the first cell 7a, a purification unit 6. The purification unit 6, for example, the following method for purifying or T SEPARATION can enable: filtration, chromatography, and in particular

Column chromatography and / or high-performance liquid chromatography, electrophoresis, extraction, precipitation, sedimentation, centrifugation and / or dialysis etc. The purification unit 6 is particularly designed _{to isolate or separate the medium 5 a} from impurities and / or undesired substances, so that the medium 5b after passing through the purification unit 6 at least essentially no longer includes the undesired substance, for example HPCs. The essentially purified medium 5b, on the other hand, only includes what is desired Substances such as target molecules, in particular target proteins and / or salts, in particular buffer substances.

The at least partially purified medium 5b leaves the purification unit 6 after the purification step and passes or passes through a further measuring cell 7b. Alternatively, it can also be the first measuring cell 7a, in particular it can be the first cleaned measuring cell 7a, but it is preferred that the cleaned medium 5b passes a non-identical further or second measuring cell 7b, since it may be in the first measuring cell 7a could be contaminated with the impurities. The second measuring cell 7b can, however, be an identical model of the first measuring cell 7a.

When the medium flows in, in particular through, the further measuring line 7b or is at least partially located therein, a further spectrum 1b can be recorded by the UV / vis spectrometer 4. The screen 9 _a in FIG. 1 exemplifies a spectrum 1 b, in particular an absorption spectrum of the at least partially purified medium 5 b, which essentially only has band b. The first band a, which was still _{visible in the first spectrum 1 a} and which was caused, for example, by the contamination, is at least no longer visible in the displayed spectrum 1 b of the purified medium 5 b. The second band b, which is caused, for example, by the target substance, for example by the target protein, is still visible, however. This indicates that the substance which generates the first band a has largely been separated from the medium 5b in the purification unit 6.

Two optical light guides 8 are arranged on and / or in the measuring cells 7a, 7b. In this case, electromagnetic radiation can be coupled into one of the two optical light guides 8, for example from the side of the UV / vis spectrometer, which emerges from the optical light guide 8 on the side of the measuring cell 7a, 7b and at least partially passes through the measuring cell 7a, 7b, so that an optical light guide 8 on the opposite side of the respective measuring cell 7a, 7b can at least partially detect the transmitted electromagnetic radiation and at least partially transmit or guide it to the UV / vis spectrometer 4. Alternatively, the process monitoring can also take place on and / or in a container. For example, at least one optical light guide can be guided to and / or into a fermentation container so that it can stimulate the medium in the fermentation container by means of coupled light and / or the light that passes or has passed through and possibly stimulated at least part of the medium, can capture or lead to a detector or a sensor. On the one hand, the monitoring can take place at least partially in situ and / or ex situ. In other words, elements that may be required for monitoring, such as sensors and / or light guides, protrude into the interior of the fermentation vessel and in particular are in direct physical contact with the medium, in particular the fermentation broth (in situ), and / or from the outside , for example on a window with optical access to the interior, so that direct physical contact with the medium can be avoided (ex situ).

The measuring cells 7a, 7b in particular have an essentially constant optical path length or thickness. The optical path length of the light or the thickness of the measuring cell 7a, 7b can be between approximately 100 nm and approximately 10 cm, in particular between approximately 500 nm and approximately 5 mm and particularly preferably between approximately 800 nm and approximately 3 mm.

FIG. 2 is an illustration of an exemplary spectrum of a protein band with marked wavelength ranges for various models. The absorbance 2 is plotted between about 240 nm and about 360 nm against the wavelength l and thus the spectrum 1 results, in particular the absorption spectrum for a band that has its local maximum between about 270 nm and 280 nm and one on both sides showing falling edge. The absorption band is caused by an exemplary protein.

The invention is based on the measurement or acquisition of absorption spectra instead of individual wavelengths and / or wavelength ranges. During the measurement there is a medium in a measuring cell, for example in a flow cell, which has an essentially constant thickness between approximately 100 nm and approximately 10 cm, in particular between approximately 500 nm and approximately 5 mm and particularly preferably between approximately 800 nm and approximately 3 mm, for example of about 1 mm. Depending on the concentration range to be measured, various multivariate models are created that are based on the evaluation of different wavelength ranges. The models can be based on PLS, for example, with the concentration of a predetermined protein being defined as the target variable and the absorption values of the respective spectral range serving as input variables.

