Configurable hand-held bioanalyzer for identifying biological products

Abstract

Configurable handheld bioanalyzers and related bioanalysis methods for Raman spectroscopy-based identification of biological products using integrated artificial intelligence (A1) are described. The bioclassification integrated model configuration (103) is loaded into a computer memory (108) of a configurable handheld bioanalyzer (102) having a processor (110) and a scanner (106). The biological ensemble classification model configuration includes a biological classification ensemble model having an unsupervised model and a supervised model. The processor is configured to receive a Raman-based spectral data set defining a biological product sample as scanned by the scanner and execute a spectral pre-processing algorithm configured by a biological ensemble classification model to reduce spectral deviations of the Raman-based spectral data set. The biological ensemble classification model identifies a biological product type based on a Raman-based spectral data set.

Description

Cross-references to related applications

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 463,187 (filed May 1, 2023), which is incorporated herein by reference in its entirety. Technical Field

[0002] This disclosure generally relates to configurable handheld bioanalyzers, and more specifically, to systems and methods for identifying or classifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy with configurable handheld bioanalyzers. Background Technology

[0003] The development and manufacture of pharmaceutical and biotechnology products often require the measurement or identification of raw materials used in their development. The purpose of identification testing is to ensure the product's identity. Situations requiring identification testing include product distribution to clinical sites, import testing, and transfers between network sites. Additionally, the measurement or identification of biological products may be important for ensuring the quality of the development or manufacturing process and the quality of the final product itself, in order to meet quality standards and / or regulatory requirements.

[0004] Using Raman spectroscopy to measure and identify biological products is a relatively new concept. Generally, Raman spectroscopy can be used to probe the chemical or biological structure of raw materials or products. Raman spectroscopy is a non-destructive chemical or biological analytical technique that measures the interaction of light with a product or material, such as the interaction of light with the biological properties or chemical bonds of the product or material. Raman spectroscopy provides a light scattering technique in which the molecules of a sample material or product are scattered by incident light from a high-intensity laser source. Typically, most of the scattered light has the same wavelength (color) as the laser source and does not provide useful information—this is called Rayleigh scattering. However, a small amount of light is scattered at a different wavelength (color), which is caused by the chemical or molecular structure of the material or product being analyzed—this is called Raman scattering, and this scattered light can be analyzed or scanned to generate Raman-based data on the material or product being analyzed.

[0005] Raman scattering analysis can yield detailed information about the properties of a material or product, including its chemical structure and / or identity, contaminants and impurities, phases, crystallinity, intrinsic stress / strain, and / or molecular interactions. This detailed information can be found in the material's Raman spectrum. Raman spectra can be visualized to show multiple peaks across various wavelengths. The Raman spectrum can show the intensity and wavelength position of the Raman scattered light. Each peak can correspond to a specific molecular bond vibration associated with the material or product being analyzed.

[0006] Typically, Raman spectroscopy provides a unique chemical or biological "fingerprint" for a specific material, molecule, or product, and can be used to verify the identity of a particular material, molecule, or product—and / or distinguish it from other substances. Additionally, Raman spectroscopy libraries—usually compilations of Raman spectra of many different materials—are commonly used to identify materials based on their Raman spectra. That is, a Raman spectroscopy library can be searched to find a Raman spectrum that matches the Raman spectrum of a given material or product being measured, thereby identifying the given material or product.

[0007] Currently, there are analyzers that use Raman spectroscopy to identify raw materials and products. For example, Thermo Fisher Scientific Inc. offers a handheld Raman-based analyzer, labeled the TruScan™ RM Handheld Raman Analyzer. However, using such existing scanners can present problems when dealing with materials and / or products with similar Raman spectra, such as pharmaceutical and biotechnology materials or products with similar Raman spectra. For example, deviations between the Raman spectra of similar products can cause existing handheld Raman-based analyzers to fail to identify them correctly, such as outputting Type 1 errors (false positives) or Type II errors (false negatives) for pharmaceutical or biotechnology products. The primary sources of deviation or error are differences between Raman-based analyzers, including differences such as variability in software, manufacturing, age, components (multiple components), operating environment (e.g., temperature), or other such differences between Raman-based analyzers.

[0008] Known methods often fail to address errors caused by biases or variability between handheld analyzers. For example, one known method involves developing static mathematical equations using data from several analyzers for use across them. However, the difficulty with this approach generally lies in the fact that instrument performance can change over time. Often, routine access to all these instruments is impractical or impossible. In particular, data for constructing static mathematical equations is often unavailable, especially for new analyzers, as manufacturers may not provide new specifications in advance. This hinders the development and maintenance of static mathematical equations, especially as new analyzers are developed over time, and given that the development of static mathematical equations typically requires a large number of samples to keep different analyzers accurate. Furthermore, without such new specifications for new analyzers, static mathematical equations may be incompatible when executed on the new analyzer. Additionally, differences in analyzer manufacturing or quality control (especially between different manufacturers) can lead to static mathematical equations being overly tolerant of variability, resulting in static mathematical equations themselves being too variable for the accurate measurement and / or identification of biological products.

[0009] In the second known approach, data from a given analyzer is standardized, where a child-to-parent instrument graph is created for a given set of analyzers. However, this approach is limited because constructing a child-to-parent instrument graph typically requires data from both the parent and child instruments, which is often difficult to implement or maintain and / or computationally expensive, especially over extended periods when several generations of analyzers have been developed, necessitating numerous permutations and types of child-to-parent instrument graphs. Furthermore, in the biopharmaceutical industry, user access to child instruments is restricted, further limiting the child-to-parent instrument graph approach. Additionally, biopharmaceutical manufacturing is subject to regulatory constraints, such as GMP requirements, which may necessitate revalidation of the child-to-parent transfer graph. Such revalidation can be time-consuming and resource-intensive.

[0010] In the third known method, data from a given analyzer are also standardized, but variability between analyzers is ignored or considered negligible. However, this method is not ideal because analyzer-to-analyst variability often affects the accurate identification and measurement of raw materials and / or biological products and should therefore be taken into account.

[0011] In a fourth approach, a trained model can be used to identify biological products based on Raman spectroscopy. This method is described in WO 2021 / 081263, filed on October 23, 2020, entitled “Configurable Handheld Biological Analyzers for Identification of Biological Products based on Raman Spectroscopy,” application number PCT / US2020 / 056961.

[0012] For the reasons stated above, there is a need for systems and methods for identifying or classifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy with configurable handheld bioanalyzers. These systems and methods are configured to reduce variability and improve compatibility among similarly configured configurable handheld bioanalyzers compared to known solutions. Summary of the Invention

[0013] The disclosure of this application describes the identification of biological products using Raman spectroscopy via handheld analyzers(s). Furthermore, the disclosure describes the use of configurable handheld bioanalysts, systems, and methods to overcome limitations typically associated with known methods for measuring biological products using Raman spectroscopy. For example, the Raman spectra of some biological products may be too similar to be distinguished using known methods of Raman spectroscopy, which often rely on generalized statistical algorithms. Raman spectroscopy measurements can be particularly problematic when instrument-to-instrument variability exists, leading to, for example, Type I and Type II errors between various analyzers. As described herein, such variability can be caused by any one or more differences in software, manufacturing, age, components, operating environment (e.g., temperature), or other differences in Raman-based analyzers. This problem is particularly evident during the development or manufacture of biological products, as analyzer-to-instrument variability can be a critical factor affecting the quality, robustness, and / or transferability of manufacturing or development processes related to pharmaceutical or biological products. Therefore, in the various embodiments disclosed herein, for example, configurable handheld bioanalyzers are described that utilize configurations using specific preprocessing algorithms and / or multivariate data analysis to (1) ensure that measurements and / or identification of materials or products are sensitive and / or explicit, and (2) ensure that compatibility and configurations developed on a first set of analyzers are transferable and / or implementable to additional analyzers, such as a “network” of analyzers or a new analyzer within a set of analyzers.

[0014] Therefore, in various embodiments herein, a configurable handheld bioanalyzer for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy is disclosed. The configurable handheld bioanalyzer may include a first housing adapted for handheld manipulation and a first scanner carried by the first housing. The configurable handheld bioanalyzer may further include a first processor communicatively coupled to the first scanner. The configurable handheld bioanalyzer may further include a first computer memory communicatively coupled to the first processor. In various aspects, the first computer memory may be configured to load a biological ensemble classification model configuration. The biological ensemble classification model configuration may include a biological classification ensemble model comprising an unsupervised model and a supervised model. The unsupervised model may be trained using Raman-based spectral training data to configure the unsupervised model to output a first indicator of one or more biological product types. The supervised model may be trained using Raman-based spectral training data to configure the supervised model to output a second indicator of one or more biological product types. Furthermore, the biological classification ensemble model configuration may include one or more spectral preprocessing algorithms. The first processor may be configured to perform one or more spectral preprocessing algorithms to reduce spectral bias in the first Raman-based spectral dataset upon receiving it. The biological classification ensemble model may be further configured to execute on the first processor, wherein the first processor is configured to (1) receive the first Raman-based spectral dataset defining a first biological product sample as scanned by the first scanner, and (2) use the biological classification ensemble model to identify the biological product type among the one or more biological product types based on the first Raman-based spectral dataset.

