Method and apparatus for determining future color values ​​or corresponding characteristics

Abstract

The present disclosure relates to determining a color value (12) or corresponding property of a protein-containing solution (2) or of a protein-containing product (3) prepared from the protein-containing solution (2), which includes applying exciting fluorescent radiation (5) to the solution (2) or product (3), measuring at least one property, preferably a spectrum (6A) or corresponding feature, of the fluorescent radiation (6), and determining a current or future color value (12) or corresponding property of the solution (2) or product (3) based on a correlation between the at least one property of the fluorescent radiation (6) and the color value (12) or corresponding property.

Description

[Technical Field]

[0001] The present invention relates to the field of color classification of protein-containing solutions or protein-containing products prepared from protein-containing solutions. In particular, the present invention relates to a method or device according to the preamble of claim 1. [Background technology]

[0002] The present invention particularly relates to the color classification of protein-containing solutions or products prepared from protein-containing solutions, where these proteins can be recombinant proteins, such as antibodies, monoclonal antibodies, or other therapeutic proteins. Such proteins can be produced in or by protein-producing structures, for example, by prokaryotic or eukaryotic cells, particularly bacteria, fungi, yeast, mammalian cells, or other biological protein-forming structures. The present invention particularly preferably relates to the color classification of protein-containing solutions or products prepared from protein-containing solutions, where the protein is a monoclonal antibody (mAb). However, the present invention can also be applied to different proteins. The protein-containing solution may contain protein-producing structures, such as prokaryotic or eukaryotic cells, to form proteins contained in the protein-containing solution or fragments of the protein-containing solution, such as polysaccharides, nucleic acids, lipids, fats, membrane fragments, low molecular weight metabolites, other host cell proteins (HCPs), at least in one stage of the production process, but these molecules, structures, or cells may be removed from the protein-containing solution prior to applying the present invention.

[0003] Coloration of the protein-containing solution or product can occur during the production of the protein, solution, or product. While the exact reason for the color and its intensity can vary and often cannot be precisely determined, it has been found that the intensity of the color corresponds to the usefulness of the protein, the protein-containing solution, or the protein-containing product prepared from the protein-containing solution. The higher the color intensity, i.e., the stronger the color, the more likely the protein, solution, or product is not suitable or will be unsuitable for the desired application, for example, to be used as a therapeutic agent. Testing of drug substances and medicinal products for the degree of color development is a pharmacopoeial requirement according to Ph. Eur. 8.0 Monograph 2031, 01 / 2012, pp. 753 to 755: "Monoclonal antibodies for human use", hereafter abbreviated as Ph. Eur. Ph. Eur. requires that monoclonal antibody products be colorless or slightly colored, thus making colored solutions unsuitable for the intended purpose. Experimental studies have shown that recombinant monoclonal antibodies (mAbs), such as those produced by Chinese hamster ovary (CHO) cells, often exhibit a yellow or yellowish-brown color. While this may at first glance seem like a minor issue, it should be noted that the color of a therapeutic drug product is a notable quality attribute with regulatory expectations, as, for example, Ph. Eur. Monograph 2031 requires that the product be colorless or slightly colored, and / or that the solution be rejected if the degree of coloration exceeds certain limits.

[0004] The color of a formulation, such as a solution or product, can be formed, for example, by tryptophan oxidation, glycation, and Maillard reactions, as well as the presence of vitamin B12. However, the molecular explanation of these effects is still under debate, which indicates that sufficient process control by removing the corresponding chemical species is problematic. Furthermore, it should be pointed out that the magnitude of the color increase with the concentration of proteins, especially mAbs, thereby limiting the effective amount of the active pharmaceutical ingredient in the final protein-containing product (drug product). More specifically, Ph.Eur. has introduced, inter alia, the so-called yellow (Y) or brown-yellow (BY) scale, which includes subcategories from 1 to 7, with 7 indicating a colorless solution and 1 relating to the maximum expected color intensity, thus making it possible to determine the degree of coloration of a liquid.

[0005] According to standard operating procedures, solutions are often manually compared to various reference solutions by visual inspection and thus classified as being part of a yellow (Y) or brown-yellow (BY) color series, and within the same procedure, the degree of coloration of the solution is also determined, as in certain instances. Therefore, this careful experimental standard procedure for classifying color values ​​by human inspection is a time-consuming and exhausting approach. To overcome these drawbacks, there is an urgent need for more robust and reliable mapping schemes, as well as methods that are sensitive and capable of high throughput. During the so-called downstream processes for purifying and / or concentrating the produced protein in solution and / or for producing a protein-containing product, the color cannot usually be completely removed. In particular, the color is chemically part of the protein, bound to the protein, or is otherwise inseparable from the protein. Therefore, it often happens that as early as the last steps of the production process, it becomes clear that the color ultimately exceeds or will exceed the limit, making the proceeding protein-containing solution or the entire batch of product formed from the protein-containing solution unusable, at least for the desired purpose, for example, as a therapeutic agent at the desired concentration.

[0006] Currently, for classification of protein-containing solutions or products formed from protein-containing solutions, it is unavoidable to prepare color standard solutions based on rule-defined recipes and to have trained experts compare the coloration of the protein-containing solution or product with the color standard solutions to assign a color value, because automatic methods for directly measuring and evaluating color, i.e., absorbance analysis, have proven unreliable. Furthermore, color classification by experts is not sensitive enough to determine whether a particular process step affects the degree of coloration of the protein-containing solution or product during the early stages of production, when the color is still very light. Natarajan Vijayasankaran et al.: "Effect of Cell Culture Medium Components on Color of Formulated Monoclonal Antibody Drug Substance," Biotechnol Prog, vol. 29, no. 5, 11 June 2013 (2013-06-11), pages 1270-1277, relates to color measurement, for example, by normalized intrinsic fluorescence intensity (NIFTY) analysis. Because color and fluorescence were observed to be correlated, the fluorescence of the antibody molecule was used as a color surrogate for NIFTY measurements. For each antibody sample, normalized fluorescence was determined by dividing the fluorescence peak area of ​​the main peak by the UV absorbance peak area of ​​the main peak, normalizing the fluorescence response for the contribution of the antibody mass. The NIFTY value was then determined by calculating the ratio of the normalized fluorescence of the sample to the normalized fluorescence of a monoclonal antibody reference sample.

[0007] US Patent Application Publication No. 2013 / 281355(A1) relates to a similar subject, namely, the use of NIFTY and the determination of a color intensity value by calculating the ratio of the normalized fluorescence of a test monoclonal antibody sample to the normalized fluorescence of a reference monoclonal antibody sample. Furthermore, it is disclosed that the NIFTY value is measured from the main peak of a size exclusion chromatogram, and this NIFTY value can be expected to remain constant throughout the purification process, preferably if colored or colorless protein molecules are not purified. However, during the complete process of protein production, it is desirable to periodically collect and purify the protein to reduce coloration. Therefore, as a result, the NIFTY value was found to be suitable as a proxy for coloration only. Summary of the Invention

[0008] It is therefore an object of the present invention to provide a method or apparatus for determining a color value or a corresponding property in order to gain control over the color of a protein-containing solution or a product formed from the protein-containing solution. This object may be achieved by a method according to claim 1 or by an apparatus according to claim 15. Advantageous embodiments are the subject of the dependent claims. According to one aspect of the invention, the method includes exciting a fluorescent radiation of the solution or product, measuring at least one property, preferably a spectrum or corresponding characteristic, of the fluorescent radiation, and determining a current or future color value or corresponding characteristic of the solution or product based on a correlation between the at least one property of the fluorescent radiation and the color value or corresponding characteristic. That is, the current or future color value of the solution or product, or if the property corresponds to the color value, the current or future property of the solution or product, is determined. Surprisingly, it has been found that this determination can be achieved by examining the fluorescence radiation of the solution or product.

[0009] When determining the color value or corresponding characteristic, a correlation between at least one characteristic of the fluorescent radiation and the color value or corresponding characteristic is used directly or indirectly. In particular, the spectrum of the measured fluorescent radiation or one or more features of this spectrum are used to determine a color value or corresponding characteristic. This determination may be achieved by directly or indirectly correlating the spectrum of the measured fluorescent radiation with a reference fluorescent radiation spectrum, or correlating one or more features of this spectrum with one or more features of the reference fluorescent radiation spectrum. Color values ​​or corresponding properties related to or corresponding to the reference fluorescent radiation spectrum with the greatest correlation, or related to or corresponding to one or more features of the reference fluorescent radiation spectrum with the greatest correlation, may then be assigned to the measured fluorescent radiation, to one or more features of this spectrum, and to the solution or product from which the measured fluorescent radiation originates. This correlation can be performed directly, i.e., by comparison of the measured spectrum with one or more reference spectra, which contain corresponding color values ​​or corresponding characteristics. The color value or corresponding characteristic of the reference spectrum that has the greatest correlation or that meets each selection criterion for correlation can then be the decision result.

[0010] However, it is particularly preferred that the correlation is performed indirectly by means of a means or tool that takes the correlation into account, for example a regression method in which regression parameters are or have been determined based on the correlation, or an artificial neural network, or another supervised machine learning method that is or has been trained based on the correlation. Particularly preferably, (advanced) multifactorial supervised regression methods can be used, which can preferably be trained to exploit the correlation. Thus, the means or tool is configured to determine the color value or the corresponding property based on (information about) the correlation between, on the one hand, the color value or the corresponding property and, on the other hand, the spectrum of the fluorescence radiation or one or more features of this spectrum. In the sense of the present invention, determining preferably means or aims directly or indirectly at using information about the correlation between at least one property of the measured fluorescent radiation, on the one hand, and a color value or a corresponding property, on the other hand, to assign a color value or a corresponding property to the measured fluorescent radiation, to at least one property of the measured fluorescent radiation, to the solution from which the measured fluorescent radiation originates and / or to the product from which the measured fluorescent radiation originates. In practice, the at least one property of the measured fluorescent radiation can be input into a tool, in particular a software tool or a computer program product, or means for determining a color value or a corresponding property can be applied accordingly.

[0011] It has been found that protein-containing solutions or products formed from protein-containing solutions exhibit fluorescence when they have or tend to have a yellowish or yellow-brownish coloration. In particular, when excited at essentially UV wavelengths, wavelengths of colors essentially complementary to a current or expected yellowish (Y) or brownish-yellowish (BY) color, and / or wavelengths that cause yellow coloration, the resulting fluorescence has surprisingly been found to exhibit significant correlations such that a current or future color, particularly a yellow / yellowish-brownish color and / or intensity / spectrum, can be determined / predicted. It has been found that one or more properties or characteristics of the aforementioned fluorescence, in particular the emission intensity and / or the wavelength of maximum intensity, one or more pairs of intensity and wavelength, and / or other properties of the fluorescence spectrum, are correlated with the current or future degree / intensity of color and / or the tendency of the protein-producing structures involved in the color generation of the protein-containing solution or protein-containing product, preferably taking into account the protein concentration of the protein-containing solution or protein-containing product and / or the production processing step of the protein production process of the protein-containing solution or protein-containing product and / or the pH value of the protein-containing solution or protein-containing product.

[0012] Therefore, the present invention preferably uses correlations between two or more of the following: - wavelength / magnitude pair of excited fluorescence of protein-containing solution or product - one or more concentrations of proteins, preferably monoclonal antibodies (mAbs), in the solution and / or product, in particular as measured and / or as expected (e.g. in the product), and - color values ​​(BY, Y), in particular color values ​​defined by GMP (Good Manufacturing Practice), for example color values ​​determined as described in the European Pharmacopoeia. According to the invention, the above correlations or further correlations are used directly or indirectly to (automatically) determine current or future color values ​​or corresponding characteristics based on the measured excitation fluorescent radiation. It has surprisingly been found that the fluorescence radiation or characteristics of the fluorescence radiation provide high sensitivity and linearity of correlation with color over a wide range of protein concentrations in the solution or product and / or across production process steps. Thus, the present invention allows for the assessment and / or prediction of color of protein-containing solutions or products even when they are not (yet) visible to the human eye or are only visible to a degree that does not allow assessment using conventional techniques.