The wavelength ranges can either be selected manually and / or determined using an algorithm and / or weighted differently. All selected wavelength ranges can be influenced by the absorbance of the target analyte (e.g. DNA or protein), the preferably linear range of the detector response or the changes in the matrix, e.g. due to particle scattering or changes in the buffer composition.

In FIG. 2, individual wavelength ranges are indicated which are each assigned to a model M1, M2, M3, M4. For example, the wavelength range assigned to model M1 is approximately between 270 nm and 280 nm. The wavelength range assigned to model M2 is between approximately 280 nm and 290 nm. The wavelength range assigned to model M3 is between approximately 290 nm and 300 nm and the wavelength range assigned to model M4 is between approximately 300 nm and 310 nm.

The individual wavelength ranges have different sensitivities with regard to the detection of the analyte concentration or the concentration of the substance in question. For example, model 1, which is assigned to the wavelength range in which the maximum of the absorption band falls, has a particularly high sensitivity compared to the other models. The models that are assigned to the wavelength ranges that are essentially in the flank of the absorption band each show a decreasing sensitivity compared to one another from model 2 to model 4. In other words, for the classification of the models shown here, the sensitivity decreases with increasing wavelength. The dynamic range thus differs depending on which model M1, M2, M3, M4 is selected. The same applies to the achievable analytical accuracy of the measurement. Compared to the trend in sensitivity, the opposite is true for the absolute measurable concentration of a substance. Model 1, which is assigned to the wavelength range in which the maximum of the absorption band falls, has a low potential for detecting high concentrations. In contrast, this potential increases with increasing wavelength in the present case. The models, which are assigned to wavelengths at the edge of the flank of the band, show a particularly high ability to indicate high concentrations.

If the model is based on the wavelengths with the highest absorbance or the maximum of a band, this model has the highest sensitivity, but the upper detection limit is reached quickly even at low substance concentrations. The M4 model, on the other hand, is mainly based on wavelength ranges that lie in the flank of the band. This means that the sensitivity is lower, but much higher substance concentrations can be measured. In particular, because of the high sensitivity, the maximum of an absorption band at a relatively low concentration reaches very high values which in particular exceed the dynamic range of a sensor or detector.

In other words, the individual wavelength ranges have different sensitivities with regard to the detection of the analyte concentration. The dynamic range thus differs depending on the model, the same applies to the achievable analytical accuracy of the measurement. Figure 2 illustrates this. It shows an example of an absorption band of a protein or a protein species. If the model in the core is based on the wavelengths with the highest absorption, this model has the highest sensitivity, but the upper detection limit is reached quickly even at low protein concentrations. In contrast, model 4 is mainly based on wavelength ranges that lie in the flank of the band. This means that the sensitivity is lower, but much higher protein concentrations can be measured.

This is also evident in the different linear areas of the different models. 4 shows the absorbance or absorption of two protein bands at different concentrations. The band at 224 nm changes a lot stronger with increasing protein concentration than the 280 band (and thus has a higher sensitivity). On the other hand, the linear range in the 280 band is much larger.

FIG. 3 shows how the absorbance of the models based on different wavelength ranges of a single band behaves with changing protein concentrations.

On the basis of the respective plots of the absorption values 2 for respectively determined wavelengths l as a function of the concentration, the previously described dependency is described in the following FIGS. 3 and 4.

FIG. 3 is a representation of the plot of absorption values 2 of the protein band shown in FIG. 2 for different concentrations c in each case at three different excitation wavelengths. However, the protein band shown in FIG. 2 corresponds to only a single selected concentration. In other words, the absorption 2 is plotted against the substance concentration c for different wavelength ranges of the protein band with the maximum at about 280 nm. In FIG. 3 it is shown in particular how the absorbance 2 of three models based on different wavelength ranges of a single band behaves with changing substance concentrations c.

The three models that were selected by way of example with regard to FIG. 3 are assigned wavelength ranges which each include the value of approximately 300 nm, 290 nm and 280 nm. The recorded neighboring data points of a model were each connected with a linear connecting line. In particular, models in areas that show essentially linear dependencies between absorbance 2 and concentration c are suitable for the reliable prediction of indeterminate concentrations. As can be seen from the illustration in FIG. 3, the wavelength ranges or the models have different, essentially linear ranges.