[0015] In another embodiment disclosed herein, a bioanalytical method for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy is disclosed. This bioanalytical method may include: loading a biological ensemble classification model configuration into a first computer memory of a first configurable handheld bioanalyzer having a first processor and a first scanner. The biological ensemble classification model configuration may include a biological classification ensemble model comprising an unsupervised model and a supervised model. The unsupervised model may be trained using Raman-based spectral training data to configure the unsupervised model to output a first indicator of one or more biological product types. Further, the supervised model may be trained using Raman-based spectral training data to configure the supervised model to output a second indicator of one or more biological product types. The bioanalytical method may further include: receiving at the first processor a first Raman-based spectral dataset, such as that scanned by the first scanner, defining a first biological product sample. The bioanalytical method may further include: the first processor executing one or more spectral preprocessing algorithms, such as those specified by the biological ensemble classification model configuration, to reduce spectral bias in the first Raman-based spectral dataset. The bioanalytical method may further include: using the biological classification ensemble model based on the first Raman-based spectral dataset to identify biological product types.

[0016] In a further embodiment disclosed herein, a tangible, non-transitory computer-readable medium (e.g., computer memory) is described for storing instructions for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy. When executed by one or more processors of a configurable handheld bioanalyzer, these instructions can cause the one or more processors of the configurable handheld bioanalyzer to load a biological integrated classification model configuration into a first computer memory of a first configurable handheld bioanalyzer having a first processor and a first scanner. The biological integrated classification model configuration can include a biological classification ensemble model comprising an unsupervised model and a supervised model. The unsupervised model can be trained using Raman-based spectral training data to configure the unsupervised model to output a first indicator of one or more biological product types. Further, the supervised model can be trained using Raman-based spectral training data to configure the supervised model to output a second indicator of one or more biological product types. Upon execution, these instructions can further cause one or more processors at a first processor to receive a first Raman-based spectral dataset, such as that defined by a first biological product sample scanned by a first scanner. These instructions, when executed, can further cause one or more processors to execute one or more spectral preprocessing algorithms, as specified by the biological ensemble classification model configuration, to reduce spectral bias in the first Raman-based spectral dataset. These instructions, when executed, can further cause one or more processors to utilize the biological ensemble model to identify biological product types based on the first Raman-based spectral dataset.

[0017] The benefits of this application include the development of multiple bioense classification models (e.g., multiple multivariate analysis models) that produce consistent results for the same pharmaceutical or biological product (e.g., therapeutic product / drug) across different analyzers (including different analyzers used to scan Raman-based datasets for constructing bioense classification models). As described herein, multiple analyzers or multiple Raman spectral datasets generated by such analyzers can be used to construct bioense classification models.

[0018] Furthermore, as described herein, the integrated biological classification model is configurable and transferable between configurable handheld bioanalysts, and can include Raman spectroscopy preprocessing, integrated model linking, and discriminant statistical analysis to reduce variability between configurable handheld bioanalysts. For example, the use of the integrated biological classification model described herein is an improvement over existing analyzers because it reduces instrument / analyst variability, can be developed without sub-instrument data, and can be used between different analyzers implementing different software, having different software or software versions, different manufacturing processes, service lives, operating environments (e.g., temperature), hardware, or other such differences.

[0019] Furthermore, the accuracy of biological ensemble classification models can be improved by applying preprocessing techniques (e.g., spectral preprocessing algorithms as described herein) to minimize statistical Type I and / or Type II errors in the output of the biological ensemble classification model, and thus improve the output of (multiple) configurable handheld bioanalysts on which the biological ensemble classification model is mounted / configured.

[0020] Additionally, in some embodiments, the configurable handheld bioanalysts can use integrated bioclassification models to distinguish between bioproducts / drugs with similar protein structures, protein concentrations, and / or formulations. This provides a flexible approach because integrated bioclassification models can be generated using various, different, and / or additional classification and predictive modeling techniques to correspond to products with multiple specifications (e.g., products related to Dinosumab).

[0021] Based on the foregoing and the disclosure herein, this disclosure includes improvements to computer functionality or other technologies, at least because the claims describe, for example, a configurable handheld bioanalyzer for identifying biological products based on Raman spectroscopy, which is an improvement over existing handheld bioanalyzers. That is, this disclosure describes improvements to the functionality of the computer itself or "any other technology or technical field," because the configurable handheld bioanalyzer is a computing device as described herein and provides reduced analyzer and analyzer variability compared to existing handheld bioanalyzers via its integrated biological classification model configuration. This is an improvement over the prior art, at least because the configurable handheld bioanalyzer described herein provides increased accuracy in the measurement, identification, and / or classification of materials and / or products (e.g., therapeutic products), a crucial feature in the manufacture and development of pharmaceutical and biological products.

[0022] Furthermore, the configurable handheld bioanalysts described herein are further improved by using a biological integrated classification model configuration that is transferable, optionally updatable (utilizing new data), and loadable into the memory of compatible (multiple) configurable handheld bioanalysts. This enables standardization, thereby reducing variability between analyzer sets or groups (i.e., analyzer "networks"). This reduces maintenance and / or deployment time for configurable handheld bioanalysts used in analyzer networks.

[0023] Furthermore, the configurable handheld bioanalyst is further improved by using a biointegrated classification model configuration that includes a biointegrated classification model. As described herein, the biointegrated classification model improves the accuracy of identification and / or classification of biological products by eliminating or reducing Type I errors (e.g., false positives) and / or Type II errors (e.g., false negatives).

[0024] Additionally, this disclosure includes the application of certain claim elements using a specific machine or by using a specific machine (e.g., a configurable handheld bioanalyzer for identifying biological products (including identifying biological products during the development or manufacture of such products) using integrated AI based on Raman spectroscopy).

[0025] Furthermore, this disclosure includes the ability to transform or restore specific items to different states or things, for example, transforming or restoring a Raman spectroscopy dataset to different states for identifying biological products based on Raman spectroscopy.

[0026] This disclosure includes specific features beyond those well-known routine activities in the art, or adds unconventional steps that limit the claims to specific useful applications. For example, it includes providing a biological integrated classification model configuration for reducing variability among a group or batch of configurable handheld bioanalyzers (i.e., a “network” of configurable handheld bioanalyzers), wherein each configurable handheld bioanalyzer can be used to identify biological products based on Raman spectroscopy. The methods and systems described herein can detect and distinguish product types with similar Raman spectral datasets or other similar Raman-related features that cannot be distinguished by conventional systems and methods.

[0027] The advantages will become more apparent to those skilled in the art from the following description of the preferred embodiments shown and described by way of illustration. As will be appreciated, embodiments of the invention may have other and different embodiments, and their details may be modified in various aspects. Therefore, the drawings and description are to be regarded in an illustrative rather than restrictive manner. Attached Figure Description

[0028] The accompanying drawings described below depict various aspects of the systems and methods disclosed therein. It should be understood that each drawing depicts an embodiment of a specific aspect of the disclosed systems and methods, and each of the drawings is intended to be consistent with its possible embodiments. Furthermore, wherever possible, the following description refers to reference numerals included in the following drawings, wherein features depicted in the plurality of drawings are indicated by consistent reference numerals.

[0029] The arrangement currently under discussion is illustrated in the accompanying drawings; however, it should be understood that embodiments of the invention are not limited to the precise arrangement and tools shown, wherein:

[0030] Figure 1 An example configurable handheld bioanalyzer for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, according to various embodiments disclosed herein, is shown.

[0031] Figure 2A Example biological classification ensemble models for identifying biological products using integrated AI based on Raman spectroscopy, according to various embodiments disclosed herein, are shown.

[0032] Figure 2B Another example flowchart is shown, illustrating a bioanalytical method for identifying biological products using integrated AI-based Raman spectroscopy, according to various embodiments disclosed herein.

[0033] Figure 3A Example visualizations of Raman-based spectral datasets, such as those scanned by various handheld bioanalyzers, are shown according to various embodiments disclosed herein.

[0034] Figure 3B Examples of various embodiments disclosed herein are shown. Figure 3A An example visualization of the modified Raman-based spectral dataset, derived from the original Raman-based spectral dataset.

[0035] Figure 3C Example visualizations of normalized Raman-based spectral datasets according to various embodiments disclosed herein are presented as... Figure 3B The normalized version of the modified Raman-based spectral dataset.

[0036] Figure 4 Example visualizations of Raman-based spectral datasets of mAb 1 (typical IgG2 monoclonal antibody) drug products (DP) and mAb 2 (typical IgG1 monoclonal antibody) DP scanned by a handheld bioanalyzer according to various embodiments disclosed herein are presented.

[0037] Figure 5A This demonstrates when a Raman-based spectral dataset (including...) is provided. Figure 4 Example visualization of Q-residuals of non-ensemble unsupervised biological classification models when those datasets are used as input.

[0038] Figure 5B This demonstrates various embodiments of the present paper when Raman-based spectral datasets (including...) are provided. Figure 4 Example visualizations of the predicted outputs of a supervised biological classification ensemble model when those datasets are used as input.