[0013] The present invention particularly takes advantage of high-throughput fluorescence spectral measurements to reap the benefits resulting from automation capabilities. The present invention can be applied in a variety of technical contexts in a synergistic manner. By measuring fluorescence and by correlating fluorescence radiation, the present invention generally enables high-throughput, automatable color sorting. Due to automation, for example, chemical species can be efficiently selected or the progress of a protein-containing solution or product production process can be closely monitored and / or adapted by frequent color checks. This can facilitate preventing excessive color intensity. The sensitivity, robustness and wide concentration range that can be determined in a synergistic manner due to the high-throughput automation capabilities facilitate access to a wide range of applications, leading to recombinant protein production methods and allowing antibody coloration to be tracked and ultimately controlled, resulting in improved product quality according to the invention. For this and other purposes, the method of the present invention can be advantageously applied to automated measurement procedures using microtiter plates. Thus, due to the high sensitivity provided, minimal sample volumes of protein-containing solutions or products are sufficient. This allows for rigorous monitoring and efficient selection of protein forming structures / chemical species.

[0014] In particular, the proposed method can be successfully applied to measuring samples of protein-containing solutions in microtiter plates, e.g., in a 96-well plate format, thereby paving the way for high-throughput automation to determine the current or future yellow / yellowish-brownish color of protein-containing solutions or products. The present invention can thus enable automatic identification and / or prediction of color. Preferably, the present invention enables automated prediction of color values ​​for drug substances, drug products, and diluents / placebos. However, automation capabilities can be advantageous in a variety of contexts. The present invention can enable color identification or prediction very early in the production process. Thus, the present invention can enable identification of process parameters and conditions that affect the degree of coloration, and can facilitate altering the process and conditions that affect the degree of coloration, or stopping and canceling a batch to prevent effort in wasted cases. As a result, the present invention can result in certain resource savings, an efficient and effective process for producing proteins in protein-containing solutions, and / or an efficient and effective process for producing protein-containing products.

[0015] Alternatively or additionally, the present invention may allow the selection and / or cloning of protein-forming structures, such as (eukaryotic) cells, that produce the desired protein and tend to produce no or minimal color. Using the method according to the present invention, multiple samples of solution can be tested and one or more specific samples that tend to produce little or no color can be selected. The one or more specific samples can form the basis for propagation or cloning to produce the protein. This sample selection may be performed as an alternative or in addition to determining process parameters that allow for the production of less color. When a protein-containing solution is treated in a downstream process to remove proteins contained therein, the present invention is particularly advantageously applied after an upstream process for producing proteins in the protein-containing solution. During the downstream process, color is removed by various means, if possible. However, due to the weak coloring at low protein concentrations, the extent to which color is removed in early downstream steps has not been evaluated in the past. However, even when only weak coloring is present, the present invention facilitates color determination and / or prediction due to its sensitivity (value or corresponding characteristic). Therefore, it is advantageous to control or appropriately change, replace, or eliminate parameters of the downstream process and process steps in the downstream process.

[0016] The present invention can be applied upstream, downstream, or both. Alternatively or additionally, the present invention can be applied to formulation development and / or product design. Again, the unique high efficiency has been demonstrated in the context of processes for producing recombinant proteins, preferably antibodies, particularly monoclonal antibodies (mAbs). However, the present invention can equally be applied in various protein production processes, particularly those prone to coloration, such as Y or BY coloration. In particular, the prediction of the color value or corresponding property of a solution or product for increasing the concentration of a protein, particularly a mAb, is beneficial in view of the fact that the final protein-containing product (drug product) contains a high protein concentration, particularly a high concentration of a mAb. In particular, the protein concentration increases significantly later in the process, significantly affecting the color of the solution. Therefore, the prediction of the color value at an early stage of the process, as enabled by the present invention, reduces the amount of troubleshooting events at later stages of the process and provides reasonable estimates for control strategies and improvements related to lead optimization. As an additional benefit, the robust classification of color values / scales for the computational method allows for a more rigorous review process by regulatory agencies. Finally, the proposed method can be easily implemented in an automated laboratory environment.

[0017] Surprisingly, it was found that the fluorescence spectra of solutions or products, even when seemingly similar, show subtle differences between different proteins, especially between mAbs, indicating the presence of different chemical species that can affect the ratio of fluorescence and coloration intensity. One aspect of the present invention relates in particular to the use of high-dimensional and / or numerical regression methods, such as machine learning techniques. The present invention preferably utilizes artificial neural network (ANN) techniques for mapping fluorescence spectra to color values. Artificial neural networks are also abbreviated as neural networks in the following. In this regard, the present invention utilizes the correlation between the integrated fluorescence spectrum and the resulting color values. As an extension or alternative to the (preferably one-dimensional) correlation, a multifaceted evaluation of the fluorescence magnitude / wavelength pair can be used, preferably in combination with the color values. Such multivariate numerical methods significantly improve accuracy and make the method applicable to all solution conditions. With respect to classification and prediction of color values, particularly with respect to determining parameters of regression / ANN, the present invention preferably facilitates that measured fluorescence spectra are used as input values, while the corresponding BY or Y values ​​are considered as corresponding output or target values, which is a significant improvement over previous approaches that rely solely on low-level classification of color from fluorescence intensity.

[0018] For example, in experiments, a high correlation coefficient of R2 = 0.94 was obtained with corresponding Y and BY color scales, thereby improving formulation and measurement times, including individual monoclonal antibodies (mAbs), diluents. It has been found that the present invention is advantageously improved by correlation using (feed-forward) neural networks and / or machine learning, where color may alternatively or additionally be determined and / or predicted by numerical regression and / or correlation techniques. In addition to more reliable classification, automated machine learning methods may be provided that allow for high-throughput characterization of colored formulations, saving development time. In previous feasibility studies, a feedforward neural network using, for example, one hidden layer containing 48 hidden nodes and a rectified linear unit activation function provided good results for predicting and classifying color values ​​of drug substances, drug products, and buffer solutions.

[0019] The present invention has proven to be particularly advantageous in the context of biopharmaceuticals when machine learning or even numerical regression (classification techniques used directly or indirectly for prediction) and classification of color values ​​are applied. For predictive capabilities, the corresponding ANN is preferably trained on multivariate fluorescence intensity / protein (especially mAb), concentration pairs and resulting color values, as well as preferably information about the corresponding protein production process and / or protein enrichment steps. Preferably, a non-linear mapping function between fluorescence intensity, protein, particularly mAb, concentration, and color value is obtained / used. As a prerequisite for the ANN process, a series of low concentrated protein solutions, particularly mAb solutions, can be prepared and the fluorescence spectrum of each measured. Regarding the surprisingly found (protein-specific) linear relationship between fluorescence intensity and protein, especially mAb, concentration, which proved to be valid for all relevant drug concentrations, the resulting fluorescence intensity for desired high protein, especially mAb, concentrations can be calculated by extrapolation.

[0020] The calculated fluorescence intensity and the concentration of the desired protein, particularly the mAb, can then be used as inputs to a trained ANN that predicts the resulting non-trivial color value. In summary, a machine learning based approach for reliable prediction and classification of color values ​​is presented. The proposed approach can be applied to improve the control of protein-containing solutions or product production processes. In combination with high-throughput fluorescence analysis, this method paves the way for new control strategies for colored solutions. Additionally, this method can be useful for detecting impurities and outliers, providing access to improved analysis of protein concentrations, especially those of mAbs, in unknown solutions. A further aspect of the present invention relates to an apparatus including a fluorescence spectrometer for measuring fluorescence radiation. The fluorescence spectrometer comprises a light source for emitting fluorescence excitation radiation and a photodetector for measuring the fluorescence radiation, in particular for measuring the spectral intensity / spectral power of the fluorescence radiation with respect to wavelength or frequency. Furthermore, the apparatus includes a device adapted to perform a method for determining a color value or a corresponding characteristic based on the measured fluorescence radiation. The described features and advantages may apply accordingly.

[0021] Color, in the sense of the present invention, is preferably a visual property that is represented by a color category, for example yellow or brown-yellow. The perception of color results from the stimulation of human photoreceptor cells (in particular the cone cells of the human eye and the eyes of other vertebrates) by electromagnetic radiation (in this case within the visible spectrum) in the sense of the present invention. The color category and the physical specifications of the color correspond to the wavelength of the reflected light and its intensity. The color preferably corresponds to electromagnetic radiation of wavelengths specific to the stimulation of human photoreceptor cells, which causes a particular color impression. In particular, the color impression or the spectrum that gives rise to this impression is controlled by specific light absorption properties. In particular, the colour and colour intensity in the sense of the present invention are specified in regulations such as the European Pharmacopoeia (Ph. Eur.) 8th Edition or newer section 2.2.2, which provides reference formulations for comparison of colour and intensity. Thus, a color and color intensity is, in the sense of the present invention, a particular color as defined in the Regulations if stimulation of human photoreceptor cells (in particular cone cells in the human eye and the eyes of other vertebrates) by electromagnetic radiation evokes the same neural response as stimulation of human photoreceptor cells (in particular cone cells in the human eye and the eyes of other vertebrates) by electromagnetic radiation resulting from a reference preparation illuminated with a light spectrum that is essentially continuous in the visible range or by electromagnetic radiation as defined in the Regulations.

[0022] A color value, in the sense of the present invention, is a specific color identifier that can be defined in a rule, and the color of an object can be determined by comparison with a reference color. A color identifier can identify a color, an intensity, or both. It is understood that a color value can be a numerical value or a specific wavelength, but this is not necessary. Therefore, a color value, in the sense of the present invention, is preferably understood broadly and covers various information for specifying a color, an intensity, or both. The color or color value preferably corresponds to a property of the solution, product, or protein. The property corresponding to the color value is therefore in particular the suitability of the solution, product, or protein to be used for its intended purpose. This property can in particular be an indicator of a drug's effect or side effects. This property preferably corresponds to a color value if the color value influences the property or if the property depends on the color or color value, for example due to regulatory requirements, because the color is an indicator of the suitability of the application, or because the color has a direct or indirect drug effect.

[0023] A color classification, in the sense of the present invention, preferably designates colors to be included in a class, i.e., a category or range of colors and / or color intensities, or a range of colors, such that similar colors or color intensity ranges are assigned to a color class, or a color class is determined by a color classification process. A color can be represented by a color value, or vice versa. Therefore, the terms color class and color value are used interchangeably in the present invention, and when the term color class is used, the term color class can be replaced with the term color value, and the term color value can be replaced with the term color class. Nevertheless, a color value can define a specific color or a range of colors, but a specific color is a range of colors with an infinitesimal range. Nevertheless, a color class preferably targets distinguishable colors and / or intensities. For the sake of brevity, a protein-containing solution, also referred to simply as a solution herein, is a liquid containing proteins in the sense of the present invention. Preferably, the protein in the protein-containing solution is dissolved in a solution such as water, but the term protein-containing solution may also refer to a suspension in which the protein is suspended, for example, in water. Therefore, the term protein-containing solution, although explicitly stated, may not refer to a suspension, but is preferably understood broadly. A protein-containing solution is a liquid used in a protein production process and may at least temporarily contain protein-producing structures. Thus, this solution may at least temporarily be a culture for producing proteins, in particular a culture broth. However, the solution during the production process may contain proteins without the protein-forming structures, which may be removed, for example, in downstream processes, to wash and / or concentrate the proteins.