For example, the plot for about 300 nm shows a linear range over a particularly broad concentration c, namely between about 0 g / l and 150 g / l. Of the The entire range can be used to predict an indeterminate concentration, although the sensitivity or sensitivity of this wavelength to the concentration c is not high compared with the other models. The plot for about 290 nm shows a high sensitivity compared to the plot for about 300 nm. The linear range within which reliable predictions can be made, however, ends for values from about 50 g / l. The plot for 280 nm shows a comparatively high sensitivity to changes in the concentration, but only the concentration range below about 10 g / l is linear. With regard to the spectrum shown in FIG. 2, the value of approximately 280 nm is relatively close to the maximum, whereas the other two values of approximately 290 nm and 300 nm lie in the region of the falling edge of spectrum 1.

In other words, depending on which model was selected, linear areas of different sizes can be used for the evaluation. The three models shown have different sensitivities, especially at low substance concentrations. The model for 280 nm shows the highest sensitivity and thus also measurement accuracy at low substance concentrations, but the model at 300 nm has a much larger linear range up to 150 g / l protein. In this way it is possible to do justice to different applications, for example the monitoring of a DNA concentration and / or the concentration of different proteins, whereby the hardware can be retained and in particular no variation of the optical path length is assumed.

FIG. 4 is an illustration of the plot of absorption values for two exemplary bands at two different excitation wavelengths for different concentrations. In other words, FIG. 4 shows the respective absorbance of two protein bands at different concentrations. The band at 224 nm changes much more strongly with increasing substance concentration than the band at 280 nm and thus has a higher sensitivity. On the other hand, the linear range for the band at 280 nm is much larger. FIG. 5 is a schematic illustration of an exemplary method 100 for controlling a process. The method begins with step 101, in which a plurality of spectra of a concentration series of a specific substance, the concentration of which is to be determined in a medium or in a sample. For example, the protein to be separated can be a photosynthesis protein. For example, a dilution series can include staggered concentration ranges. In particular, the concentrations of a substance can be at constant intervals for predetermined areas, wherein they can increase and / or decrease the constant intervals in areas. Additionally or alternatively, the distances can also be selected to be non-constant for at least one area. Such an example of a concentration series is given below: 100 mg / l; 400 mg / l; 700 mg / l; 1g / l; 1.5 g / l; 2 g / l; 2.5 g / l; 5 g / l; 7.5 g / l; 10 g / l; 20 g / l; 30 g / l; 40 g / l; 50 g / l; 75 g / l; 100 g / l; 150 g / l etc.

The concentration series preferably only comprises the substance to be determined in different concentrations dissolved in a suitable, essentially physiological solvent, in particular in an aqueous buffer solution, for example in a phosphate-buffered saline solution or a TBS buffer. If necessary, the solution can also comprise a detergent and / or an addition of another substance to prevent possible denaturation of the substance.

In particular, a measuring cell with a constant gap width or sample layer thickness is used to record the spectra of the concentration series. Alternatively, two optical light guides can be arranged at a constant distance at least partially within a container in which the medium is at least partially, one non-conductor being designed to emit light or electromagnetic radiation into the medium and the other light guide designed to do so is to at least partially absorb the transmitted radiation and forward it to an evaluation unit and / or the UV / vis spectrometer.

For each concentration sample of the concentration series, at least one spectrum is recorded, in particular in a range between approximately 200 nm and 700 nm. Using the spectra of the concentration series, in which absorption values are usually plotted against the wavelength, suitable models for different concentration ranges are created in step 102. In particular, multivariate models are created, preferably using partial least square (PLS) regression, orthogonal RLS (OPLS) regression and comparable regression concepts.

In particular, hierarchical models can be selected, each model covering a specific concentration range. For hierarchical models, the rule is that the precision is high for low concentrations and decreases at higher concentrations and, as a result, model errors may occur.

The method 100 offers two combinable options, namely option A, a global model and option B, a local model. The global model of option A includes the exemplary sub-models, namely models M1, M2 and M3, each based on spectra measured, for example, between approximately 0 g / l and 5 g / l, between approximately 0 g / l and 25 g / l and between about 0 g / l and 75 g / l. Additionally or alternatively, a local model can also be used. The local model of option B includes the exemplary sub-models, namely models M4, M5 and M6, each based on spectra measured, for example, between approximately 0 g / l and 5 g / l, between approximately 5 g / l and 25 g / l and between about 25 g / l and 75 g / l. The main difference between option A and option B is that for option B, the models M4, M5 and M6 do not all start at around 0 g / l, only the first model M4.