[0039] Figures 6A to 6C A list of example computer programs is provided showing pseudocode for a biological ensemble classification model configuration according to various embodiments disclosed herein, the biological ensemble classification model configuration including the configuration of the unsupervised portion of the biological classification ensemble model.

[0040] Figures 7A to 7C A list of example computer programs is provided showing pseudocode for a biological ensemble classification model configuration according to various embodiments disclosed herein, the biological ensemble classification model configuration including the configuration of a supervised portion of the biological classification ensemble model.

[0041] The accompanying drawings depict preferred embodiments for illustrative purposes only. Alternative embodiments of the systems and methods shown herein may be employed without departing from the principles of the invention described herein. Detailed Implementation

[0042] Figure 1 An example configurable handheld bioanalyst 102 is shown, according to various embodiments disclosed herein, for identifying biological products 140 using integrated artificial intelligence (AI) based on Raman spectroscopy. Figure 1 In one embodiment, the configurable handheld bioanalyzer 102 includes a first housing 101 molded or otherwise adapted for handheld manipulation. Additionally, the configurable handheld bioanalyzer 102 includes a first scanner 106 carried by the first housing (e.g., such as being directly or indirectly coupled to or connected to the first housing). The configurable handheld bioanalyzer 102 also includes a first processor 110 communicatively coupled to the first scanner 106. The configurable handheld bioanalyzer 102 may further include a first computer memory 108 communicatively coupled to the first processor 110. Furthermore, the configurable handheld bioanalyzer 102 may include an input / output (I / O) component 109 for receiving input from a navigation wheel 105. For example, a user can manipulate the navigation wheel 105 to select or scroll data or information of a specific sample of a biological product, such as data or information scanned from a scanning biological product 140. The input / output (I / O) component 109 may also control the display of measurement information, identification information, classification information, or other information as described herein on a display screen 104. Each of the display screen 104, navigation wheel 105, first scanner 106, first computer memory 108, I / O component 109, and / or first processor 110 is communicatively coupled via an electronic bus 107 configured to send and / or receive electronic signals (e.g., control signals) or information between the various components (including 104 to 110). In some embodiments, the configurable handheld bioanalyzer 102 may be a Raman-based handheld analyzer, such as the TruScan™ RM handheld Raman analyzer supplied by ThermoFisher Scientific Inc.

[0043] In various embodiments, the first computer memory 108 is configured to load a biological ensemble classification model configuration, such as biological ensemble classification model configuration 103. Biological ensemble classification model configuration 103 can be used for implementation. Figure 2A and / or Figure 2B Bioanalytical methods for identifying biological products based on Raman spectroscopy, as further described herein.

[0044] In another embodiment, the computer memory is configured to load a new biological classification ensemble model. The new biological classification model may include an updated unsupervised model and / or an updated supervised model trained on a new and / or updated Raman spectroscopy dataset.

[0045] exist Figure 1 In this embodiment, the biological ensemble classification model configuration 103 is implemented as an XML file in Extensible Markup Language (XML) format. As described in the various embodiments herein, Figures 6A to 6CA list of example computer programs is shown, including several code sections 602, 650, and 675, which include pseudocode for a biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) in XML format. Figures 6A to 6C This includes example pseudocode for configuring the unsupervised portion of an ensemble model for biological classification. Similarly, Figures 7A to 7C Several code sections, including code segments 702, 750, and 775, demonstrate example pseudocode for configuring a biological ensemble classification model, which includes the configuration of the supervised portion for the biological classification ensemble model. For example, in... Figures 6A to 6C In the code segment 1 of the embodiment computer program list, the biological ensemble classification model configuration 103 is in XML format, wherein the biological ensemble classification model (“ <model>) or parts thereof (e.g., Figure 2A The unsupervised model 202m is defined in the biological ensemble classification model configuration 103. Similarly, in Figures 7A to 7C In the code segment 1 of the embodiment's computer program list, the biological ensemble classification model configuration 103 is in XML format, wherein the biological ensemble classification model (" <model>) or parts thereof (e.g., Figure 2A The supervised model 204m is defined in the integrated biological classification model configuration 103. The integrated biological classification model configuration 103 can be transferred, installed, and / or otherwise implemented or executed on similarly configured configurable handheld bioanalysts (e.g., configurable handheld bioanalysts 112, 114, and / or 116). It should be understood that... Figures 6A to 6C and Figures 7A to 7C The pseudocode can be combined into a single file, or alternatively, can be a separate file. Such files(s) can be stored, linked, or / or otherwise stored or referenced for access by one or more processors (as described herein) via a biological integrated classification model configuration 103, which can be transferred, installed, and / or otherwise implemented or executed on a similarly configured configurable handheld bioanalyst (e.g., configurable handheld bioanalysts 112, 114, and / or 116).

[0046] Each of the configurable handheld bioanalysts 112, 114, and 116 includes the same components as the configurable handheld bioanalyst 102; therefore, the disclosure of the configurable handheld bioanalyst 102 also applies to each of the configurable handheld bioanalysts 112, 114, and 116. Each of the configurable handheld bioanalysts 102, 112, 114, and 116 may be part of the same analyzer group or set (i.e., including an analyzer "network" or group). In some embodiments, each of the configurable handheld bioanalysts 102, 112, 114, and / or 116 may have the same or similar characteristics or features and / or different mixtures of characteristics or features, such as the same or similar software version or type(s), manufacture(s), service life(s), operating environment(s), components(s), or other such similarities or differences and / or different mixtures of these characteristics or features in Raman-based analyzers.

[0047] Regardless of the similar or identical characteristics or features and / or different mixtures of characteristics or features among the configurable handheld bioanalysts 102, 112, 114, and 116, the biointegrated classification model configuration 103 and its associated biointegrated classification model allow the network of configurable handheld bioanalysts (e.g., configurable handheld bioanalysts 102, 112, 114, and 116) to produce consistent results when measuring or identifying pharmaceutical or biological products (e.g., therapeutic products / drugs). That is, despite the similarities or differences among a given analyzer network of configurable handheld bioanalysts, when such a configurable handheld bioanalyst is configured with a biointegrated classification model configuration as described herein, it can accurately identify or measure a given pharmaceutical or biological product.

[0048] In various embodiments, multiple analyzers can be used to generate or construct the biological ensemble classification model configuration 103 and its associated biological ensemble classification model. For example, in some embodiments, any one or more of configurable handheld biological analyzers 102, 112, 114, and 116 and / or other analyzers (not shown) can be used to generate or construct the biological ensemble classification model.

[0049] The generation of the integrated biological classification model configuration 103 and its associated integrated biological classification model typically requires a set of analyzers or a network of analyzers to scan samples (e.g., of biological product 140) to generate Raman-based spectral datasets of those samples. For example, scanning biological product 140 using any of the configurable handheld bioanalyzers 102, 112, 114, and 116 can generate detailed information about biological product 140. This detailed information may include, for example, multiple Raman-based spectral datasets defining (e.g., of biological product 140) samples of (multiple) biological product samples. Examples of biological product 140 may include any of mAb 3 DP, mAb 2 active pharmaceutical ingredient (DS), mAb 1 DP, and / or other substances described herein. However, it should be understood that additional biological products are considered herein, and biological product 140 is not limited to any particular biological product or its grouping.

[0050] In some embodiments, the configurable handheld bioanalyzer 102 can define instrument- or analyzer-based spectral acquisition parameters (e.g., integration time, laser power, etc.) for scanning samples (e.g., biological product 140). For example, a user can select spectral acquisition parameters for scanning samples via navigation wheel 105. In some embodiments, the configurable handheld bioanalyzer 102 can generate an output file (e.g., an output file of type ".acq") specifying the spectral acquisition parameters.

[0051] In some embodiments, a configurable handheld bioanalyst (e.g., configurable handheld bioanalyst 102) can load a file (e.g., a ".acq" file) to configure the configurable handheld bioanalyst for scanning target products using spectral acquisition parameters. As described herein, one or more configurable handheld bioanalysts (e.g., configurable handheld bioanalyst 102) can scan multiple Raman-based spectral datasets to generate a biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103). In some embodiments, multiple samples of biological products (e.g., biological product 140) can be selected (e.g., multiple batches) as representative target products for scanning. Generally, as described herein, a "target product" refers to a biological product used to train or otherwise configure the biological ensemble classification model configuration and its associated models. Generally, target products are selected based on their biological specifications. Once the spectral acquisition parameters are set for scanning the target product, a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) can (e.g., using a first scanner 106) scan a sample of the target product, in some cases multiple times (e.g., fourteen (14) times), where each scan generates detailed information, including multiple Raman-based spectral datasets of the target product.