[0024] A protein-containing product, which for the sake of brevity is referred to herein simply as a product, is, in the sense of the present invention, a product comprising a solution containing the protein, but from which other components of the solution have preferably been removed, so that the product is preferably suitable for pharmaceutical application for administration or is otherwise preferably in a form for direct use, in particular for example injection for therapeutic purposes. Fluorescence radiation, in the sense of the present invention, is radiation resulting from the effect of fluorescence of a substance, such as a solution or product containing a fluorescent component. Fluorescence, in the sense of the present invention, preferably means the emission of light (fluorescence emission) by a substance that has absorbed light that causes luminescence (fluorescence excitation radiation) or other electromagnetic radiation, the emitted light preferably having a longer wavelength and therefore lower energy than the absorbed radiation. For example, fluorescence, in the sense of the present invention, is the case when radiation is absorbed in the ultraviolet region of the electromagnetic spectrum, while the emitted light is in the visible region, which can give the fluorescent substance an individual color that can only be seen or measured when exposed to UV light.

[0025] The excitation radiation according to the present invention is preferably light that is directed at and absorbed by a substance that has the ability to generate fluorescence radiation using the energy of the absorbed excitation radiation. When fluorescence is excited at a particular wavelength, electromagnetic radiation of this wavelength is applied to the substance to cause fluorescence and thus excite the substance to generate fluorescence radiation. Thus, fluorescent radiation is radiation produced by a fluorescent material when it is exposed to excitation radiation, and as used herein, the fluorescent material is a solution or product. Wavelengths in the sense of the present invention are preferably wavelengths of electromagnetic radiation / waves (of intensity maximum) in the optical range. Alternatively or additionally, the term "wavelength" can be used as a substitute for or replaced by the corresponding frequency. Intensity, in the sense of the present invention, is preferably a metric for measuring or expressing the spectral power of light, preferably at each wavelength. Although the term "intensity" is primarily used below, intensity corresponds to and can be replaced by the terms power (of electromagnetic waves) and magnitude (of electromagnetic radiation / at wavelength).

[0026] It is particularly preferred that the property or characteristic of the fluorescence radiation is one or more of a spectrum or spectral features, such as intensity and / or wavelength and / or intensity-wavelength pair, intensity maximum, wavelength and / or intensity of intensity maximum or side maximum, characteristics such as the presence, shape and / or intensity of shoulders in addition to the spectral maximum, the spectral integral of the fluorescence radiation, i.e., the area under the curve, or other related or inferred property of the fluorescence radiation.

[0027] A wavelength maximum or spectral maximum, in the sense of the present invention, is preferably the maximum intensity or power at a specific wavelength. A maximum is preferably the absolute maximum intensity or power in a spectrum or a section of a spectrum. A maximum is at least a local intensity / power maximum, although one or more other maxima may be included in the spectrum. However, maxima are preferably peaks with an intensity / power significantly above the noise level and / or above a threshold, so that merely eight, preferably five, and in particular four or fewer peaks in a spectrum are considered maxima. Therefore, in particular, variations close to the noise level are not considered maxima, but peaks with a power / intensity multiple of the noise level are considered maxima. Preferably, merely absolute maxima within a partial range of the spectrum are considered maxima at a specific wavelength, preferably the infrared (IR), in particular the near-infrared (NIR), the visible wavelength range (VIS), or a partial range referring to a specific color visible to the human eye, and / or the ultraviolet range (UV), or alternatively the UVA, UVB, and / or UVC partial ranges. Further aspects of the invention emerge from the claims and the following description of preferred embodiments with reference to the drawings. [Brief explanation of the drawings]

[0028] [Figure 1] 1 shows a schematic apparatus according to the present invention; [Figure 2] FIG. 1 illustrates a simplified example of an artificial neural network. [Figure 3]FIG. 1 shows a schematic flow chart for predicting color values. [Figure 4] FIG. 1 shows a schematic flow chart of a downstream process. [Figure 5] FIG. 1 shows a microtiter plate. [Figure 6A] FIG. 1 shows a chart of intensity versus wavelength. [Figure 6B] FIG. 1 shows a chart of intensity versus wavelength. [Figure 7] FIG. 1 shows a chart of area under the curve versus concentration. [Figure 8] FIG. 1 shows a chart of fluorescence versus concentration. [Figure 9] FIG. 10 is a chart showing a plot of the goodness of fit between predicted and measured BY values ​​for the ET model. [Figure 10] FIG. 10 is a chart showing a plot of goodness of fit between predicted and measured BY values ​​for an ANN model. [Figure 11] FIG. 10 shows the scaling factor pX and the standard deviation of the scaling factor σ(pX) using the values ​​in Table 1 for the calculation of integrated fluorescence intensity. [Figure 12] Figure 1 shows the measured and predicted fluorescence intensities with prediction error for selected mAb1 at different process steps X and at specific concentrations. The connecting solid grey lines are merely guides for ease of viewing. [Figure 13] Figure 10 shows a chart plotting the goodness of fit of predicted versus measured BY values ​​for the RF model, with the black line indicating perfect agreement with a slope of 1. [Figure 14] Figure 10 shows a chart plotting the goodness of fit of predicted versus measured BY values ​​for an ANN model, with the black line indicating perfect agreement with a slope of 1. [Figure 15]FIG. 1 shows a chart showing measured BY values ​​for a particular mAb1 at a particular process step and concentration, along with predicted BY values ​​from the ANN and ET models; the solid and dashed lines are merely guides for ease of viewing. [Figure 16] FIG. 1 shows a chart showing measured BY values ​​for a particular mAb2 at a particular process step and concentration, along with predicted BY values ​​from the ANN and RF models; the solid and dashed lines are merely guides for ease of viewing. DETAILED DESCRIPTION OF THE INVENTION

[0029] In the following description of preferred embodiments, the same reference signs are used for the same or similar parts, and the same or similar effects and advantages may be achieved even when repeated descriptions are avoided. FIG. 1 shows a schematic apparatus 1 according to the present invention for determining the color value 12 or a corresponding property of a protein-containing solution 2, hereinafter abbreviated as "solution 2," or a protein-containing product 3 prepared from the protein-containing solution 2, hereinafter abbreviated as "product 3." In the example shown in Figure 1, solution 2 is contained in bioreactor 2A, and the arrow indicates that product 3 may be produced, optionally using intermediate further processing steps, and product 3 is symbolized by vial 3A. Either Solution 2 or Product 3 may be tested directly as proposed or by collecting a sample and testing a sample of Solution 2 or Product 3, for example, in sample chamber 2B.

[0030] The protein of Solution 2 may be produced by a protein production structure, in particular a cell culture, which may include, for example, eukaryotic cells. Preferably, the protein production structure produces a recombinant protein, in particular an antibody, such as a monoclonal antibody (mAb). However, the present invention may also be applied to a different Solution 2. The protein-containing product 3 is preferably suitable for administration, symbolized in Figure 1 by a vial 3A for administration by injection, e.g., via an injection device not shown, such as a syringe. Of course, the use of a vial 3A is not essential. Product 3 and Solution 2 are preferably liquid (at room temperature), however, it is not essential that Product 3 be in liquid form. The device 1 comprises a light source 4 for generating fluorescence excitation radiation 5 that can be applied to a solution 2 or a product 3 . Fluorescence radiation 6 emitted by the solution 2 or the product 3 can be received by a spectrometer 7 of the device 1 .

[0031] The pictograms assigned to the fluorescence excitation radiation 5 and the fluorescence radiation 6 respectively show schematic illustrations of exemplary spectra 5A, 6A in a diagram of light intensity P (also called power or magnitude) against wavelength λ. As can be seen from the pictogram of spectrum 5A of fluorescence excitation radiation 5, this radiation 5 preferably has a maximum value within a range of wavelengths λ that is smaller than the range of wavelengths λ within which the maximum value of fluorescence radiation 6 occurs, as shown in the pictogram of spectrum 6A, which shows an example of fluorescence radiation 6 caused by fluorescence of solution 2 or product 3. Preferably, the solution 2 or product 3 during protein production develops a yellowish (Y) or yellow-brownish (BY) coloration. In the first step below, a typical manual approach for color classification as shown in Figure 1 is described to facilitate the explanation of the differences that the present invention allows and to show how a reference measurement can be performed.

[0032] The color can be inspected manually by an expert eye 8, where an at least essentially uniform spectrum light source 9 illuminates the solution 2 or product 3 with light 10 (see light spectrum FIG. 10A). This spectrum 10A is partially absorbed by the solution 2 or product 3, causing different parts of the spectrum 10A to be reflected, transmitted or scattered as light 10, which typically has a yellowish or brownish-yellowish color. The yellowish or brownish-yellowish color preferably corresponds to electromagnetic waves having a maximum wavelength λ in the range greater than 560 nm and / or less than 620 nm. The corresponding pictogram of spectrum 11A of reflected light 11 shows a maximum in this range, although it should be understood that spectrum 11A may vary while still essentially corresponding to a yellow or brownish-yellow color. In the manual approach, the yellowish or brownish-yellowish color of Solution 2 or Product 3 is compared to the color of a reference solution having a yellowish or brownish-yellowish color to assign a color value 12 to Solution 2 or Product 3 by manual classification.

[0033] According to the present invention, a different approach is followed since the fluorescence excitation radiation 5 can be merely a single wavelength maximum or at least a non-continuous spectrum. The fluorescence radiation 6 caused by the fluorescence of the solution 2 or product 3, measured by the spectrometer 7, is usually not yellow or brownish-yellowish, or at least does not have to be yellow or brownish-yellowish. Nevertheless, it has surprisingly been found that this fluorescence radiation 6 correlates with the yellowish or brownish-yellowish color of the solution 2 or product 3. This correlation is therefore used to replace the manual classification process of assigning a color value 12, although neither the fluorescence excitation radiation 5 nor the fluorescence radiation 6 need be yellow or brownish-yellow. According to the present invention, the fluorescence radiation 6 of the solution 2 or product 3 excites at least one property, preferably spectrum 6A or a corresponding feature, of the fluorescence radiation 6, which is measured, and based on the correlation between the at least one property of the fluorescence radiation 6 and the color value 12 or a corresponding property, the current or future color value 12 or a corresponding property of the solution 2 or product 3 is determined. The current or future color value 12 or corresponding property of the solution 2 or product 3 is determined using the surprisingly discovered correlation. While it is possible to use the spectrum 6A of the fluorescence radiation 6, which is a property of or represents the fluorescence radiation 6a, it is preferable to use one or more features corresponding to the spectrum 6A. Such features can be, for example, an intensity P vs. wavelength λ pair (a point on the curve of the spectrum 6A / a pair of values ​​of the spectrum 6A), or a specific shape of the spectrum 6A, the intensity P of a maximum, the wavelength λ of a maximum, or any pair thereof. All of these are features corresponding to the spectrum 6A, and individual features or combinations of features can be used to determine the color value 12 or corresponding property of the solution 2 or product 3.

[0034] The relevance of the color value 12 has already been discussed. However, based on the present invention, it can be concluded that the color value 12 can be replaced by any indicator or other color-related label that indicates the suitability of the application of the solution 2 or the product 3. On the one hand, such an indicator is the color value 12 and vice versa, and on the other hand, such an indicator is in any case determinable according to the present invention and is covered by the properties corresponding to the color value 12. For clarity, the illuminating light source 9 and eye 8 are shown in FIG. 1 merely to illustrate the different approaches, but preferably neither the continuous light source 9 nor the eye 8 form part of the device 1 . Preferably, the fluorescence excitation radiation 5 has a maximum intensity at a wavelength λ greater than 310 nm and / or less than 540 nm. More preferably, the fluorescence excitation radiation 5 has a wavelength λ greater than 360 nm and / or less than 420 nm. It is particularly preferred that the wavelength λ of the fluorescence excitation radiation 5 is greater than 380 nm and / or less than 400 nm.