Optionally, a hierarchical model can be selected that includes a global model (option A) over the entire concentration range (for example between about 0 g / l and 75 g / l) and represents a top hierarchy and further sub-models M4, M5, M6 (option B ) for individual concentration ranges (for example staggered between about 0 g / l and 75 g / l).

Sub-models are generally based on unsaturated spectra or spectral ranges, which are determined, for example, by means of an OPLS / PLS algorithm and / or by plotting the absorbance against the concentration of the relevant wavelengths. In other words, is the goal of a model in particular, to build on unsaturated spectral ranges. This applies to both the global model and the sub-models. In particular, the linear range of each plot is determined with respect to each selected wavelength, which is followed by a decision as to whether the selected wavelength of a model can be used for a specific concentration range. In the linear range, the slope of the detector signal is usually maximum when the concentration changes. Above the linear range, the slope gradually decreases until there is essentially no change in the detector signal with a further increase in concentration. The non-linear area can still be evaluated, but the prediction accuracy is lower in this area and the model is more complex, for example due to additional main components. Standard MVDA methods are usually multilinear methods and therefore these non-linearities or non-linear areas cannot be optimally modeled. Therefore, linear areas of the plot are particularly suitable for determining an undetermined concentration of a substance.

A data set for modeling comprises a large number of spectra of samples with a known analyte content (eg protein). Input variables are a large number of intensities at different wavelengths (X values) as well as reference values such as protein content or concentration (Y values). The algorithm finds correlations in the multivariate space of the X values, which correspond to the course of the Y values. It is in particular the reference book "Multi- and Megavariate Data Analysis - Basic Principles and Applications" ^(3rd Edition) by L. Eriksson et al. referenced, in which (especially the 4th chapter) a detailed introduction to modeling, especially using PLS, can be found.

In the following step 103, an online measurement or a real-time measurement of at least one sample spectrum of a sample or a medium, which in particular includes the substance to be determined, is recorded. The term online measurement or real-time measurement relates to the fact that the medium or the sample is examined immediately before and / or after and / or during a process, so that a dynamic or direct based on the determined concentration of the substance is examined or several process parameters can be influenced. In step 104, the sample spectrum that was recorded in step 103 is classified in or assigned to at least one suitable model. Possible methods for classification include, for example, PLS-DA (Partial Least Squares Discriminant Analysis), SIMCA (Soft Independent Modeling of Class Analogy) calculations of the Euclidean distances or a dendrogram. The sample spectrum is compared with the spectrum sets of the individual models and a classification according to the best fit or best fit result is made. In other words, in particular at least one, but preferably several absorption values of a sample spectrum are compared with the corresponding absorption values of the spectra of different models.

Additionally or alternatively, the classification can also be based on process knowledge or known process parameters, e.g. on known concentration ranges of the process step.

The sample spectrum results in particular from a measurement, in particular from an online measurement in the process. In particular, an optimally applicable model can be selected. One criterion can, for example, be the expected concentration of the sample. Due to the process, the user can therefore classify relatively roughly whether the concentration of the sample is, for example, around 15 g / l or around 150 g / l, but a fine classification can only be made using the correspondingly selected model. It may therefore not be possible to estimate whether the concentration is around 12 g / l, 15 g / l or 17 g / l without a suitable model. A suitable model would be, for example, a model between 5 and 20 g / l.

In step 105, the model is applied to the sample spectrum in order to essentially at least approximately determine or predict the unknown concentration of the substance. The prediction value or the determined concentration of the substance is then output by the computing unit in step 106 and passed on or sent to the process control in step 107. On the basis of the concentration value determined, the process control, to which this value is passed, can, in step 108, refer to the process in the desired Take influence. In this way, a process can be controlled and / or regulated based on a specific prediction value.

In other words, in step 104 a current spectrum or a sample spectrum is assigned to a suitable model. With the aid of the selected model, the current or the concentration of the substance to be determined is determined in step 105. In step 106 the determined value for the concentration of the substance is output to an interface or an interface and passed on to a control unit in step 107. On the basis of the determined concentration value, the process control, which receives the prediction value, can influence the process, in particular one or more process parameters, to the desired extent in step 108.