[0052] In similar embodiments, multiple configurable handheld bioanalysts (configurable handheld bioanalysts 102, 112, 114, and / or 116) can load output files (e.g., ".acq" files) to configure each configurable handheld bioanalyst for scanning biological product samples using spectral acquisition parameters. Once configured, each configurable handheld bioanalyst (e.g., any one of configurable handheld bioanalysts 102, 112, 114, and / or 116) is configured to (e.g., using the first scanner 106) scan the sample, in some cases multiple times (e.g., fourteen (14) times), wherein each scan generates detailed information about the target product, including (multiple) Raman-based spectral datasets. By utilizing different / multiple scanners to scan a given target product, the (multiple) Raman-based spectral datasets captured by these scanners become robust because the (multiple) Raman-based spectral datasets capture any differences between scanners (e.g., caused by software, manufacturing, age, operating environment (e.g., temperature), etc.). In this way, multiple Raman-based spectral datasets provide ideal training datasets for reducing variability among multiple scanners as described herein. For example, each of the multiple Raman-based spectral datasets scanned by multiple scanners (e.g., any one of configurable handheld bioanalyzers 102, 112, 114, and / or 116) can be output and / or saved as a Raman spectral file, for example, of type ".spc".

[0053] It should be understood that multiple Raman-based spectral datasets can also be captured for the challenge product in the same or similar manner as for the target product. As used herein, "challenge product" describes a biological product (e.g., selected from biological product 140) that is configured to be identified, classified, or measured when a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) is loaded or otherwise configured with a biological integrated classification model configuration as described herein (e.g., biological integrated classification model configuration 103) and its associated biological integrated classification model.

[0054] Multiple Raman-based spectral datasets of a challenge product can be captured in the same or similar manner as the target product, wherein the challenge product can be selected based on its biological specifications, and wherein a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) can load an output file (e.g., a ".acq" file) to configure the configurable handheld bioanalyzer for scanning the challenge product using spectral acquisition parameters. Once set up, the configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) is configured to (e.g., using a first scanner 106) scan a sample of the challenge product, in some cases multiple times (e.g., three (3) times), wherein each scan generates detailed information about the challenge product, including multiple Raman-based spectral datasets. For example, multiple Raman-based spectral datasets scanned by configurable handheld bioanalyzer 102 can be output and / or saved as Raman spectral files, for example, of type ".spc".

[0055] In some embodiments, it can be achieved through, such as Figure 1 The processor of the computer 130 shown is used as a remote processor to perform the generation of a biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103). For example, multiple Raman-based spectral datasets generated for biological products (e.g., selected from biological product 140) as described herein can be imported into and / or analyzed by modeling software executed on computer 130 and configured to analyze multiple Raman-based spectral datasets. An example of such modeling software includes SOLO (Standard Chemometrics Software) provided by Eigenvector Research, Inc. However, it should be understood that other modeling software implemented to perform the features described herein can also be used, including custom or proprietary software. The modeling software can build or generate a biological ensemble classification model based on multiple Raman-based spectral datasets. For example, in some embodiments, as described herein, multiple Raman-based spectral datasets scanned or captured for multiple target or challenge products can be used to build or generate a biological ensemble classification model. Furthermore, multiple Raman-based spectral datasets (e.g., for target or challenge products) can be used to cross-validate biological ensemble classification models. For example, multiple Raman-based spectral datasets can be used against cross-validation datasets of multiple Raman-based spectral datasets to evaluate Type I errors (e.g., false positives) and Type II errors (e.g., false negatives) of biological ensemble classification models.

[0056] In various embodiments, a biointegrated classification model and / or its associated biointegrated classification model configuration (e.g., biointegrated classification model configuration 103) can be generated to include algorithms (e.g., scripts) and parameters to be used by a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) for the identification, classification, and / or measurement of biological products as described herein. Figure 2A , Figure 2B , Figures 6A to 6C and Figures 7A to 7C Examples of algorithms (e.g., scripts) and / or parameters are described. For example, a bio-ensemble classification model configuration (e.g., bio-ensemble classification model configuration 103) may include parameters that define the details of the bio-ensemble classification model. For example, such parameters may include the number of classification components, loadings, etc., of the bio-ensemble classification model. The term "classification component" as used herein may include principal components determined by principal component analysis (PCA). More generally, in other embodiments, classification components may be coefficients or variables of a multivariate model (such as a regression model or a machine learning model). The bio-ensemble classification model is configured to identify the bio-product type of a given bio-product sample (e.g., selected from bio-product 140) based on classification components. For example, in one embodiment, the number of classification components may be determined, for example, by modeling software via singular value decomposition (SVD) analysis, where the classification components include one or more principal components of the PCA. The PCA implementation indicates the use of multivariate analysis (e.g., as implemented by a configurable handheld bioanalyst 102 configured with bio-ensemble classification model configuration 103) to distinguish bio-products (e.g., bio-product 140), such as therapeutic products / medications with similar formulations (e.g., as described herein for...). Figure 5A and Figure 5B (As described herein). For example, biological or pharmaceutical products are often associated with high-dimensional data. High-dimensional data can include multiple features, such as the expression of many genes measured on a given sample (e.g., a sample of the biological product 140). PCA provides a technique, as used by a configurable handheld bioanalyzer 102, for simplifying the complexity of high-dimensional data (e.g., multiple Raman spectroscopy datasets) while preserving trends and patterns useful for predictive and / or identification purposes (e.g., identification of biological products as described herein). For example, the application of PCA includes (e.g., by a first processor 110) transforming a dataset (e.g., a Raman-based spectroscopy dataset) into fewer dimensions. The transformed, less-dimensional dataset provides a summary or simplification of the original dataset. Furthermore, the transformed dataset reduces computational costs when manipulated by a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) as described herein. Further, as described herein, implementing PCA can also reduce the error rate(s), thereby eliminating the need to apply multiple test corrections to higher-dimensional data when testing the association of each feature with a specific result.

[0057] Furthermore, PCA, as implemented by the configurable handheld bioanalyzer 102, reduces data complexity by geometrically projecting the data onto a lower dimension called principal components (PCs) and by using a finite number of PCs for optimal data summarization (and therefore PCs). The first PC is selected to minimize the total distance between the data and its projection onto the PC. Similarly, any second (subsequent) PCs are selected, where it is further required that they are unrelated to all previous PCs.

[0058] PCA is an unsupervised learning method and is similar to clustering—it finds trends or patterns without relying on prior knowledge about whether samples come from different sources (such as different configurable handheld bioanalysts (e.g., configurable handheld bioanalysts 102, 112, 114, and / or 116)). For example, in some embodiments, the taxonomical components of a bio-ensemble classification model can be the first principal components of a PCA model. In such embodiments, the first principal components can be determined by the first processor 110 based on singular value decomposition (SVD) analysis. The use of the first principal components by the configurable handheld bioanalyst 102 limits or reduces the amount of analyzer variability considered by its bio-ensemble classification model. In some embodiments, the first principal component (PC) can be a unique principal component. In other embodiments, the bio-ensemble classification model can include a second taxonomical component, wherein the bio-ensemble classification model is configured to identify the (multiple) bio-product types of a given bio-product sample (e.g., bio-product 140) based on multiple taxonomical components (the first taxonomical component and the second taxonomical component).

[0059] Modeling software can be configured to set statistical confidence levels to determine taxonomic components (e.g., principal components) to be included in or otherwise used by a biological ensemble classification model. For example, in... Figures 6A to 6C In an embodiment of the computer program list, at code segment 1, the biological ensemble classification model configuration indicates the biological ensemble classification model (e.g., defined as "..."). <model>This includes PCA-type ensemble classification models of organisms. This indicates that the classification components of an ensemble classification model will include principal components. For example, in... Figures 6A to 6C In one embodiment, code segment 2 indicates that the number of principal components to be determined via SVD analysis ("algorithm: SVD") to be performed on the first processor 110 of the configurable handheld bioanalyzer 102 will be a (single) principal component ("PC number: 1"). Similarly, modeling software can be configured to set statistics to determine equations (e.g., best-fit or least-squares linear equations) to be included in or otherwise used by the bioensemble model. For example, in Figures 7A to 7C In code segment 1 of the embodiment's computer program list, the biological ensemble classification model configuration instruction (e.g., defined as "...") <model>This includes biological ensemble classification models of the PLSDA type. This indicates that the classification values ​​of biological ensemble classification models will include PLSDA values. For example, in... Figures 7A to 7C In one embodiment, code segment 2 indicates that the model includes a PLSDA model with specific "axis units", Y block values, and linear values ​​("LV number: 1"), which will be determined via, for example, a PLSDA model to be executed on a first processor 110 of a configurable handheld bioanalyzer 102.