[0035] It has been found that fluorescence excitation radiation 5 at these wavelengths λ is particularly suitable for examining solution 2 or product 3 with respect to yellowish or brownish-yellowish colorations that may develop during the process of using solution 2 to produce product 3. Alternatively or additionally, the fluorescence radiation 6 is preferably produced by the solution 2 or the product 3 and / or detected within a wavelength λ range of 330 nm to 800 nm, i.e., on that basis, the spectrum 6A of the fluorescence radiation 6 typically contains information suitable for subsequent classification at least in the wavelength range of 330 nm to 800 nm, and therefore the spectrometer 7 can, and is preferably configured to, cover at least this wavelength λ range. The fluorescence excitation radiation 6 preferably has an excitation maximum intensity at a wavelength greater than 330 nm and / or less than 800 nm, which corresponds to the minimum detection range of the spectrometer 7, but it is even more preferred that the fluorescence radiation 6 has a maximum or absolute maximum wavelength λ greater than 420 nm and / or less than 600 nm, in particular greater than 450 nm and less than 530 nm. The intensity maximum of the fluorescence radiation 6 is preferably at a wavelength λ that is more than 50, 60, or 70 nm beyond, and / or less than 130, 120, or 110 nm beyond, the wavelength λ at which the fluorescence excitation radiation 5 has its maximum intensity, as shown schematically in spectra 5A, 6A in the pictograms of Figure 1 and will be explained in more detail below.

[0036] According to the present invention, the current or future color value 12 or corresponding property of the solution 2 or product 3, such as a color value from a regulatory specification, can be predicted directly or indirectly based on the properties of the excitation fluorescence radiation 6, which are particularly accurate and reliable when the above-mentioned wavelength λ range is met, individually and especially in a synergistic way when combined. Determining the current or future color value 12 or corresponding property of the solution 2 or product 3 preferably involves direct or indirect prediction based on the properties of the exciting fluorescent radiation 6 using correlations as described. The aforementioned correlation of the property of the exciting fluorescent radiation 6 and the current or future color value 12 or corresponding property is preferably determined or is determined, for example, by manual classification of a sample of the solution 2 or product 3. The resulting pair of reference properties of the exciting fluorescent radiation 6 on the one hand and the current or future color value 12 or corresponding property on the other hand can be used for a direct correlation or, particularly preferably, for developing tools or means that indirectly represent the correlation, allowing a prediction of the current or future color value 12 or corresponding property based on the property of the exciting fluorescent radiation 6. Both of these are described below, starting with the direct approach, followed by advantageous advanced approaches and particularly preferred properties or characteristics used.

[0037] To represent the correlation, a reference, preferably a reference fluorescence radiation spectrum 14 or corresponding information, is assigned to the color values ​​12 or corresponding characteristics. The pair of reference fluorescence radiation spectrum 14 and assigned color values ​​12 or corresponding characteristics is also referred to as a pre-sorted reference spectrum 14. The found correlation is preferably represented by the pre-sorted reference spectrum 14. The properties of the reference fluorescence spectra 14 resulting from the reference color samples may be characterized by a reference spectrometer 15. Corresponding color values ​​12 or corresponding properties are assigned to the reference fluorescence spectra 14, respectively. This assignment may be achieved by manual inspection and entered using an input device 16 for assigning color values ​​12 to properties of the reference spectra 14. Optionally, a reference database 17 may be provided that stores pairs of reference spectra 14 and color values ​​12. The determination of the color value 12 or corresponding property is preferably performed by a correlation device 18. The correlation device 18 is preferably able to use the correlation found, directly or indirectly, to examine a property such as the measured fluorescence radiation 6, i.e., spectrum 6A, or one or more features thereof, to determine, derive, or predict a color value 12 or corresponding property that can be assigned to the solution 2 or product 3.

[0038] In a direct approach, the reference fluorescence spectra 14 can be checked for correlation with the measured fluorescence spectrum 6A, and one of the reference fluorescence spectra 14 with the best correlation can be selected to assign the color value 12 of the best-correlated reference spectrum 14 to the solution 2 or product 3, which can then be output by the output device 13 as required. That is, the determined color value 12 or corresponding characteristic may include or be formed by a previously obtained color value 12 or corresponding characteristic previously obtained by comparison of expert color reference solutions, e.g., prepared based on definitions of a regulatory agency. Preferred advanced techniques are based on (numerical) regression and / or artificial intelligence and / or machine learning methods or systems based on trained machine learning structures. Herein, correlation is preferably represented by tools or means that are adapted / trained based on pre-classified reference fluorescence spectra 14 and that do not necessarily require the use of pre-classified reference fluorescence spectra 14. The correlation device 18 preferably comprises or is formed by an artificial intelligence module 19. The artificial intelligence module 19 is configured to predict the color value 12 or a corresponding characteristic based on a characteristic of the fluorescence radiation 6, e.g. spectrum 6A or a feature thereof, by means of an artificial intelligence that takes into account or represents a correlation between the characteristic of the fluorescence radiation 6 and the color value 12 or a characteristic thereof.

[0039] Artificial intelligence module 19 preferably includes a trained artificial intelligence structure such as an artificial neural network 20 (ANN), as illustrated by the respective pictograms of the boxes representing artificial intelligence module 19 in FIG. A simplified example of a possible neural network 20 is shown in Figure 2. The neural network 20 consists of multiple nodes 23 and edges 24 connecting the nodes 23. There can be multiple layers of nodes 23A, 23B, 23C. In the example shown in Figure 2, there are three layers, the middle layer is called the hidden layer, while the input node 23A is on the left side of the diagram and the output node 23C is on the right side. The artificial intelligence module 19, which particularly preferably uses a neural network 20, is configured to assign or predict a color value 12 or corresponding property of the solution 2 or product 3 based on the measured fluorescent radiation 6, preferably the spectrum 6A, or features thereof. The artificial intelligence module 19 can thus implicitly exploit correlations between properties of the fluorescent radiation 6 and the color values ​​12 or corresponding properties, in particular by learning or having learned the correlation based on the reference spectra 14 or features thereof and the assigned color values ​​12 or corresponding properties.

[0040] Alternatively or additionally, the correlation device 18 comprises a regression module 21 for predicting the color value 12 or corresponding property based on the fluorescence radiation 6, the spectrum 6A or features thereof by a (numerical) regression method. To that end, regression parameters 22, shown in the example represented by a spectral pictogram, can be determined or have been determined using one or more reference spectra 14 or their features. The regression parameters 22 thus take into account or represent the correlation between the properties of the fluorescence radiation 6 and the color value 12 or corresponding property. Therefore, the prediction and classification of the color values ​​12 may be performed by machine learning methods, in particular by one or more (feed-forward) neural networks 20, for example by an artificial intelligence module 19. It should be noted that different or further numerical regression or classification schemes such as partial least-squares regression (PLS2), or other machine learning techniques such as support vector machines, random forests, etc. may also be used, for example, by regression module 21 as an alternative or in addition.

[0041] Thus, although the present invention need not be limited to the use of an ANN 20, it has surprisingly been found that an ANN 20 provides a high level of accuracy in this situation. The correlation device 18, neural network 20 or regression method may be implemented, for example, in Python by using further open source modules such as TensorFlow, Keras, PyTorch or licensed program code toolkits such as MatLab. The neural network 20 may be pre-trained using pairs of reference spectra 14 (or features thereof) and color values ​​12 (or corresponding properties). By training the neural network 20 using the reference spectra 14 or one or more features thereof and the assigned (current or future) color values ​​12 or corresponding properties of the solution 2 or product 3, the behavior of the neural network 20 is such that an input spectrum 6A of the fluorescent radiation 6 from the solution 2 or product 3, or features thereof, may be input to the neural network 20 to result in an output (prediction) of the most likely current or future color value 12 or corresponding property. In particular, by training the neural network 20, the weights W of the neural network 20 are determined or adapted.

[0042] In one particularly preferred example, pairs of intensity P and wavelength λ are specified and input for a plurality of pre-classified spectra 14, while a corresponding color value 12 or corresponding characteristic of each of the plurality of pre-classified spectra 14 is specified as a target. Thus, at least one weight W of the neural network 20 is determined or adapted (trained) such that, when each intensity P and wavelength λ pair or (characteristic of) the spectrum 6A of the fluorescence radiation 6 containing or corresponding thereto is input, the neural network 20 outputs or is configured to output (a prediction of) the specified color value 12 or corresponding characteristic. The hyperparameter optimization procedure of the neural network 20 can be carried out in particular by cross-validation and / or a back-propagation algorithm can be used for iterative adjustment of the weights W of the neural network 20. To train the ANN, a set of pre-classified fluorescence spectra 6A can be used as training data, and it has surprisingly been found that a set containing more than 100 and / or less than 200 spectra 6A from different solutions and / or different products is sufficient for adequate accuracy.

[0043] Alternatively or additionally, for training, testing, and / or validation of the ANN, the training to testing data ratio can be selected from a data ratio greater than 70 / 30 and / or less than 90 / 10, preferably from 80 / 20. Alternatively or additionally, to train the ANN, corresponding magnitude P and wavelength λ pairs of the spectrum 6A can be used as standard descriptors and can be iteratively mapped to color values ​​12, preferably in a training phase on pre-classified Y and BY values. Random selection of pre-classified fluorescence spectra 6A as training data with multiple iterations, preferably more than 20 or 50, and in particular more than 100 iterations, can make it possible to estimate the average accuracy of the method for such small data sets. After the training phase, there is approximately a 12-byte error between the predicted and measured 12-byte color values. <r>In addition to an average Pearson correlation coefficient of =0.92±0.06, an average root-mean-square error (RMSE) of 0.49 was obtained. The obtained results therefore highlight the advantages of the present invention even in the case of small data sets.

[0044] In particular, the accuracy of the prediction improves with increasing amount of pre-classified fluorescence spectra 6A for training purposes. Thus, the presented approach provides a reliable classification of the actual color value 12 of the considered solution 2 or product 3. In particular, the present invention makes it possible to predict the color value 12 of a formulation for successively increasing concentrations of proteins, in particular mAbs. Herein, a linear correlation between the integrated fluorescence spectrum of the fluorescence radiation 6 (fluorescence intensity P versus wavelength λ) and the concentration of proteins, in particular mAbs, can be used as a prerequisite. The neural network 20 can be trained on the correlation between color values ​​12 (target values), concentrations of proteins, particularly mAbs, and fluorescence intensities P (input values). For determining the regression parameters 22 and / or training the neural network 20, in particular the current and / or future protein concentration of the solution 2 or product 3 and the integral 29 of the intensity P of the fluorescence radiation 6, in particular the area under the curve, of the spectrum 6A of the fluorescence radiation 6, for a plurality of samples in each case are used as input. Furthermore, a current and / or future color value 12 or a corresponding property is specified as a target. Based thereon, at least one weight W of the neural network 20 or at least one regression parameter 22 of the regression method can be determined or adapted. Furthermore, this is preferably performed in such a way that a color value 12 / corresponding property is predicted when each protein concentration and the integral of the fluorescence radiation intensity are input.

[0045] In this context, it is further preferred that pairs of intensity P and wavelength λ of the fluorescence radiation 6A and / or the generation process step for several (reference) samples are used as further inputs, as this has been found to improve the reliability of the results. A schematic flow chart of the comparison of predicted and measured color values ​​12 is shown in Figure 3. As shown on the left, multiple spectra 6A of different solutions 2 or products 3 are input into a trained artificial neural network 20, resulting in different predicted color values ​​12, in the right-hand diagram, the Y-axis values ​​of the BY (brownish-yellowish) color values ​​12, while the final color of the product 3 or solution 2 is measured, forming the X-axis values ​​of the BY color values ​​12 in the right-hand diagram of Figure 3. On the other hand, the predicted color values ​​12 and the measured color values ​​12 are close, or at least within the range of identity of the measured and predicted BY values ​​indicated by the lines in said diagram, indicating that the prediction based on the artificial intelligence method works well.