The modeling, in particular the generation of several quantitative models based on the spectra of the concentration samples, is expressly different from a usual calibration. For example, a typical calibration can essentially be based on a linear regression between a wavelength and a protein concentration. A linear regression is essentially based on a fit or a comparison calculation with a linear model. A calibration can also be based on non-linear models. In contrast, the modeling described here is based in particular on PLS models.

In particular, PLS models ("Partial Least Squares") are based on the main components of two matrices X and Y. The main components should be calculated separately for the matrices X and Y in order to create a regression model between the entries or scores of the main components (and not the original data). For this purpose, the matrix X is broken down into a matrix T (the "Score" matrix) and a matrix P "(the" Loadings "matrix) plus an error matrix E. The matrix Y is broken down into the matrices U and Q and the error term F. These two equations can be referred to as the "external relationships". The aim of PLS is, in particular, to minimize the norm of F and at the same time to obtain a correlation between X and Y, in which the matrices U and T are related to one another as follows: U = BT. This equation can also be referred to as the "inner relationship". The control can in particular have valves and pumps. Otherwise, the corresponding collected volume can be discarded, especially if it contains impurities. In addition, changes in the pH value, conductivity, concentration and / or other parameters can be achieved by means of appropriate additions of substances and, in particular, these and / or other parameters can be regulated.

The term spectrum essentially means the absorption spectrum. However, the term can also refer to an intensity spectrum, provided that absorption spectroscopy is not involved. This is the case, for example, for Raman spectroscopy, which does not measure the absorbance of light, but rather light with a “shifted” wavelength, triggered by the inelastic scattering of light on molecules. Without being restricted to UV / vis spectroscopy, however, the present embodiments essentially relate to the absorption spectra generated thereby.

Mainly absorption bands of proteins are shown and discussed here. However, the invention can also be based on all other substances, in particular molecules, which can be examined by means of molecular spectroscopy.

The term “sample” is not limited to a random sample that can be examined ex-situ outside a container, for example outside a bioreactor. Rather, the term sample relates to the medium, in particular a liquid, a gas, a solid and / or any mixture thereof, which comprises the substance and is essentially located within the container during the measurement. The sample is measured in particular in situ, i.e. essentially within the container in which the process also takes place. In particular, the sample is also not discarded but essentially remains with the entire medium in the container and, if necessary, remains exposed to all further process steps. A sample may not be an identical medium. That is, the sample can mean the medium that happens to get into the measuring cell, for example into a flow cell, and is analyzed there. The molecules and or atoms must must not necessarily be the same for each measurement of a permanent monitoring, rather the sample is determined by chance with each measurement. When measuring in containers, the sample is particularly representative of the bulk. When measuring in hose lines, the sample is particularly representative of the amount in the corresponding hose area.

Host cell proteins (HCPs) can be process-related contaminants expressed by the host cell used to make biopharmaceutical proteins. On the one hand, the host cell can express HCPs or, on the other hand, it itself consists of HCPs, which are released upon cell disruption or cell death. During the cleaning process, most of the HCPs are usually removed (especially> 99%). If necessary, however, the remaining amounts of HCP can remain in the products, such as monoclonal antibodies (mAbs), antibody-drug conjugates (ADCs), therapeutic proteins, vaccines and other protein-based biopharmaceuticals, among others.

A UV / vis spectrometer, which is designed to detect essentially spectral information, can have a broadband light source. On the other hand, an essentially broadband emission spectrum can also be generated through the use of several narrow and / or broadband emitting light sources. An example of a light source emitting essentially broadband is a deuterium lamp (D2 lamp) or a halogen lamp.

The term “narrowband ig” with regard to the emission spectrum of an LED light source relates to light sources that emit electromagnetic radiation with an FWHM (spectral half-width) of values that are less than about 50 nm, in particular less than about 30 nm and particularly preferably less than are about 15 nm. For example, a light source with an FWHM of approximately 10 nm is a narrow-band light source. Often one or more narrow-band emitting light sources are used to determine the extinction and, accordingly, a concentration of a very specific substance that absorbs electromagnetic radiation at a specific wavelength. An example of the use of several such light sources is based on two light sources with each one FWHM of about 10 nm, each of which has maximum emission values or peak wavelengths at about 250 nm and about 280 nm.