[0060] As another example, a biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) may include computer code or scripts for defining or implementing (multiple) spectral preprocessing algorithms, such as those described above. Figures 3A to 3C As described herein. For example, a first processor (e.g., first processor 110) may be configured to execute one or more spectral preprocessing algorithms to reduce spectral bias in the first Raman-based spectral dataset upon receiving it. More generally, computer code or scripts for defining or implementing the spectral preprocessing algorithms(s) may be executed on the processor (e.g., first processor 110), wherein the processor receives the Raman-based spectral dataset(s) of a biological product (e.g., biological product 140). A handheld bioanalyzer may then be configured to execute the computer code or scripts defining or implementing the spectral preprocessing algorithms(s) to prepare / preprocess the data for input into the taxonomic components(s) of a biological integrated taxonomy model, thereby identifying, measuring, or classifying the biological product (e.g., a challenge product) as described herein. For example, in Figures 6A to 6C In code segment 2 of the embodiment's computer program list, the biological ensemble classification model configuration includes an execution sequence of an example spectral preprocessing algorithm (e.g., "Preprocessing: First Derivative (Order: 2, Window: 21 pt, Include Only, Tail: polyinterp), SNV, Mean Centering"), which includes determining the first derivative on a Raman-based spectral dataset scanned for a specific product (e.g., target product or challenge product), applying the Standard Normal Variable (SNV) algorithm, and further applying the mean centering function. Figures 7A to 7C Code snippet 2 illustrates a similar embodiment of the PLSDA model. Further, this paper discusses... Figures 3A to 3C as well as Figures 6A to 6C and Figures 7A to 7C Code segments 4 to 6 describe and visualize example implementations of this execution sequence.

[0061] As another example, a biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) may include multiple Raman-based spectral datasets for generating the biological ensemble classification model. For example, in Figures 6A to 6C In code segment 3 of the embodiment's computer program list, the biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) includes a function for... Figures 6A to 6C Examples (multiple) of pseudocode-generated biological ensemble classification models or parts thereof are based on Raman spectral datasets. Similarly, in Figures 7A to 7C In code segment 3 of the embodiment's computer program list, the biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) includes a function for... Figures 7A to 7C Examples (multiple) of pseudocode-generated biological ensemble classification models or parts thereof are based on Raman spectral datasets.

[0062] In some embodiments, the biointegrated classification model configuration (e.g., biointegrated classification model configuration 103) may also define a threshold (e.g., as a statistical acceptance criterion) to determine whether the configurable handheld bioanalyzer 102 has successfully identified or measured a biological product. For example, such a threshold may define a pass / fail threshold for the Q residual value (e.g., as described herein for...). Figure 2A , Figure 5A and Figure 5B (as described herein), to determine whether the configurable handheld bioanalyzer 102 has successfully identified or measured a biological product. In other embodiments, thresholds can be configured independently of the biointegrated classification model configuration (e.g., biointegrated classification model configuration 103), for example, by a user manually configuring and / or defining thresholds via the navigation wheel 105 and display screen 104 described herein.

[0063] Once generated, the bio-ensemble classification model and its associated bio-ensemble classification model configuration (e.g., bio-ensemble classification model configuration 103) can be exported to a file (e.g., an XML file as described herein) to be transferred (e.g., via computer network 120 or other devices described herein) to a configurable handheld bioanalyzer (e.g., any one or more of configurable handheld bioanalyzers 102, 112, 114, and / or 116) and / or loaded into the memory of the configurable handheld bioanalyzer. In some embodiments, output files (e.g., ".acq" files as described herein) can also be transferred (e.g., via computer network 120 or other devices described herein) to a configurable handheld bioanalyzer (e.g., any one or more of configurable handheld bioanalyzers 102, 112, 114, and / or 116) and / or loaded into the memory of the configurable handheld bioanalyzer.

[0064] Biological ensemble classification models can be generated by a remote processor located away from a given configurable handheld bioanalyst. For example, in Figure 1 In some embodiments, computer 130 includes a remote processor located remotely from configurable handheld bioanalyzer 102. Computer 130 may generate (e.g., as described herein) and store one or more biointegrated classification model configurations and / or biointegrated classification models in database 132. In various embodiments, computer 130 may transfer (e.g., any one of biointegrated classification model configurations 103, 113, 115, and / or 117) to configurable handheld bioanalyzers (e.g., to configurable handheld bioanalyzers 102, 112, 114, and / or 116, respectively) via computer network 120. In some embodiments, each of biointegrated classification model configurations 103, 113, 115, and / or 117 may be a copy of the same file (e.g., the same XML file). Computer network 120 may include wired and / or wireless (e.g., 802.11 standard networks) implementing computer packet protocols (such as, for example, Transmission Control Protocol (TCP) / Internet Protocol (IP)). In other embodiments, the biological integrated classification model configuration (e.g., biological integrated classification model configuration 103) may be transferred via a Universal Serial Bus (USB) cable (not shown), a memory drive (e.g., flash memory or thumb drive) (not shown), a disk (not shown), or other transfer or storage device cables for transferring data files (such as the XML files disclosed herein). In further embodiments, the biological integrated classification model configuration 103 may be transferred via wireless standards or protocols such as Bluetooth, WiFi, or via cellular standards such as GSM, EDGE, CDMA.

[0065] A biointegrated classification model configuration (e.g., biointegrated classification model configuration 103) can be transferred between configurable handheld bioanalysts. Once transferred, the biointegrated classification model configuration can be loaded into the memory of the configurable handheld bioanalyst to calibrate or configure the configurable handheld bioanalyst with reduced variability relative to other configurable handheld bioanalysts implementing or executing the biointegrated classification model. For example, in one embodiment, biointegrated classification model configuration 103 may include a biointegrated classification model. The biointegrated classification model of biointegrated classification model configuration 103 may be configured to execute on a first processor 110. For example, the first processor 110 may be configured to (1) receive a first Raman-based spectral dataset of a first bioproduct sample as defined by a first scanner (e.g., scanning bioproduct 140), and (2) identify the bioproduct type based on the first Raman-based spectral dataset using the biointegrated classification model. For example, in some embodiments, the bioproduct type may be a therapeutic product having a therapeutic product type. Furthermore, in some embodiments, the bioproduct type may be identified by the biointegrated classification model during the manufacture of the bioproduct having a bioproduct type. The manufacture of such (multiple) biological products can include the processing and storage of pharmaceutical products as well as the production of bioreactors.

[0066] The biointegrated classification model of biointegrated classification model configuration 103 can be electronically transferred to the configurable handheld bioanalyzer 112 via, for example, through a computer network 120 via biointegrated classification model configuration 113. Like the configurable handheld bioanalyzer 102, the configurable handheld bioanalyzer 112 may include: a second housing adapted for handheld operation, a second scanner coupled to the second housing, a second processor communicatively coupled to the second scanner, and a second computer memory communicatively coupled to the second processor. The second computer memory of the configurable handheld bioanalyzer 112 is configured to load the biointegrated classification model configuration 113. The biointegrated classification model configuration 113 includes the biointegrated classification model of biointegrated classification model configuration 103. When implemented or executed on the second processor of the configurable handheld bioanalyzer 112, the second processor is configured to (1) receive a second Raman-based spectral dataset of a second biological product sample as defined by the second scanner of the second configurable handheld bioanalyzer 112 (e.g., obtained by scanning biological product 140), and (2) identify the biological product type based on the second Raman-based spectral dataset using the biological integrated classification model. In such an embodiment, the same biological product or product type can be identified by using the same biological integrated classification model as transferred through the biological integrated classification model configuration file, wherein the second biological product sample is a new sample of that biological product type (e.g., the same biological product type analyzed by the first configurable handheld bioanalyzer 102).

[0067] In various embodiments, new or additional Raman-based spectral datasets can be scanned by a configurable handheld bioanalyzer and used to update the ensemble biological classification model. In such embodiments, as described herein, the updated ensemble biological classification model can be transferred to a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102).

[0068] In some embodiments, the computer memory (e.g., first computer memory 108) of a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) can be configured to load a new bioencoded classification model, wherein the new bioencoded classification model may include updated taxonomic components. For example, new taxonomic components may be generated or determined for a new bioencoded classification model configuration 103, such as when received with a new bioencoded classification model configuration (e.g., bioencoded classification model configuration 103).

[0069] As described in the various embodiments herein, a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) can be configured by loading a biointegrated classification model configuration and its associated biointegrated classification model. Once configured, the configurable handheld bioanalyzer 102 can be used to identify, classify, or measure products of interest (e.g., challenge products and / or samples) as described herein.

[0070] Figure 2A An example biological classification ensemble model 200 for Raman spectroscopy-based identification of biological products is shown, according to various embodiments disclosed herein. Figure 2A As shown, the biological classification ensemble model 200 includes an unsupervised model 202 and a supervised model 204. In Figure 2A In the example, unsupervised model 202 is configured based on principal component analysis (PCA), while supervised model 204 is configured based on partial least squares discriminant analysis (PLSDA). The configured models are models trained on data (e.g., Raman spectral data), and the trained models can be used for prediction, classification, or other subsequent product identification purposes. In each respect, the models can be independent, where each model can be independently trained or otherwise configured on the same and / or different Raman-based spectral training data, and where each model has its own index output (e.g., pass / fail is independently determined and / or output). For example, this paper considers multiple models, such as two, three, or more models, each with its own index output.

[0071] A biological taxonomy ensemble model (e.g., biological taxonomy ensemble model 200) can be configured to identify the type of biological product when the first indicator passes a first pass / fail threshold and the second indicator passes a second pass / fail threshold. For example, as Figure 2A As illustrated in the example, the biological classification ensemble model 200 may include chained or otherwise sequential outputs, whereby one model passes its output to another model to identify (e.g., pass or fail) a specific target product (e.g., a pharmaceutical product). In this way, a bioconfigurable handheld bioanalyzer 102 on which the biological classification ensemble model 200 is loaded may be implemented, executed, or otherwise accessed to identify (e.g., pass or fail) a specific target product (e.g., a pharmaceutical product).