[0046] The fluorescence intensity P calculated from the linear regression fit and the concentration of the protein, in particular the mAb, of the considered solution 2 or product 3 can be used as input values ​​for the neural network 20, which can then predict the resulting color value 12. The corresponding value of the Pearson correlation coefficient between the predicted and experimental color values ​​is approximately <r2>=0.92±0.05, and this value will likely improve in the near future due to a larger training dataset that includes a wide range of mAb concentrations. In the case of regression using a regression module 21 instead of training a neural network 20, a regression can be performed to determine regression parameters 22 to associate the spectrum 6A of the fluorescent radiation 6 with and determine the (current or future) color value 12 or corresponding property of the solution 2 or product 3. When utilizing an artificial intelligence module 19, the reference spectra 14 are already taken into account for training the artificial intelligence, e.g., neural network 20, and do not need to be used directly to use the correlation to determine the color values ​​12 or corresponding characteristics.

[0047] To determine the color value 12 or corresponding characteristic, in particular the application of a neural network 20 or regression preferably takes into account the production process stage and / or the protein concentration of the solution 2 or product 3 (which may be measured in addition to determining the fluorescence radiation spectrum 6A of the fluorescence radiation 6). For example, a light color at an early process step may be more or less significant than a similar light color after purification of a protein based on Solution 2. Thus, either the basis of evaluation or the determined color value 12 may be evaluated or modified based on or taking into account the production process stage and / or protein concentration. The weights W or regression parameters 20 of the neural network 20 or the neural network 20 or regression method can have respective inputs to take into account the stage and / or concentration. This is particularly advantageous when the coloring tendency is to be evaluated or the likely future color of the solution 2 or product 3 is to be predicted, or when the current color has to be properly evaluated with respect to the validity of the determined color. Alternatively or additionally, different regression parameters or neural networks 20 may be used depending on each stage and / or protein concentration. Preferably, the stage and / or concentration is or has been taken into account when training each neural network 20 or determining the regression parameters 22.

[0048] In the case of a direct approach, for correlation purposes, there may be a reference set comprising different pre-classified spectra 14, i.e., reference spectra 14 or corresponding features, each assigned to a color value 12 or corresponding characteristic, for a particular production process stage and / or for different protein concentrations. Finally, there may be adaptations or modifications applied depending on the individual production process step or protein concentration. Herein, an essentially linear relationship between fluorescence intensity P and concentration has been found, on the basis of which extrapolations can be performed. The correlation can be performed, or alternatively, can take into account the similarity of wavelength, intensity, or intensity-wavelength pairs of the spectrum 6A of the fluorescent radiation 6, and the current or future color value 12 or corresponding property of the solution 2 or product 3. That is, it is not necessary to utilize the spectrum 6A of the fluorescent radiation 6 in such a way, and alternatively or additionally, features of its spectrum, such as wavelength-intensity pairs, can be used for the correlation. The measured fluorescence radiation 6 is preferably directly or indirectly related to a current or future color value 12 or corresponding property of the solution 2 or product 3. It is particularly preferred that the intensity P and wavelength λ pair of the spectrum 6A is used to determine the color value 12 or corresponding property.

[0049] The color value 12 is preferably determined using a correlation of the spectrum 6A of the fluorescent radiation 6 with a reference 14, or features of the spectrum 6A such as the maximum intensity wavelength, the shape of the shoulder at or adjacent to the maximum intensity wavelength, site maxima, or the overall shape and / or progression of the spectrum 6A of the fluorescent radiation 6. The present invention particularly preferably facilitates prediction of future color values ​​12 or properties of a solution 2 or product 3, for example, with respect to suitability for administration as a pharmaceutical. In one aspect of the present invention, a current color or corresponding color value 12 is determined by determination / correlation, which can then serve as the basis for prediction of the color, color value 12, or property expected for a product 3 in a subsequent process performed using the solution 2. However, alternatively or additionally, a correlation may be performed directly between (the spectrum 6A of) the fluorescent radiation 6 and the future or final color, color value 12, or corresponding property. In particular, by such a direct correlation of the current spectrum 6A with the future or expected property or coloration of the solution 2 or the final property or coloration of the product 3, any extrapolation may be avoided.

[0050] The correlation can be performed by regression, but as already indicated, it is particularly preferred to make use of artificial intelligence, in particular machine learning and / or neural networks 20 . For prediction purposes, based on the current fluorescence radiation spectrum 6A of the reference sample and the future or final color value 12 or corresponding characteristic, a neural network 20 can be trained and regression parameters 22 can be determined. A series of experimental fluorescence measurements were performed at low protein concentrations, especially low mAb concentrations. Standard numerical regression methods, such as least squares, were used to calculate the corresponding linear slope and offset of the associated data set. The correlation was found to be essentially linear, i.e., it did not change with increasing / higher protein concentrations, especially the concentration of mAb, and therefore the results could be extrapolated to the concentration of product 3 (of the relevant drug). In this regard, a linear regression fit is then preferably used to calculate the resulting fluorescence intensity P for desired high protein concentrations, particularly high concentrations of mAb. In a further aspect of the invention, a production process for producing a protein in a solution 2 or a product 3 can be controlled based on the correlation. In particular, the production process can be controlled based on a color value 12 or a corresponding property, in particular a color value 12 or a corresponding property determined based on (a property of) the fluorescence radiation 6, which can be one or more features of the spectrum 6A, on which at least one process parameter of the production process, in particular of a purification step, is preferably controlled.

[0051] In other words, the results of the correlation according to the present invention can be used to modify parameters or complete steps within the production process, in particular purification steps, in order to improve the results, i.e. to reduce the color (intensity) of the final protein-containing solution 2 or the product 3 produced therefrom. In this context, Figure 4 shows an example of a downstream process for purifying a protein produced in a protein-containing solution 2. Prior to applying this method, the protein-containing solution 2 can be, for example, a cell culture. The process shown starts with harvesting the cell culture fluid, followed by a chromatography step, a preparation step to allow subsequent filtering, and finally, one or more chromatography steps and further optional steps to purify the protein, in particular to produce a (purified antibody) product 3. According to the invention, the parameters for controlling each step can be checked or changed, or some of the steps can be omitted or replaced depending on the correlation results. In Figure 1, process control 27 is symbolized by arrows.

[0052] Techniques for producing protein-containing solutions within the meaning of the present invention are known per se to those skilled in the art. Protein-containing solutions are preferably produced biotechnologically by culturing, preferably fermenting, suitable prokaryotic or eukaryotic cells, in particular bacterial, fungal, or mammalian cells. This means that the cultured cells express the protein of interest in cell culture in a suitable medium under conditions that allow growth and / or protein production / expression. When considering feeding strategy batch culture, fed-batch or continuous cell culture, or a combination thereof, are known and are individually selected taking into account the requirements of the cells and the intended production protocol. Cells suitable for producing secreted recombinant therapeutic proteins are sometimes called "host cells." When considering the physical setting, the cells can be cultured in an adherent, encapsulated, and / or suspended form. A suspension of cells in the respective medium is often preferred. In certain embodiments, the host cell may further comprise one or more expression cassettes encoding a heterologous protein, such as a therapeutic protein, e.g., a secreted recombinant therapeutic protein. Expression of the protein of interest then occurs in the cells containing the DNA sequence coding for the biological product or recombinant protein of interest, which is transcribed and translated into a protein sequence including post-translational modifications to produce the biological product or recombinant protein of interest in cell culture.

[0053] In certain embodiments, production of such proteins of interest involves culturing bacterial cells, such as Escherichia coli as an example of a Gram-negative bacterium or Bacillus subtilis as an example of a Gram-positive bacterium, both of which have advantageously been previously transformed with genetic elements coding for the respective protein. The protein of interest can then be purified from the cells (e.g., from the periplasm of Gram-negative bacteria) or directly from the cell culture medium as a secreted protein. In other embodiments, production of such proteins of interest involves culturing eukaryotic cells, such as fungal cells. Cells of strains of the genus Pichia (e.g., Pichia pastoris) and yeast (e.g., Saccharomyces cerevisiae) are preferred, especially those that secrete proteins into the cell culture medium.

[0054] In certain embodiments, the eukaryotic host cell is an animal cell, such as an insect cell or a mammalian cell, such as a hamster cell or a mouse cell, such as a mouse myeloma cell, such as NS0 and Sp2 / 0 cells, or a rodent cell, such as a mouse cell, or a derivative / progeny of any such cell line. Such mammalian cells can be isolated cells or cell lines, preferably transformed and / or immortalized cell lines. In certain embodiments, the mammalian cells are adapted for continuous passage in cell culture and do not include primary non-transformed cells or cells that are part of an organ structure. In certain embodiments, the mammalian cells are BHK21, BHK TK-, Jurkat, 293, HeLa, CV-1, 3T3, CHO, CHO-K1, CHO-DXB11 (also called CHO-DUKX or DuxB11), CHO-S, and CHO-DG44 cells, or a derivative / progeny of any such cell line. In certain embodiments, the mammalian cells are CHO cells, such as CHO-DG44, CHO-K1, and BHK21, and more preferably, CHO-DG44 and CHO-K1 cells. In certain embodiments, the mammalian cells are CHO-DG44 cells. Glutamine synthetase (GS)-deficient derivatives of mammalian cells, particularly CHO-DG44 and CHO-K1 cells, are also encompassed. In one embodiment, the mammalian cells are Chinese hamster ovary (CHO) cells, such as CHO-DG44 cells, CHO-K1 cells, CHO DXB11 cells, CHO-S cells, CHO GS-deficient cells, or derivatives thereof.

[0055] A suitable technique for producing a protein-containing solution advantageously comprises, in the sense of the present invention, the following steps: I. Cultivating cells expressing the protein of interest in cell culture as described above. II. Harvesting the protein of interest from the cell culture, for example by centrifugation, known per se to those skilled in the art, thereby obtaining a liquid form (harvested cell culture fluid (HCCF)) containing the protein of interest and one or more impurities, buffer components, or other components as disclosed above. III. One or more chromatographic steps to capture or purify the protein of interest, including affinity, anion, and / or exchange chromatography of the liquid containing the protein of interest. The choice and sequence of these chromatographic steps is known per se to those skilled in the art and can be designed individually for each protein and liquid. IV. Optionally, one or more further steps, such as viral filtration and / or inactivation, concentration (e.g., by ultrafiltration), buffer exchange (e.g., by diafiltration). V. Optionally, forming the protein of interest into a pharmaceutically acceptable formulation suitable for administration.

[0056] The ultrafiltration (UF) and diafiltration (DF) steps can advantageously be combined as follows: (a) First ultrafiltration (UF1), followed by (b) a first diafiltration (DF1) using a high ionic strength buffer, followed by (c) a second diafiltration (DF2) using a buffer of low ionic strength, in particular one having an ionic strength lower than that of the buffer of DF1, followed by (d) followed by a second ultrafiltration (UF2) All of these steps (a)-(d) are described in further detail in WO 2018 / 033482.

[0057] For the purification of antibodies and antibody-like proteins, for example, the following purification protocol may be applied. i. Starting with a cell culture (pre-harvest cell culture fluid (CCF)) from a fermentation, preferably mammalian cells. ii. Harvesting by centrifugation, resulting in harvested cell culture fluid (HCCF). iii. Affinity chromatography using Protein A. iv. Viral inactivation (VI). v.Depth filtration (DF). vi. Anion exchange chromatography (AIEX), for example in elution mode. vii. Cation exchange chromatography (CIEX), for example in bind / elute mode, alternatively hydrophobic interaction chromatography (HIC), or a mixed mode of both these chromatographies. viii. Viral filtration (VF). ix. Preferably, ultrafiltration and / or diafiltration (UF, DF), or vice versa, as disclosed above. x. The resulting bulk drug substance (BDS) is: xi. Optionally, further processed, re-buffered, and / or filled into vials or other suitable devices.