The sensor can be selected in accordance with the selected light source and the corresponding emission spectrum, regardless of whether it emits narrow or broadband, in order to be able to detect at least part of the emission spectrum. In other words, it makes sense that the emission spectrum of the light source overlaps with the sensitive area of the sensor, in particular by more than about 50% and preferably by more than about 70%.

Sensors can also have, in particular, sensor elements with different sensitive wavelength ranges, which are designed to be able to detect electromagnetic radiation in wavelength ranges that are not identical to one another.

The light source or light sources, whether emitting narrow or broadband, is or are preferably designed to emit electromagnetic radiation in a wavelength range that is visible to humans, i.e. between about 380 nm and about 780 nm and in the UV range is between about 10 nm and about 380 nm.

Alternatively or additionally, the light source can also be designed to emit electromagnetic radiation in the infrared range, that is to say between approximately 780 nm and approximately 50 mhti.

As already described, the sensor can be matched to the emission range.

In the following, further designs and features are mentioned in other words, these features in particular being combinable with previous features.

An exemplary method for controlling or regulating a process can proceed as follows: 1. Detection of UV-Vis spectra from a concentration series

2. a. Creation of models based on different wavelength ranges, the highest concentration of a sample within a model is determined by the maximum absorbance in the corresponding wavelength range (model 1 with focus on 224 nm only includes samples from 0-2 g / l, model 5 with Focus on 300 nm includes all samples between 0 and 200 g / l b. As an alternative to 2a, various calibration data sets can be created from the concentration series. From these, various models are then created using statistical methods (e.g. PLS). In this case, the Algorithm to redefine the wavelength ranges for each calibration data set (and will thus mainly select wavelengths in the flank of the protein band for models with high protein concentrations (and also not use the band at 224 nm)

3. Online measurement in the process

4. Classification of the online spectrum with the models and selection of the most suitable model. Here algorithms such as PLS-DA or the calculation of the Euclidean or Mahalanobis distance or SIMCA can be used

5. Application of the model to the online spectrum

6. As an alternative to 4 and 5, the required measuring range is clear for each unit operation. With this knowledge, the valid model can also be selected.

At the same time, by recording the complete UV-Vis spectrum, cross-sensitivity between DNA and protein is reduced or avoided and it is possible to record HCPs and target proteins separately. 6 shows the prediction against the reference value for DNA at different protein concentrations. The color of the individual points corresponds to the different protein contents of the sample. The model error is acceptable despite the vastly different protein concentrations. The cross sensitivity is low. In the downstream process, protein and DNA concentrations are usually important parameters that are determined again and again throughout the entire process chain. This determination takes place conventionally mostly offline by measuring the UV-Vis absorbance with the excitation of individual wavelengths (eg with narrow-band emitting LEDs).

Depending on the process step, the protein concentrations differ considerably (0-200 g / l), which is why the samples can usually only be measured conventionally in different dilution stages. Due to the limited dynamic range (area within which essentially no saturation occurs) of the detectors, available online sensors allow e.g. a determination of protein contents up to about 10 g / l. The optical path length must be changed depending on the concentration range in order to be able to expand the detection range if necessary. This means that one and the same hardware cannot be used for all applications. Furthermore, existing UV sensors show considerable cross-sensitivity between DNA and protein. An increase in the DNA concentration therefore also leads to a higher measured protein content.

It should be mentioned, however, that a sensor can also detect concentrations of about 50 g / l in the usual way if the path length is reduced accordingly. It must therefore be decided in advance whether a higher sensitivity and a limited concentration range or a broader concentration range and lower sensitivity is to be achieved. This is then determined in advance by choosing the path length. The method according to the invention, on the other hand, has the advantage that, without changing the hardware, a decision can be made retrospectively as to whether a high sensitivity or a higher concentration range or higher concentrations should generally be measured.

As already mentioned above, there are online UV sensors in conventional systems that work with 1 or 2 individual wavelength ranges (Optek, Pendotech), but which are already saturated at relatively low protein concentrations and therefore have a low dynamic range (see “Classic UV meter maxed out "in FlowVPE.pptx). The wavelength of the absorbance of different proteins can vary. Existing systems adapt the excitation and / or The detection wavelength generally does not match the protein currently to be measured or the excitation wavelength must be defined in advance and cannot be changed ad hoc on the software side depending on the application. Thus, the protein can be measured randomly at the maximum of the absorption wavelength (highest sensitivity), but it will be far more common that the excitation radiation lies in the flank of the protein absorption (lower sensitivity). For the same reason, host cell proteins (HCPs) are usually recorded together with the target protein (cross-sensitivity is close to 100%). With the conventional measurement, only in the case that the absorption maxima and / or the band width or band shape of the HCPs differ from the target protein, the detection of the entire spectrum enables an analytical differentiation.