[0072] Unsupervised model 202 may include artificial models that perform principal component analysis (PCA), Euclidean distance or correlation, neighborhood-based training or implementation, K-means training or implementation, quality threshold (QT) training or implementation, centroid training or implementation, Ward and / or fuzzy C-means clustering and / or training thereon. In various aspects, unsupervised model 202 may be trained using Raman-based spectral training data to configure the unsupervised model to output a first indicator (e.g., pass / fail indicator) for one or more biological product types (e.g., one or more target products).

[0073] exist Figure 2A In the example, unsupervised model 202 is a PCA model 202m or other clustering model configured to classify Raman-based spectral data. For example, unsupervised model 202 may include a PCA model configured to implement a dimensionality reduction clustering algorithm that uses principal components to summarize important variability in the dataset. Although Raman spectra are often highly multicollinear, principal components can still be used to identify the target product. Specifically, a PCA model can be generated or constructed using the Raman spectra of the target product, using different sample batches and instruments, where several principal components (e.g., 1 to 2 principal components) can be identified that summarize important variability in the Raman spectra (typically instrument variability and batch-to-batch differences). The data used for identification may include column-based values ​​(e.g., variables in MVA literature (1000 or more intensity measurements for Raman)) and row-based values ​​(e.g., sample spectra, typically a few or dozens of samples). Unsupervised model 202 is configured to distinguish or otherwise identify products with slightly different protein concentrations, formulations, or other differences (e.g., 100 mg / mL mAb 2 versus 90 mg / mL mAb 3 (a typical IgG2 monoclonal antibody, which is different from mAb 2)).

[0074] Unsupervised model 202 is given Raman-based spectral data (e.g., related to the test or challenge product) as input and outputs a pass / fail indicator. The Raman-based spectral data may be data indicating a specific biological product (e.g., biological product 140), and if unsupervised model 202 fails to detect a specific biological product, it provides a FAIL output. Such a FAIL output may occur if the value produced by unsupervised model 202 is higher or lower than a threshold for a specific biological product, for example, as described in this paper. Figure 5A and 5B As described. In Figure 5A and 5B In the example shown, a Q residual value greater than 1 will result in a FAIL.

[0075] refer to Figure 2A If a FAIL result 202f is generated, no further analysis is required, and the biological classification ensemble model 200 may produce an unacceptable or negative output as the result. On the other hand, if a acceptable result is generated, the biological classification ensemble model 200 can initiate the execution of the supervised model 204.

[0076] The supervised model 204 may include artificial models that perform partial least squares discriminant analysis (PLSDA), linear discriminant analysis (LDA), k-nearest neighbor (KNN) analysis, soft independent modeling of analogies (SIMCA), and / or logistic regression discriminant analysis (LREGDA) and / or are trained thereon. In various aspects, the supervised model 204 may be trained using Raman-based spectral training data to configure the supervised model to output a second indicator (e.g., a pass / fail indicator) for one or more biological product types.

[0077] exist Figure 2A In the example, the supervised model 204 is a PLSDA model 204m, for example, a linear classification model configured to determine label values ​​indicating a particular product type. The PLSDA model 204m can perform or implement a dimensionality reduction classification algorithm that uses latent variables to summarize or otherwise detect significant variability in the dataset. The PLSDA model 204m can be trained to identify or select latent variables that maximize the covariance (i.e., Cov(X,Y)) between the data block (X) and the class label (Y) matrix. In various aspects, the X values ​​can include Raman spectra, for example, a data matrix (M × N) with approximately 30 × 2048 data values. The Y values ​​can include a class matrix (N × 1), where, for example:

[0078] If the sample is the target product, then Y i = 1; or

[0079] If the sample is not the target product, then Y i = 0

[0080] This data may include Raman-based spectral data of a target product and / or a challenge sample with similar values. This allows the supervised model 204 to discover latent variables that distinguish the target product from the challenge sample with similar values. In this way, the supervised model 204 is configured to differentiate or otherwise identify products with substantially the same formulation, protein concentration, molecular class, and / or other similar properties.

[0081] In some respects, multiple PLSDA models (not shown) can be used Figure 2A The biological classification ensemble model 200. For example, the biological classification ensemble model 200 can be configured or updated to have two supervised models (multi-class models). In this respect, the biological classification ensemble model 200 can be configured or updated to employ a smaller range of models to reduce the complexity of the dataset and provide simplified linear discrimination between the target Raman spectroscopy dataset and the challenge Raman spectroscopy dataset.

[0082] Further reference Figure 2A Similar to the unsupervised model 202, the supervised model 204 is given Raman-based spectral data (e.g., relevant to the test or challenge product) as input and outputs a pass / fail indicator. The Raman-based spectral data may be data indicating a specific biological product (e.g., biological product 140). If the supervised model 204 cannot identify or detect a specific biological product, it provides a FAIL output. This FAIL output may occur if the value produced by the supervised model 204 is below the threshold or confidence value for a specific biological product, for example, as described in this paper. Figure 5A and Figure 5B As described.

[0083] Further reference Figure 2A If a FAIL result 204f is generated, no further analysis is required, and the biological classification ensemble model 200 may produce an unacceptable or negative output (e.g., "0" or FALSE) as the result. On the other hand, if a valid result is generated, the biological classification ensemble model 200 outputs a positive (PASS) or other positive output (e.g., "1" or TRUE).

[0084] although Figure 2A Specific models are illustrated, but it should be understood that the biological classification ensemble model 200 is not limited to PCA and PLSDA models. For example, in one embodiment, a class of powerful but computationally expensive nonlinear methods (all discriminant analyses with the suffix DA) can be used, including support vector machines (SVM), artificial neural networks (ANN), and extreme gradient boosting (XGB). Alternatively, in additional embodiments, combining KNN or LDA with PCA can be considered to obtain complementary and / or efficient results.

[0085] Figure 2B Another example flowchart of a bioanalytical method 250 for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, according to various embodiments disclosed herein, is shown. At block 252, the bioanalytical method 250 includes loading a biological integrated classification model configuration (e.g., biological integrated classification model configuration 103) into a first computer memory of a first configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) having a first processor (e.g., first processor 110) and a first scanner (e.g., first scanner 106). The biological integrated classification model configuration includes a biological classification integrated model (e.g., biological classification integrated model 200), which includes an unsupervised model (e.g., unsupervised model 202) and a supervised model (e.g., supervised model 204).

[0086] In various aspects, the configuration of unsupervised models is based on one or more of the following: Principal Component Analysis (PCA), Euclidean distance or correlation; neighborhood-based algorithms, K-means algorithms, quality thresholding (QT) algorithms, centroid algorithms, Ward's algorithm, or fuzzy C-means clustering algorithms. For example, Figure 2A As described, unsupervised models can include PCA models comprising a simplified set of principal components. The configured model is a model trained on data (e.g., Raman spectroscopy data), and the trained model can be used for prediction, classification, or other subsequent product identification purposes. For example, in various aspects, unsupervised models are trained using Raman-based spectroscopy training data to configure the unsupervised model to output a first indicator (e.g., a pass / fail indicator) for one or more biological product types.

[0087] In some aspects, unsupervised models (e.g., unsupervised model 202) are configured to detect and identify variability associated with one or more types of biological products. For example, in some aspects, variability includes instrument (e.g., handheld analyzer) variability or batch-to-batch variability.

[0088] In addition, unsupervised models (e.g., unsupervised model 202) can output indicators (e.g., first indications) based on whether one or more biological product types meet thresholds. Figure 2A As shown, the unsupervised model determines pass / fail based on a threshold. The threshold can be based on one or more of the following (but not limited to) or can be measured by one or more of the following: simplified Q-residuals, Hotelling T-squared values, Mahalanobis distance values, or a specific range of principal component scores. For example, a bio-ensemble classification model for a configurable handheld bioanalyst (e.g., configurable handheld bioanalyst 102) may include classification components selected to reduce the Q-residuals of the bio-ensemble classification model. In this way, the bio-classification model is configured to identify the bio-product type of a given bio-product sample based on the classification components. Generally, Q-residuals are best suited for methods with bio-products of a single specification, where batch-to-batch variability is the primary source of bias between analyzers. Accordingly, as... Figure 5A As shown, the Q residual can be used as a discriminant statistic to determine models that can tolerate the variability of analyzers (e.g., the biological ensemble classification model described in this paper).

[0089] Alternatively, Hotelling T 2 The value can also be used with or in place of the Q residual. Generally, Hotelling T 2 The value represents a measure of variation for each sample within a model (e.g., a biological ensemble classification model). Hotelling T 2 The value indicates how far each sample is from the "center" of the model (value 0). In other words, Hotelling T 2 The value is an indicator of the distance to the model center. Due to the variability of the analyzer, distances from the center often arise. Using Hotelling T... 2 These values ​​are useful for identifying biological products with multiple specifications. In these cases, the different concentrations of active ingredients, excipients, etc., lead to greater variability in Raman spectra compared to batch-to-batch variations.