[0058] The present invention is particularly concerned with color classification of protein-containing solutions or products prepared therefrom. It is therefore particularly preferred that the correlation is carried out based on the spectrum 6A of the fluorescence radiation 6 arising from the excited solution 2 or product 3 during, before or after one or more of the following process steps, or using a sample obtained in one or more of the following process steps: (i) culturing eukaryotic cells expressing the recombinant protein of interest in cell culture; (ii) harvesting the recombinant protein; (iii) purifying the recombinant protein; and (iv) optionally, forming the recombinant protein into a pharmaceutically acceptable formulation suitable for administration; and (v) obtaining at least one sample containing the recombinant protein in steps (ii), (iii), and / or (iv).

[0059] The methodology used in the present invention can preferably be fully automated using appropriate laboratory equipment and program code, with the numerical regression methodology and corresponding fluorescence measurements, with human labor being used only for the preparation and mixing of the corresponding solution 2. The correlation according to the present invention is preferably performed automatically based on samples of solution 2 or product 3 that are loaded onto a microtiter plate 29 such as that shown in FIG. 5, which shows an example of a microtiter plate 29 containing 96 wells 30 for loading solution 2 or product 3. A light source 4 can provide fluorescence excitation radiation 5 to each of the wells 30, either stepwise or simultaneously, and one or more spectrometers 7 can measure the spectrum 6A of the fluorescence radiation 6 arising from the solution 2 or product 3 contained in each well 30. This facilitates automating the characterization and / or prediction of the correlation value 12 or its corresponding property. 6A shows an example of a spectrum 6A, a chart of the intensity versus wavelength of the fluorescent radiation 6 emitted by solution 2 or product 3 when excited at 390 nm. The light intensity of the fluorescent radiation 6 typically depends on the protein concentration in solution 2 or product 3. It is therefore understood that the concentration or corresponding process step is taken into account when evaluating the measurement results. It is particularly preferred that the integral 29 (area under the curve) is used to determine the corresponding color value 12.

[0060] Figure 6B shows six overlaid charts of fluorescence intensity versus wavelength of fluorescent radiation 6 emitted from different products 3 or solutions 2 when excited at 390 nm. All products 3 or solutions 2 were adjusted to a protein concentration of 10 mg / mL. 7 shows a chart of area under the curve versus concentration. It has thus been found that the area under the curve of the spectrum of fluorescence radiation 6 is essentially linearly related to the current or future protein concentration of solution 2 or product 3. This can be used to interpolate or extrapolate the intensity P of fluorescence radiation 6 versus concentration and / or to predict the fluorescence radiation 6 or current or future color value 12. This is preferably taken into account to determine color value 12 or a corresponding characteristic. In particular, neural network 20 is trained to take such a relationship into account, or regression parameters 22 are adapted or determined based on or taking into account this. FIG. 8 illustrates the essentially linear relationship between measured fluorescence and protein concentration.

[0061] In a particularly preferred embodiment, a machine learning or regression model is used to determine the future color values ​​12 or corresponding properties of the solution 2 or product 3, the model being trained using the reference spectra 14 or their features and the corresponding current or future color values ​​12 or corresponding properties of the solution 2 or product 3. Particularly preferably, the future color value 12 or corresponding property of the solution 2 or product 3 is determined using an artificial neural network 20 trained using the reference spectrum 14, the current color value 12 and corresponding property of the solution 2 or product 3. The machine learning model, in particular the artificial neural network 20, can be dynamically and / or continuously adapted. Alternatively or additionally, the regression model can be dynamically and / or continuously adapted. To do this, the determined properties of the fluorescence radiation 6 based on the solution 2 during the production process (e.g. after it has been used for prediction) and the measured current or future color value 12 or corresponding properties of the solution 2 or product 3 are used as inputs for adapting the machine learning model, in particular the artificial neural network 20 and / or the regression model. When determining or predicting color value 12, the production process steps may be taken into account in that the characteristics of the fluorescence radiation 6 (preferably spectrum 6 or corresponding features) are normalized and / or extrapolated based on the expected change in protein concentration within the process from the step where a sample of solution 2 is collected / fluorescence radiation 6 or its characteristics is determined to the future step where future color value 12 or its characteristics is predicted.

[0062] For example, a scaling factor specific to the step in processing may be preferably determined empirically that represents the expected (preferably concentration-independent) effect of the process on color value 12 due to refinement in future process steps. This scaling factor can be specific to the process step in which a sample of solution 2 is collected and analyzed to predict future color values ​​12 or their corresponding properties 12. This scaling factor can be applied, in particular multiplied, as an input variable directly (or indirectly) to a property (preferably a spectrum or corresponding feature) or normalized or extrapolated value of fluorescence radiation 6 in the prediction process to determine future color values ​​12 or their corresponding properties. The properties (preferably spectrum or corresponding characteristics) of the fluorescence radiation 6 are: divided by the protein concentration of the protein in protein-containing solution 2 (of the collected sample), multiplied by the protein concentration of the desired bulk drug substance (the protein concentration of the protein in the desired future protein-containing solution 2 or product 3), It is particularly preferred that the production process steps can be taken into account in that they are preferably processed (multiplied) using a scaling factor specific to the step in question.

[0063] The results are finally used as input parameters for at least one of a regression method, preferably an advanced multifactorial regression, or an artificial intelligence method, preferably a supervised machine learning method, in particular an artificial neural network 20, to predict the degree of coloration of the bulk drug substance (product 3), i.e., color value 12 or corresponding property. Alternatively, one or more of the steps may be implemented in a supervised machine learning method, such as a regression technique, preferably an advanced multifactorial regression, or an artificial intelligence technique, preferably an artificial neural network 20. Each parameter "protein concentration", "desired bulk drug substance protein concentration", and / or "scaling factor specific to the step in the process" is then input in place of the corresponding result. In this way, the production process steps can be taken into account completely or in part by regression techniques, preferably advanced multifactorial regression, or artificial intelligence techniques, preferably supervised machine learning methods, in particular artificial neural networks 20. In one option, one or more intermediate results of the steps and parameters of other steps selected from "protein concentration of the protein in protein-containing solution 2," "protein concentration of the desired bulk drug substance," and "optionally, a specific scaling factor" are used as input parameters.

[0064] In another option, the characteristics of the fluorescence radiation 6 and the parameters "protein concentration of the protein in the protein-containing solution 2", "protein concentration of the desired bulk drug substance", and "optionally, a specific scaling factor" are used as input parameters. In this way, a supervised machine learning method, such as a regression technique, preferably an advanced multifactorial regression, or an artificial intelligence technique, preferably an artificial neural network 20, is preferably trained using the corresponding current or future color value 12 or corresponding characteristic and one or more of the following parameters (sets) of the following alternative input parameter groups A to D: A: ·Characteristics of Fluorescent Radiation 6 Protein concentration of the protein in protein-containing solution 2 Desired bulk drug substance protein concentration Optional: A scaling factor specific to the step being processed B: Characteristics of fluorescence radiation specific / extrapolated to protein mass Desired bulk drug substance protein concentration Optional: A scaling factor specific to the step being processed C: Predicted intrinsic / extrapolated bulk drug substance fluorescence radiation characteristics Desired bulk drug substance protein concentration D: · Predicted intrinsic / extrapolated bulk drug substance fluorescence radiation6 characteristics multiplied by the desired bulk drug substance protein concentration

[0065] The following is an exemplary description of a procedure for predicting the color (represented by color value 12 or a corresponding property) of a bulk drug substance (product 3) by measuring an in-process sample (of protein-containing solution 2) of any process step. The degree of color development of the bulk drug substance (Product 3), as expressed by the color value 12 or corresponding property, is: excitation of the fluorescence radiation 6 of the solution 2 derived from a sample of a processing step (of the production or purification of a protein in the protein-containing solution 2), for example at 390 nm, measuring at least one characteristic, preferably a spectrum 6A or a corresponding feature (such as intensity or, for example, the area under the curve of integral 29 / intensity), of the fluorescence radiation 6, for example, between 420 nm and 600 nm, preferably after blank subtraction; The measured property is further divided by the protein concentration of the sample of protein-containing solution 2, resulting in a fluorescence intensity specific / normalized to the mass of the protein, which is then predicted. Hereinafter, "fluorescence intensity" is an example of a property of fluorescence radiation 6 and can be replaced with "property of fluorescence radiation 6".

[0066] This protein mass-specific / normalized fluorescence intensity is then processed using a scaling factor specific to the step in the process, resulting in a predicted / extrapolated bulk drug substance-specific fluorescence intensity. In the next step, the predicted bulk drug substance intrinsic fluorescence intensity is multiplied by the protein concentration of the desired bulk drug substance, and the result is used as an input parameter for an advanced multifactorial regression or supervised machine learning method to ultimately predict the degree of color development (color value 12 or corresponding property) of the bulk drug substance (product 3). Consequently, this high-throughput compatible procedure, together with the method's sensitivity, robustness, and wide concentration range, provides access to a wide range of applications, leading to a recombinant protein manufacturing method that unlocks the ability to predict, track, and ultimately control the degree of color development of product 3 / bulk drug substance, ultimately resulting in improved product quality and reduced process development time and manufacturing costs.

[0067] In the following, the proof-of-concept results are explained with reference to Figures 9-16. In this proof-of-concept, protein concentrations of 100 mg / mL for bulk drug substance (BDS) after ultrafiltration (UF) and 24 mg / mL for protein-containing solution 2 were selected to compare the predicted degree of color (color value 12 or corresponding characteristic) of each bulk drug substance (product 3) of monoclonal antibody 1 (mAb1, BDS), selected as an exemplary monoclonal antibody preparation of molecular type IgG1 at the BDS stage, and mAb2 (UF), selected as an exemplary monoclonal antibody preparation of molecular type ZweiMab+ (e.g., as disclosed in WO 2019 / 234220), with the measured degree of color according to Ph.Eur. (see Figure 15 (mAb1) and Figure 16 (mAb2)).

[0068] 1. Classification of BY values ​​from fluorescence spectra The following describes the application of a machine learning method for determining the BY value (an example of "color value 12", which may be replaced by "color value 12" hereinafter) from the fluorescence spectrum (6A). This method is preferably supported such that any advanced multi-factor supervised regression technique can be used, but is not limited to a specific machine learning method. 1.1. Training Data and Verification Data As training data and verification data for the machine learning model, a set of 123 spectra (6A) of different monoclonal antibodies (mAbs) in a buffer solution (solution 2) was used. The concentration of the mAb varied within the range of 0.00 mg / mL to 84.20 mg / mL. Regarding the corresponding BY values, the corresponding solution 2 was classified by human inspection. The observed BY values were within the range of BY = 7.5 to BY = 1.5. More specifically, the human classification introduced only integer values within the range of BY = 0 to BY = 7. For intermediate results, solution 2 was classified as "less than X" meaning BY < X. This classification corresponds to the irregular definition X > X + 1. Since Boolean logic is difficult to implement in a supervised machine learning regression technique, all classifications BY < X were encoded as X.5 for training and prediction purposes.

[0069] 1.2. Calculation Details All source code was written in Python 3.7.4 [1] using the modules NumPy 1.16.5 [2] and Scikit-Learn 0.21.3 [3]. The artificial neural network (ANN) model, random forest (RF) model, and extra trees (ET) model, the results of which are shown in Figures 9, 10, 13, 14, and 15, were implemented using the functions MLPRegressor, RandomForestRegressor, and ExtraTreesRegressor as part of Scikit-Learn [3]. Unless otherwise noted, hyperparameters were set to default values. For the ANN model, a MinMax scaler was used, as implemented in the function MinMaxScaler as part of Scikit-Learn.