With regard to this fact, a fiber was alternatively designed to be movable beforehand for the UV transmission measurement, so that the path length can be varied reproducibly and can also be chosen to be much smaller than the minimum path length of, for example, about 1 mm that is usual for UV sensors. In this way, protein concentrations above 100 g / l can also be recorded. However, the distance between the fibers must be adjustable in a highly precise and reproducible manner. Furthermore, due to the complex measuring arrangement, implementation can only be transferred to use in single-use systems under certain circumstances.

The method described herein for controlling or regulating a process is based in particular on the measurement of complete absorption spectra - instead of individual wavelength ranges - preferably using a flow cell with a constant optical path length. A preferred flow cell thickness or sample layer thickness is approximately 1 mm. Depending on the concentration range to be measured, various multivariate models (e.g. PLS with protein as the target variable and the absorption values of the respective spectral range as input variables) are created, which insist on the evaluation of different wavelength ranges.

These wavelength ranges can either be selected yourself or, with all selected wavelength ranges being influenced by the absorbance of the target analyte (DNA or protein). The invention allows a precise determination of concentrations for DNA and / or proteins. In particular, absorption spectra (instead of values at discrete wavelengths) are recorded in a cell at a constant optical path length of, for example, approximately 1 mm. In particular, a data set is compiled into a concentration range and multivariate models are generated with regard to the results within different wavelength ranges. Selected wavelength ranges show different sensitivities to DNA and / or protein. Individual wavelength ranges have different sensitivities. The dynamic range (dynamic ranks) is different.

In FIG. 2, for example, model 1 is assigned to the absorbance at the highest sensitivity, the dynamic range being correspondingly limited. Model 4 is generated with reduced sensitivity, but the concentration range to be used can be expanded. The concentration values are determined by comparing the collected data with predefined models.

This allows a particularly simple determination of protein and DNA concentrations in the downstream process and enables the concentration ranges of DNA and protein, which vary greatly in practice, to be recorded. In addition, the invention makes it possible to avoid the usually required dilution of the samples during the measurement with regard to the limited dynamic range of the sensor. In addition, cross-sensitivity when measuring between DNA and protein can be reduced or avoided.

List of reference symbols

1 optical spectrum

1 _a Optical spectrum of the sample during the process before purification

1b Optical spectrum of the sample during the process after purification

2 absorption value (s)

3 process, for example downstream process

4 UV / vis spectrometers

5 _a Medium or sample in the process before purification 5b Medium or sample in the process after purification

6 Purification unit

7 measuring cell

8 Optical light guide

9 Evaluation Z arithmetic unit

9a desktop or screen

100 Method according to one embodiment

101 Recording of spectra of a concentration series

102 Creation of several models for individual concentration ranges

103 Online measurement: recording a sample spectrum during the process

104 Classification of the sample spectrum in a suitable model

105 Application of the model to the sample spectrum

106 Output of forecast value

107 Forwarding of the forecast value to process control

108 Changed process control based on the prediction l wavelength

Ml-6 models

Claims

1. A method for controlling or regulating a process (3), in particular a downstream bioprocess, based on the prediction of an unknown concentration of one or more substances in a sample (5 _a , 5 b) by means of spectroscopy, in particular UV / vis spectroscopy , comprising the steps:

Acquisition (101) of spectra (1) of a plurality of concentration samples, at least two concentration samples having mutually different concentrations of the one or more substances;

Generation (102) of several quantitative models (M) based on the spectra of the concentration samples, the models (M) each having an assignment of at least one spectral measured variable (2) of the spectra (1) to concentrations in concentration ranges, the concentration ranges of two models ( M) are not identical;

Acquisition (103) of at least one sample spectrum (1; 1 _a , 1b) of the sample (5a, 5b);

Assignment (104) of the sample spectrum (1; 1a, 1b) to at least one quantitative model (M) of the generated quantitative models (M);

Application (105) of the at least one quantitative model (M), which was assigned to the sample spectrum (1; 1 _a , 1b), to the sample spectrum (1; 1 _a , 1 _b ) in order to determine a predictive value for the undetermined concentration; and

Controlling and / or regulating (108) the process (3) with respect to at least one parameter on the basis of the prediction value for the undetermined concentration.