[0090] exist Figures 6A to 6C In code segment 7 of the embodiment's computer program list, the biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) defines a set of PCA predictions specified for its biological ensemble classification model. Figures 6A to 6C Code snippet 7 also provides the definition of fit summary statistics (e.g., Hotelling T). 2 A script for calculating the Q residual / value. Similarly, in Figures 7A to 7C In code segment 7 of the embodiment's computer program list, the biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) defines a set of PLSDA predictions specified for its biological ensemble classification model. Figures 7A to 7C Code snippet 7 also provides a script to define the calculation of the fitted summary statistics. For Figure 6C and Figure 7C In each of these, the script of code segment 7 can be executed by the first processor 110 as part of an integrated model (including the corresponding models as described herein (e.g., models 202m and 204, respectively)).

[0091] In various aspects, supervised models (e.g., supervised model 204) can be trained using Raman-based spectral training data to configure the supervised model to output a second indicator for one or more biological product types. The second indicator output by the supervised model can be based on whether one or more biological product types meet a predicted threshold for that biological product type. For example, the supervised model can output a pass / fail judgment based on the predicted threshold for the biological product type, such as... Figure 2A As described.

[0092] In various aspects, one or more of the following algorithms are used to train supervised models: Partial Least Squares Discriminant Analysis (PLSDA), Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), Soft Independent Modeling with Analogous Representation (SIMCA), or Logistic Regression Discriminant Analysis (LREGDA). For example, as this paper addresses... Figure 2A As described, a supervised model (e.g., supervised model 204) is a PLSDA model that includes a set of simplified or otherwise optimized latent variables.

[0093] Further reference Figure 2B At box 254, the bioanalysis method 250 further includes receiving at a first processor (e.g., first processor 110) a first Raman-based spectral dataset defining a first biological product sample as scanned by a first scanner.

[0094] At box 256, the bioanalysis method 250 further includes: executing one or more spectral preprocessing algorithms, as specified by the biological integrated classification model configuration, by a first processor (e.g., first processor 110), to reduce spectral bias in the first Raman-based spectral dataset. In various aspects, spectral bias is analyzer-to-analyzer spectral bias between the first Raman-based spectral dataset and one or more other Raman-based spectral datasets from one or more corresponding other handheld bioanalyzers. For example, spectral bias may exist between a Raman-based spectral dataset scanned by a configurable handheld bioanalyzer 102 and a Raman-based spectral dataset scanned by a configurable handheld bioanalyzer 112. Spectral bias may exist even if each Raman-based spectral dataset scanned by each analyzer represents the same type of biological product. Such spectral bias may be caused by analyzer-to-analyzer variability and / or differences (such as software), including version, manufacture, age, operating environment (e.g., temperature), component differences, or other differences in Raman-based analyzers as described herein.

[0095] Spectral preprocessing algorithms are configured to reduce or otherwise mitigate analyzer-to-analyzer spectral bias between a first Raman-based spectral dataset and one or more other Raman-based spectral datasets. For example, in various embodiments, (e.g., on the first processor 110) the spectral preprocessing algorithm is implemented or executed to minimize statistical Type I (e.g., false positives) and / or Type II errors (e.g., false negatives) associated with the identification of a biological product (e.g., biological product 140). In various embodiments, the spectral preprocessing algorithm may reduce analyzer-to-analyzer spectral bias between multiple configurable handheld bioanalysts (e.g., any one of configurable handheld bioanalysts 102, 112, 114, and / or 116).

[0096] At box 258, the biological analysis method 250 further includes utilizing a biological classification ensemble model (e.g., biological classification ensemble model 200) based on a first Raman-based spectral dataset (e.g., as targeted at...). Figures 3A to 3C Visualization and description of Raman-based spectral datasets are used to identify or classify biological product types. For example, in various embodiments, once, as described herein... Figures 3A to 3C and / or Figures 6A to 6C or Figures 7A to 7C The described (e.g., executed by the first processor 110) execution sequence of the spectral preprocessing algorithm allows a configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) to use the preprocessed Raman-based dataset (e.g., such as...) Figure 3C The depicted aligned and / or normalized Raman-based spectral datasets (e.g., including Raman-based spectral datasets 322a, 322b and 322c) are used to identify or classify biological products (e.g., biological product 140).

[0097] Throughout this paper, the configuration of the integrated biological classification model can be transferred or otherwise deployed to another analyzer (e.g., a second configurable handheld biological analyzer). For example, further reference... Figure 2B At box 260, the bioanalysis method 250 includes transferring the configuration of the integrated biological classification model to a second configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 112).

[0098] At block 262, the bioanalysis method 250 includes loading a biological taxonomy ensemble model configuration into a second computer memory (e.g., a configurable handheld bioanalyzer 112), the biological taxonomy ensemble model configuration including a biological taxonomy ensemble model.

[0099] At block 264, the bioanalysis method 250 further includes receiving a second Raman-based spectral dataset, as defined by a second scanner, of a second processor of a second configurable handheld bioanalyzer (e.g., configurable handheld bioanalyzer 112), such that the second scanner scans the sample.

[0100] At block 266, the bioanalysis method 250 further includes identifying the type of biological product based on a second Raman-based spectral dataset by a second processor implementing a biological classification ensemble model (e.g., biological classification ensemble model 200). The second biological product sample may include a new sample of the biological product type.

[0101] Figures 3A to 3C Example execution sequences of spectral preprocessing algorithms for configurable handheld bioanalysts (e.g., configurable handheld bioanalyst 102) are shown. The execution of the spectral preprocessing algorithms (e.g., by a first processor 110) mitigates, reduces, or otherwise mitigates the impact of differences unique to each analyzer (e.g., configurable handheld bioanalysts 102, 112, 114, and / or 116) and reduces bias between Raman-based spectral datasets generated by the scans of these analyzers. One or more spectral preprocessing algorithms can be applied to modify and / or align Raman-based spectral data or other information. In one aspect, the spectral preprocessing algorithm may include applying a derivative transformation to a first Raman-based spectral dataset to generate a modified Raman-based spectral dataset. In some aspects, the derivative transformation is applied to consecutive groupings of 5 to 15 Raman intensity values ​​across the Raman displacement axis. It should be understood that additional and / or different ranges may be used, such as other ranges including 1 to 20 Raman intensity values. Further, in some aspects, the modified Raman-based spectral dataset may be centered.

[0102] On the other hand, the spectral preprocessing algorithm can further include a Raman-based spectral dataset aligned across the Raman displacement axis. In some aspects, the corresponding derivatives for consecutive groups of 5 to 15 Raman intensity values ​​are determined across the Raman displacement axis.

[0103] On the other hand, the spectral preprocessing algorithm can further include normalizing the modified Raman-based spectral dataset across the Raman intensity axis.

[0104] In another aspect, one or more spectral preprocessing algorithms may be performed to modify at least one of the following: (a) training data used to train one or both of a supervised or unsupervised model; or (b) production data used to generate outputs from one or both of a supervised or unsupervised model.

[0105] Figure 3A Example visualizations 302 are shown, such as example Raman-based spectral datasets (e.g., including Raman-based spectral datasets 302a, 302b, and 302c) scanned by one or more handheld bioanalyzers according to various embodiments disclosed herein. Figure 3A Raman-based spectral datasets may include Raman-based spectral datasets (e.g., Raman-based spectral datasets 302a, 302b, and 302c) for generating biological ensemble classification model configurations (e.g., biological ensemble classification model configuration 103) and their associated biological ensemble classification models (e.g., biological ensemble classification model 200) as described herein. Figure 3A Raman-based spectral datasets can be in Figure 6A and / or Figure 7A The datasets identified in code segment 3.

[0106] In some embodiments, Figure 3A Each Raman-based spectral dataset (e.g., including Raman-based spectral datasets 302a, 302b, and 302c) can represent a scan performed by different configurable handheld bioanalyzers (e.g., any configurable handheld bioanalyzer 102, 112, 114, and / or 116). However, in other embodiments, Figure 3A Each Raman-based spectral dataset (e.g., including Raman-based spectral datasets 302a, 302b, and 302c) can represent multiple scans performed by the same configurable handheld bioanalyzer (e.g., any configurable handheld bioanalyzer 102).

[0107] Figure 3A Several Raman-based spectral datasets (e.g., including Raman-based spectral datasets 302a, 302b, and 302c) are depicted, visualizing trans-Raman intensity values ​​(on Raman intensity axis 304) and light wavelength / frequency values ​​(on Raman shift axis 306). Raman intensity axis 304 indicates the intensity of scattered light at a given wavelength across Raman shift axis 306. Raman intensity axis 304 can show a number of photons scattered by a biological product sample, as scanned by an analyzer (e.g., a configurable handheld bioanalyzer 102) (e.g., where data / value 3 is a relative measure of photon intensity measured / scanned by a first scanner 106). Raman shift axis 306 indicates the wavenumber (e.g., inverse wavelength) of the scattered light. The unit of wavenumber is (i.e., wavenumber per centimeter (cm)). -1 This provides an indication of the frequency or wavelength difference between the incident and scattered light. Figure 3A In visualization 302, the displacement axis 306 includes 600 to 1500 cm. -1 The range. Raman intensity axis 304 includes a Raman intensity range of 1 to 5. For example... Figure 3A As shown, each Raman-based spectral dataset (e.g., including Raman-based spectral datasets 302a Raman, 302b, and 302c) pairs at 600 to 1500 cm⁻¹ -1 The Raman intensity values ​​measured within the spectral range are visualized.