[0070] 1.3. Training and Validating Machine Learning Models The corresponding fluorescence spectrum 6A S(W) is s =420nm~W s All were monitored using a step size of 2 nm for a wavelength of λ = 600 nm. For training purposes and subsequent application of the machine learning model, the fluorescence spectra 6A were used as input values, the individual magnitude values ​​of specific wavelengths M(W) were used as feature values, and the corresponding wavelengths were used as features or descriptors. As target values, the corresponding pre-classified BY values ​​(characteristics or spectra 6A) by human inspection were used. Different machine learning models were trained and validated using the leave-one-out cross-validation (LOOCV) method [4], including, inter alia, artificial neural networks (ANN), random forests (RF), extra trees (ET), or gradient boosting. This means that N-1 data points were used for training, and the remaining data points were used for validation. The random permutation of the training data and the LOOCV method ultimately allowed for the evaluation of the statistical predictive accuracy of the individual models. The root mean square error of prediction was used as the standard statistic.

number

number

number

[0071] 1.4.Results Figures 9 and 10 show plots of the goodness of fit of the validation data for the LOOCV approach between predicted and measured BY values ​​for the ET model (Figure 9) and the ANN model (Figure 10). The ET model was trained using 100 estimators. The ANN model contained three hidden layers with 30, 20, and 10 nodes. A logarithmic activation function was used. The corresponding results show reasonable accuracy of the model's predictions. The ET model performs slightly better than the ANN model in this example, with RMSE values ​​of 0.13 (ET) and 0.33 (ANN). As already explained in Section 1.1, BY values ​​are classified in integer units. With RMSE << 1 for all model predictions, it can be concluded that the prediction and classification accuracies fully meet the individual objectives. It should be noted that a larger dataset containing a greater amount of evenly distributed BY values ​​can improve the prediction accuracy.

[0072] 2. Prediction of color values ​​12 for future process steps and densities In the following sections, methods are described for the prediction of color values ​​12 and integrated fluorescence intensity for selected mAb concentrations after specific process steps, where a protein-containing solution 2 (also considered as the probe in process) is obtained after protein production, known per se, in which a transgene coding for each protein is expressed by a fed-batch culture of Chinese hamster ovary (CHO) cells and secreted into the cell culture medium, from which the protein is purified in various successive steps, each of which then results in a further purified protein-containing solution 2. Affinity chromatography (AF), depth filtration (DF), cation exchange chromatography (CIEX), mixed mode chromatography (MM), virus filtration (VF), ultrafiltration (UF) and / or diafiltration, as well as pure bulk drug substance (BDS) in buffer. Further process steps such as anion exchange chromatography or liquid-liquid separations can be easily integrated into the framework.

[0073] 2.1.Scaling Factor First wavelength selected W s and the final wavelength W e Fluorescence spectra 6A S measured for selected process steps X ∈ [AF,DF,CIEX,MM,VF,UF,BDS] over various wavelengths W between X (W) was integrated according to the following formula:

number

number

number

[0074]

number

number

[0075] [Table 1]

[0076] It is important to note that these scaling factors are universally applicable, i.e., independent of the specific format. In particular, the influence of different process steps on the resulting fluorescence intensity can be clearly recognized, as subsequent process steps result in lower fluorescence intensities. Here, the corresponding predicted fluorescence intensity

number

number

[0077]

number

number

number

[0078] 2.2. Training and Validation A process-step-independent dataset consisting of various spectra for different mAbs was used to train the machine learning model. This dataset contained 153 fluorescence intensities with minimum and maximum values ​​of 5824.24 au and 180030.66 au. The corresponding minimum and maximum concentrations were 0.00 mg / mL and 143.15 mg / mL, with minimum and maximum BY values ​​of 7.5 and 1.5, as defined in Section 1.1. As machine learning methods, an artificial neural network (ANN) model and a random forest (RF) model were trained. The ANN consisted of one hidden layer containing 100 nodes. A rectified linear unit (ReLU) activation function and a constant learning rate of 0.001 were used. The RF model was trained using 100 tree estimators. The corresponding results for the validation data, as calculated by the leave-one-out method, are shown in Figure 12: Measured fluorescence intensity and predicted fluorescence intensity with prediction error for a selected mAb1 at different process steps X and specific concentrations. The connecting lines are merely guides for clarity. As input features, the concentrations and fluorescence intensities associated with the corresponding BY values ​​as targets were used. The corresponding accuracy of predictions for the experimental dataset, when split into test and training data according to the leave-one-out approach, is shown in Figures 13 and 14: Goodness-of-fit plots of predicted and measured BY values ​​for the RF model. Figure 13: Results for the RF model. Figure 14: Results for the ANN model. The black line indicates perfect agreement with a slope of 1. As shown in the figures, the RF model performs slightly better in terms of lower normalized root-mean-squared errors (nRMSE) of predictions when compared to the ANN model. It becomes clear that various regression-based machine learning methods can yield reasonable prediction accuracy.

[0079] 2.3. Predicting BY values ​​for future process steps The corresponding predictions of the ET and ANN models for two mAbs (one lgG1 (mAb1) and the other Zweimab+ (mAb2)) at specific concentrations at different process steps are shown in Figures 15 and 16. Figure 15: Predicted BY values ​​from the ANN and ET models are shown along with measured BY values ​​at specific process steps and concentrations for a specific mAb1. Figure 16: Predicted BY values ​​from the ANN and RF models are shown along with measured BY values ​​at specific process steps and concentrations for a specific mAb2. The solid and dashed lines are merely guides for ease of viewing. As can be seen in the figures, the predicted BY values ​​are consistent with the BY values ​​measured for validation purposes. Furthermore, it can be seen that the differences between the RF and ANN model predictions are only slight. Therefore, it can be concluded that the proposed method, including the scaling factor combined with machine learning, provides reliable prediction of BY values ​​for future process steps and arbitrarily selected concentrations. As a prerequisite, only the concentration of the mAb and the corresponding fluorescence intensity for AF need to be known. All other fluorescence intensity and color values ​​can be easily predicted after applying the scaling factor and the trained machine learning model.

[0080] 3.References [1] G. Van Rossum and FL Drake, Python 3 Reference Manual, CreateSpace, Scotts Val-ley, CA, 2009. [2] CR Harris, KJ Millman, SJ van der Walt, R. Gommers, P. Virtanen, D. Courna-peau, E. Wieser, J. Taylor, S. Berg, NJ Smith, R. Kern, M. Picus, S. Hoyer, MH van Kerkwijk, M. Brett, A. Haldane, JF del Rio, M. Wiebe, P. Peterson, P. Gerard-Marchant, K. Sheppard, T. Reddy, W. Weckesser, H. Abbasi, C. Gohlke and TE Oli-phant, Nature, 2020, 585, 357-362. [3] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot and E. Duchesnay, J. Mach. Learn. Res., 2011, 12, 2825-2830. [4] T.-T. Wong, Pattern Recogn., 2015, 48, 2839

[0081] Further aspects of the present invention are as follows. 1. A method for determining the color value (12) or a corresponding property of a protein-containing solution (2) or of a protein-containing product (3) prepared from the protein-containing solution (2), comprising: the method comprising exciting a fluorescent radiation (6) of the solution (2) or product (3), measuring at least one property, preferably a spectrum (6A) or a corresponding feature, of the fluorescent radiation (6), and determining a current or future color value (12) or a corresponding feature of the solution (2) or product (3) based on a correlation between the at least one property of the fluorescent radiation (6) and the color value (12) or the corresponding feature.

[0082] 2. a. the fluorescence excitation radiation (5) has a maximum intensity (P) at a wavelength (λ) greater than 310 nm and less than 540 nm, preferably greater than 360 nm and less than 420 nm, in particular greater than 380 nm and less than 400 nm, and / or b. Fluorescence radiation (6) is detected within a wavelength (λ) range of at least 330 nm to 800 nm; and / or c. the fluorescent radiation (6) has a maximum intensity (P) at a wavelength (λ) greater than 330 nm and less than 800 nm, preferably greater than 420 nm and less than 600 nm, in particular greater than 450 nm and less than 530 nm, and / or d. the intensity (P) maximum of the fluorescence radiation (6) is at a wavelength (λ) that is more than 60 nm and / or less than 130 nm beyond the wavelength (λ) at which the fluorescence excitation radiation (5) has a maximum intensity (P); and / or e. The method of aspect 1, wherein the color value (12) or corresponding property corresponds to a yellowish or brownish-yellowish coloration, preferably characterized in that light can be transmitted, reflected, or scattered by the solution (2) or product (3) with a maximum transmittance, reflectance, or scattering at wavelengths (λ) greater than 560 nm and / or less than 620 nm.

[0083] 3. The method according to aspect 1 or 2, characterized in that the solution (2) or the product (3) comprises a recombinant protein or antibody, preferably a monoclonal antibody, and the color value (12) or the corresponding property is a property of the recombinant protein or antibody, preferably a pharmacological utility. 4. The method according to any one of aspects 1 to 3, characterized in that the production process stage is taken into account to determine the color value (12) or a corresponding characteristic, and / or the protein concentration is taken into account to determine the color value (12) or a corresponding characteristic. 5. The method according to any one of aspects 1 to 4, characterized in that the fluorescent radiation (6) has a wavelength (λ) and / or an intensity (P), in particular a pair of wavelength (λ) and corresponding intensity (P) values, based on which a current or future color value (12) or corresponding property of the solution (2) or product (3) is determined. 6. The method according to any one of aspects 1 to 5, characterized in that the current or future color value (12) or corresponding property of the solution (2) or product (3) is determined based on at least one property of the fluorescence radiation (6) using regression techniques or artificial intelligence, preferably using machine learning, in particular using an artificial neural network (20).

[0084] 7. The method according to any one of aspects 1 to 6, characterized in that determining the color value (12) or the corresponding characteristic is achieved by a trained neural network (20) or a regression technique for a numerical regression or classification procedure using at least one regression parameter (22). 8. The method of aspect 7, wherein the artificial neural network (20) is pre-trained and at least one regression parameter (22) is determined, comprising pre-training the artificial neural network (20) or determining at least one regression parameter (22), preferably wherein pairs of intensity (P) and wavelength (λ) of the fluorescence radiation (6) obtained from the fluorescence radiation spectrum (6A) are specified as inputs (25) for a plurality of solutions (2) or products (3), and color values ​​(12) or corresponding characteristics are specified as targets (26), and wherein at least one weight (W) or at least one regression parameter (22) defining a characteristic of the neural network (20) is configured, determined or adapted such that, upon input of each pair of intensity (P) and wavelength (λ), the regression method based on the artificial neural network (20) or the regression parameter (22) outputs or is configured to output the specified color value (12) or corresponding characteristic.

[0085] 9. The method according to aspect 7 or 8, characterized in that based on at least one property of the fluorescent radiation (6), a color value (12) or a corresponding property of the solution (2) or product (3) is predicted for a future stage of a production method for producing the solution (2) or product (3). 10. An artificial neural network (20) is pre-trained and at least one regression parameter (22) is determined, comprising pre-training an artificial neural network (20) or determining at least one regression parameter (22), for a plurality of samples, in each case: a. the protein concentration of the solution (2) or product (3), in particular the current and / or future, and / or b. The integral of the intensity (P) of the fluorescence radiation (6), specifically the area under the curve of the intensity (P) of the fluorescence radiation (6) 10. The method according to any one of aspects 7 to 9, wherein weights (W) defining the characteristics of the artificial neural network (20) or at least one regression parameter (22) of the regression method are determined or adapted such that, when the integrals of the respective protein concentrations and the intensities (P) of the fluorescence radiation (6) are input, a current and / or future color value (12) or corresponding property of the solution (2) or product (3) is designated as a target, and the color value (12) or corresponding property is predicted.