2. The method according to claim 1, wherein the spectra include a one-to-one assignment of the spectral measured variable (2) to several wavelengths (l) and preferably at least a subset of the spectral measured variable (2) that is assigned to the corresponding wavelengths (l), not on the detector side Shows saturation.

3. The method according to claim 1 or 2, wherein the quantitative models (M) comprise at least one global model (M).

4. The method according to claim 1, wherein the quantitative models (M) include hierarchically organized models (M) corresponding to at least a first hierarchy level and a second hierarchy level, wherein in particular the first hierarchy level comprises a global model and the second hierarchy level comprises at least one local model.

5. The method according to any one of the preceding claims, wherein the assignment of the sample spectrum (1; 1 _a , 1b) to at least one quantitative model (M) takes place by means of Euclidean distance and / or Mahalanobis and / or PLS-DA and / or SIMCA.

6. The method according to any one of the preceding claims, wherein the control and / or regulation takes place with respect to the at least one parameter and the parameter comprises at least one of the following parameters: a valve circuit, a pump control, a media flow, a media pressure, a gas pressure, a pH - Value, a filtration step, a fractionation step, a time control, a process time setting, a temperature, an ion concentration, a chromatography step.

7. The method according to any one of the preceding claims, wherein the detection of the at least one sample spectrum (1; 1 _a , 1b) of the sample (5 _a , 5b) takes place in or online and the control and / or regulation preferably also takes place during the process as a direct consequence of the determined predictive value for the indeterminate concentration.

8. The method according to any one of the preceding claims, wherein the generation of quantitative models (M) takes place by means of multivariate model creation, in particular by means of multivariate regression, such as preferably by means of MLR, and / or PLS and / or OPLS.

9. The method according to any one of the preceding claims, wherein at least a partial range, in particular individual wavelengths (l) of the spectra (1) of the plurality of concentration samples are weakened or excluded by weighting.

10. The method according to any one of the preceding claims, wherein at least one sub-range, in particular individual wavelengths (l) of the spectra (1) of the plurality of concentration samples, preferably sub-ranges for which there is a linear dependence between wavelengths (l) and measured values, for the application the sample spectrum (1; 1 _a , 1b) are weighted more heavily, the weighting manually or by means of an algorithm, in particular factor analysis regressions in PLS and / or OPLS.

11. The method according to any one of the preceding claims, wherein the acquisition of a plurality of sample measured values takes place with a single, essentially identical gap width of one or more measuring cells.

12. Device for controlling and / or regulating a process, in particular a downstream bioprocess (3), based on the prediction of an undetermined concentration of at least one substance in a sample (5 _a , 5 b) by means of spectroscopy

At least one interface for obtaining spectra of a plurality of concentration samples of the at least one substance and of at least one sample spectrum (1; 1 _a , 1b) of the sample (5 _a , 5b);

At least one unit for

Generation of quantitative models (M) based on the spectra of the concentration samples,

Assignment of the sample spectrum (1; 1 _a , 1b) to at least one quantitative model (M) of the generated quantitative models (M); and

Application of the at least one quantitative model (M), which was assigned to the sample spectrum (1; 1 _a , 1b), to the sample spectrum (1; 1 _a , 1b) in order to determine a predictive value for the undetermined concentration; At least one process unit for controlling and / or regulating the process with respect to at least one parameter on the basis of the prediction value for the undetermined concentration.

13. The device according to claim 12, further comprising a spectrometer (4) for detecting the spectra of the plurality of concentration samples of the at least one substance and / or the at least one sample spectrum (1; 1 _a , 1b) of the sample (5 _a , 5b).

14. The device according to claim 13 or 14, further comprising a measuring cell, in particular a disposable measuring cell, which is designed to _{receive and guide the sample (5 a} , 5 b) with the substance and on which the at least one sample spectrum (1; 1 _a , 1b) can be recorded.

15. Computer program product, comprising computer-readable instructions which, when the program is executed by a suitable computer-aided system, cause the program to carry out the steps of the method according to one of claims 1 to 11.