[0086] 11. The method according to aspect 10, characterized in that for a plurality of samples, pairs of intensity (P) and wavelength (λ) of the fluorescence radiation (6) and / or the generation process step are used as further inputs, and / or a parameter of the fluorescence radiation (6) or a color value (12) or a corresponding property of the solution (2) or product (3) is predicted. 12. The method according to any one of aspects 10 or 11, characterized in that the production process is controlled based on a parameter of the fluorescent radiation (6) or on a color value (12) or a corresponding property determined based on the fluorescent radiation (6), preferably at least one process parameter of the production process, in particular of the purification step, is determined or controlled based on a parameter of the fluorescent radiation (6) or on a color value (12) or a corresponding property. 13. The method according to any one of aspects 1 to 12, wherein the parameter of fluorescence radiation (6) is determined from a sample of solution (2) or product (3) when held in a well (30) of a microtiter plate (29).

[0087] 14. For producing a protein-containing solution (2) or a product (3) produced from a protein-containing solution (2), (vi) culturing eukaryotic cells expressing the recombinant protein of interest in cell culture; (vii) harvesting the recombinant protein; (viii) purifying the recombinant protein; and (ix) optionally forming the recombinant protein into a pharmaceutically acceptable formulation suitable for administration; and (x) obtaining at least one sample comprising the recombinant protein of steps (ii), (iii), and / or (iv), The method, wherein the sample is a solution (2) or a product (3), and the method further comprises carrying out the steps of the method according to any one of aspects 1 to 13.

[0088] 15. An apparatus (1) comprising a light source (4), a spectrometer (7) for measuring fluorescence radiation (6), and a device (18) adapted to perform the method according to any one of embodiments 1 to 14 based on the fluorescence radiation (6). The various aspects of the invention can be realized independently or in combination, and various synergistic effects may be obtained, even if not expressly mentioned herein. [Explanation of symbols]

[0089] 1 device 2. Protein-containing solution 2A Bioreactor 2B Sample Room 3 products 3B Small bottle 4 light source 5. Fluorescence Excitation Radiation 5A Spectrum 6. Fluorescent Radiation 6A Spectrum 7 Spectrometer 8 eyes 9 Continuous Light Source 10 Irradiated Light 10A Continuous Light Spectrum 11 Reflected light 11A Reflectance Spectrum 12 color values 13 Output Devices 14 Reference Spectra 15 Reference spectrometer 16 Input Devices 17 Databases 18 devices 19 Artificial Intelligence Module 20 Neural Networks 21 Regression Module 22 Regression parameters 23 nodes 24 Edge 25 inputs 26 outputs / targets 27 Process Control 28 Integral 29 Microtiter Plates 30 wells AF Affinity Chromatography AEX Anion Exchange Chromatography BDS Bulk Drug Substances CIEX Cation Exchange Chromatography DF Depth Filtration MM mixed-mode chromatography P Power / Strength / Size UF Ultrafiltration / Diafiltration VF Virus Filtration W weight λ wavelength < / r>

Claims

1. 1. A method for determining a future color value (12) or a corresponding future property of a protein-containing solution (2) or of a protein-containing product (3) prepared from said protein-containing solution (2), comprising: i) exciting the fluorescence radiation (6) of said solution (2) or product (3); ii) measuring at least one property, preferably a spectrum (6A) or a corresponding feature, of said fluorescent radiation (6); and iii) a1) said at least one characteristic of said fluorescent radiation (6); b1) determining a future color value (12) of the solution (2) or the product (3), or a property corresponding to the future color value, based on a correlation between the current color value (12) of the solution (2) or the product (3), the current color value (12) forming the basis for predicting a future color value (12) of the solution (2) or the product (3) or a property corresponding to the future color value, predicted in a later process of the solution (2) or the product (3); Or, a2) said at least one characteristic of said fluorescent radiation (6); b2) determining the future color value (12) of the solution (2) or the product (3) or a property corresponding to the future color value (12) based on a correlation between the future color value (12) of the solution (2) or the product (3) or a property corresponding to the future color value; The method comprising:

2. 2. The method of claim 1, wherein the correlation is performed directly between the at least one characteristic of the fluorescence radiation (6) and the future color value (12) of the solution (2) or the product (3) or a future characteristic corresponding to the future color value (12).

3. the current or future color value (12) of the solution (2) or the product (3) or a property corresponding to said color value is determined based on said at least one property of the fluorescent radiation (6) by using an artificial intelligence, preferably a machine learning model, in particular an artificial neural network (20), or preferably a regression model for numerical regression or classification using at least one regression parameter (22); 3. The method according to claim 1 or 2, characterized in that the artificial intelligence, preferably the machine learning model, in particular the artificial neural network (20) is trained or the regression model is determined using a reference spectrum (14) or its features and the current or future color value (12) or a property of the solution (2) or the product (3) corresponding to the current or future color value (12).

4. 4. The method of claim 3, wherein the artificial intelligence, preferably the machine learning model, in particular the artificial neural network (20) is trained or the regression model is determined, wherein pairs of intensity (P) and wavelength (λ) of the fluorescence radiation (6), preferably obtained from fluorescence radiation spectra (6A), are specified as inputs (25) for a plurality of solutions (2) or products (3), and current or future color values ​​(12) or properties corresponding to said color values ​​are specified as targets (26), and wherein at least one weight (W) defining a property of the neural network (20) is configured, determined or adapted such that, when each pair of intensity (P) and wavelength (λ) is input, the artificial neural network (20) based on the artificial neural network (20) outputs or is configured to output the specified current or future color value (12) or property corresponding to said color value.

5. 4. The method of claim 3, wherein based on the at least one characteristic of the fluorescent radiation (6), the future color value (12) of the solution (2) or product (3) or a future characteristic corresponding to the color value is predicted for a future stage of the production method for producing the solution (2) or product (3).

6. The artificial neural network (20) is pre-trained or the method comprises pre-training the artificial neural network (20), for a plurality of samples: a. the protein concentration of said solution (2) or product (3), in particular the current and / or future, and / or b. The integral of the intensity (P) of said fluorescence radiation (6), in particular the area under the curve of the intensity (P) of said fluorescence radiation (6).

4. The method of claim 3, wherein weights (W) defining a characteristic of the artificial neural network (20) are determined or adapted such that, when the integrals (P) of the respective protein concentrations and / or the intensity (P) of the fluorescence radiation (6) are input, the current and / or future color values ​​(12) of the solution (2) or product (3) or characteristics corresponding to said color values ​​are designated as targets, and the integrals (P) of the respective protein concentrations and / or the intensity (P) of the fluorescence radiation (6) are input, the current or future color values ​​(12) or characteristics corresponding to said color values ​​are predicted.

7. 7. The method of claim 6, wherein for the plurality of samples, pairs of intensity (P) and wavelength (λ) of the fluorescence radiation (6) and / or a generation process stage are used as further inputs and / or the at least one property of the fluorescence radiation (6) or the current or future color value (12) of the solution (2) or product (3) or a property corresponding to the color value is predicted.

8. 3. The method according to claim 1 or 2, characterized in that the production process of the protein-containing solution (2) or of a protein-containing product (3) prepared from the protein-containing solution (2) is controlled based on the at least one characteristic of the fluorescent radiation (6) or based on the future color value (12) or a corresponding future characteristic determined based on the fluorescent radiation (6).

9. 9. The method according to claim 8, characterized in that at least one process parameter of the production process, in particular of a purification step, is determined or controlled based on the at least one characteristic or future color value (12) of the fluorescence radiation (6) or a corresponding future characteristic.

10. the fluorescence excitation radiation (5) has a maximum intensity (P) at a wavelength (λ) greater than 310 nm and less than 540 nm, preferably greater than 360 nm and less than 420 nm, in particular greater than 380 nm and less than 400 nm, the fluorescence radiation (6) is detected within a wavelength (λ) range of at least 330 nm to 800 nm; the fluorescence radiation (6) has a maximum intensity (P) at a wavelength (λ) greater than 330 nm and less than 800 nm, preferably greater than 420 nm and less than 600 nm, in particular greater than 450 nm and less than 530 nm, the maximum intensity (P) of the fluorescence radiation (6) being at a wavelength (λ) greater than 60 nm and / or less than 130 nm beyond the wavelength (λ) at which the fluorescence excitation radiation (5) has a maximum intensity (P); and / or 3. The method according to claim 1 or 2, characterized in that the current or future color value (12) or the property corresponding to the color value corresponds to a yellowish or brownish-yellowish coloration, preferably characterized in that light can be transmitted, reflected or scattered by the solution (2) or product (3) with a maximum transmittance, reflectance or scattering at a wavelength (λ) greater than 560 nm and / or less than 620 nm.

11. 3. The method according to claim 1 or 2, characterized in that the solution (2) or product (3) comprises a recombinant protein or antibody, preferably a monoclonal antibody, and the current or future color value (12) or a property corresponding to the color value is a property of the recombinant protein or antibody, preferably a pharmacological utility.

12. 3. The method according to claim 1 or 2, characterized in that a production process stage is taken into account to determine the current or future color value (12) or a characteristic corresponding to the color value.

13. 3. The method according to claim 1 or 2, characterized in that the protein concentration is taken into account to determine the current or future color value (12) or a characteristic corresponding to the color value.

14. The solution (2) (i) culturing eukaryotic cells expressing said recombinant protein of interest in cell culture; (ii) harvesting the recombinant protein; (iii) Preferably, (a) affinity chromatography (AF); (b) depth filtration (DF); (c) anion exchange chromatography (AIEX); (d) cation exchange chromatography (CIEX); (e) mixed-mode chromatography (MM); (f) viral filtration (VF), and / or (g) purifying the recombinant protein, comprising one or more of: ultrafiltration / diafiltration (UF); and (iv) a sample comprising a recombinant protein obtained in one of steps (i) to (vi) or (a) to (g) of a process comprising forming said recombinant protein into a product (3), in particular a pharmaceutically acceptable formulation suitable for administration as a pure bulk drug substance (BDS) in a buffer solution; 3. The method according to claim 1 or 2, characterized in that the fluorescence radiation (6) is excited using the sample and the future color value (12) or a property corresponding to the color value is determined for different subsequent steps (i) to (iv) or (a) to (g).

15. 3. The method according to claim 1 or 2, characterized in that the fluorescent radiation (6) has a wavelength (λ) and / or an intensity (P), in particular a pair of wavelength (λ) and corresponding intensity (P) values, on the basis of which the current or future color value (12) of the solution (2) or product (3) or a property corresponding to the color value is determined.

16. 3. The method according to claim 1 or 2, wherein the parameter of the fluorescence radiation (6) is determined from a sample of the solution (2) or product (3) when held in a well (30) of a microtiter plate (29).

17. For producing a protein-containing solution (2) or a product (3) produced from the protein-containing solution (2), (i) culturing eukaryotic cells expressing the recombinant protein of interest in cell culture; (ii) harvesting the recombinant protein; (iii) purifying the recombinant protein; and (iv) optionally forming the recombinant protein into a pharmaceutically acceptable formulation suitable for administration; and (v) obtaining at least one sample comprising said recombinant protein in steps (ii), (iii), and / or (iv), The method, wherein the sample is the solution (2) or the product (3), further comprising carrying out the steps of the method according to claim 1 or 2.

18. 1. An apparatus (1) comprising a light source (4), a spectrometer (7) for measuring fluorescence radiation (6), and a device (18) adapted to perform the method of claim 1 or 2 on the basis of said fluorescence radiation (6).

Images