[0108] In addition, in various embodiments, Figure 3A Each Raman-based spectral dataset (e.g., including Raman-based spectral datasets 302a, 302b, and 302c) can represent a scan of samples of the same biological product having the same biological product type. In such an embodiment, as Figure 3A As shown, even if any one or more of the configurable handheld bioanalysts may have scanned the same biological product sample of the same biological product type, the Raman intensity values ​​(on the Raman intensity axis 304) of the Raman-based spectral datasets (e.g., including Raman-based spectral datasets 302a, 302b, and 302c) can still be variable across the light wavelength / frequency values ​​(on the Raman shift axis 306). As described herein, this variability may be caused by differences in software, manufacturing, age, optical components, operating environment (e.g., temperature), or other aspects between the configurable handheld bioanalysts (e.g., any one of configurable handheld bioanalysts 102, 112, 114, and / or 116).

[0109] Figure 3B Showing how from Figure 3A Example visualization of the modified Raman-based spectral dataset, derived from the original Raman-based spectral dataset, 312. For example, Figure 3B This can represent the first stage of the execution sequence of the spectral preprocessing algorithm. Figure 3B The visualization 312 includes those related to the present article. Figure 3A The same Raman intensity axis 304 and Raman displacement axis 306 are described. Figure 3B In one embodiment, the processor (e.g., the first processor 110) applies the derivative transformation to... Figure 3A Raman-based spectral datasets (e.g., including Raman-based spectral datasets 302a, 302b, and 302c) are used to generate, for example... Figure 3B The modified Raman-based spectral datasets depicted (e.g., including Raman-based spectral datasets 312a, 312b, and 312c). Specifically, in Figure 3B In one embodiment, a first derivative with 11 to 15 points of data smoothing is applied (i.e., determining the Raman weighted average of consecutive groups of 11 to 15 Raman shift values, and then applying a first derivative transformation to these groups). In other words, Figure 3B The derivative transformation illustrated involves a processor (e.g., first processor 110) determining the Raman weighted average of 11 to 15 consecutive Raman shift values ​​on the Raman shift axis 306 (of the Raman intensity axis 304), and then the processor (e.g., first processor 110) determining the corresponding derivatives of those Raman weighted averages on the Raman shift axis 306. The application of the derivative transformation mitigates the effects of background curvature, for example, due to Rayleigh scattering / suppression optics and / or other dispersive elements. Figure 3A Visualization 302 and Figure 3B The comparison of visualization 312 is illustrated graphically, where Raman-based spectral datasets (e.g., including Raman-based spectral datasets 302a, 302b, and 302c, such as...) are used. Figure 3A Deviations (e.g., vertical and / or horizontal deviations) are removed or reduced to produce a result as shown. Figure 3B The modified Raman-based spectral datasets with fewer changes are depicted (e.g., Raman-based spectral datasets 312a, 312b, and 312c).

[0110] pass Figures 6A to 6C and Figures 7A to 7C The list of computer programs further demonstrates, such as Figure 3B The application of the visualized derivative transformation. For example, in Figures 6A to 6C and Figures 7A to 7C In the code segment 4 of the embodiment's computer program list, for each set of graphs, the biological integrated classification model configuration includes a script executable by a first processor 110 of the configurable handheld bioanalyzer 102, which applies as described herein. Figure 3B The described derivative transformation algorithm. For example, it can be executed... Figure 6A The script in code snippet 4 is used to train unsupervised models (e.g., Figure 2A The model 202m) applies the derivative transformation algorithm and can also perform... Figure 7A The script in code segment 4 is used to supervise the model (e.g., Figure 2A The derivative transformation algorithm is applied to model 204m. It should be understood that the script can be the same script in which the same transformation is applied to the data, and this transformation can be performed only once for the data to be used across both models, in order to, for example, reduce the utilization of computational resources. Alternatively, as... Figure 6A and Figure 7A As shown, each script can be executed independently for each different model.

[0111] Figure 3C An example visualization of a normalized Raman-based spectral dataset is shown, as... Figure 3B The modified, normalized version of the Raman-based spectral dataset. For example, Figure 3C This can represent one or more subsequent stages in the execution sequence of a spectral preprocessing algorithm. Figure 3C The visualization 322 includes those related to the present article. Figure 3A and Figure 3B The same Raman intensity axis 304 and Raman displacement axis 306 are described. For example, in one embodiment, as... Figure 3B The modified Raman-based spectral datasets depicted (e.g., including Raman-based spectral datasets 312a, 312b, and 312c) are aligned by a processor (e.g., a first processor 110) across the Raman shift axis 306 to produce, as shown in the image. Figure 3C The depicted aligned Raman-based spectral datasets (e.g., including Raman-based spectral datasets 322a, 322b, and 322c) are aligned to correct for subtle y-axis shifts (i.e., Raman intensity axis 304) caused by analyzer-to-analyzer bias / differences, as described herein. Figures 6A to 6C and Figures 7A to 7C The list of computer programs further demonstrates, such as Figure 3C The application of the visualized alignment algorithm. For example, in... Figures 6A to 6C and Figures 7A to 7C In code segment 6 of the embodiment's computer program list, the biological ensemble classification model configuration (e.g., biological ensemble classification model configuration 103) includes a script executable by a first processor 110 of a configurable handheld bioanalyzer 102, which applies a mean-centering algorithm to adjust such... Figure 3B The alignment of the depicted modified Raman-based spectral datasets (e.g., including Raman-based spectral datasets 312a, 312b, and 312c) is performed to remove or reduce spectral biases (e.g., vertical and / or horizontal biases) in these modified Raman-based spectral datasets. This adjustment results in... Figure 3C The aligned Raman-based spectral datasets depicted (e.g., including Raman-based spectral datasets 322a, 322b, and 322c). For example, it is possible to perform... Figure 6B The script in code snippet 6 is used to process unsupervised models (e.g., Figure 2A The model 202m) applies the mean-centering algorithm and can also execute Figure 7B The script in code segment 6 is used to supervise the model (e.g., Figure 2A The model 204m applies a mean-centered algorithm. It should be understood that the script can be the same script in which the same algorithm is applied to the data, and this algorithm can be executed once for the data to be used across both models, in order to, for example, reduce the utilization of computational resources. Alternatively, as... Figure 6B and Figure 7B As shown, each script can be executed independently for each different model.

[0112] Alternatively, in another embodiment, such as Figure 3B The modified Raman-based spectral datasets depicted (e.g., including Raman-based spectral datasets 312a, 312b, and 312c) are normalized across the Raman intensity axis 304 by a processor (e.g., a first processor 110) to produce, Figure 3C The depicted aligned Raman-based spectral datasets (e.g., including Raman-based spectral datasets 322a, 322b, and 322c) are normalized using robust normalization algorithms to account for analyzer-to-analyst bias / differences in intensity axis variations (i.e., variations in intensity values ​​across Raman intensity axis 304) as described herein. Figures 6A to 6C and Figures 7A to 7C The list of computer programs further demonstrates, such as Figure 3C The application of the visualized normalization algorithm. For example, in... Figures 6A to 6C and Figures 7A to 7C In the code segment 5 of the embodiment's computer program list, the biological integrated classification model configuration includes a script that can be executed by a first processor 110 of the configurable handheld bioanalyzer 102, applying a normalization algorithm, such as... Figure 3B The modified Raman-based spectral datasets (e.g., including Raman-based spectral datasets 312a, 312b, and 312c) are normalized to remove or reduce spectral biases (e.g., vertical and / or horizontal biases) in these modified Raman-based spectral datasets. This normalization results in... Figure 3C The normalized Raman-based spectral datasets depicted (e.g., including Raman-based spectral datasets 322a, 322b, and 322c). Specifically, in Figures 6A to 6C and Figures 7A to 7C In some embodiments, for example, the first processor 110 applies the Standard Normal Variable (SNV) algorithm to, for example, Figure 3B The modified Raman-based spectral datasets depicted (e.g., including Raman-based spectral datasets 312a, 312b, and 312c) are used to produce results such as Figure 3C The depicted aligned Raman-based spectral datasets (e.g., including Raman-based spectral datasets 322a, 322b, and 322c). For example, it is possible to perform... Figure 6B The script in code snippet 5 is used to train unsupervised models (e.g., Figure 2A The model 202m applies the SNV algorithm and can also execute... Figure 7B The script in code snippet 5 is used to supervise the model (e.g., Figure 2A The SNV algorithm is applied to model 204m. It should be understood that the script can be the same script in which the same algorithm is applied to the data, and this algorithm can be executed once for the data to be used across both models, in order to, for example, reduce the utilization of computational resources. Alternatively, as... Figure 6B and Figure 7B As shown, each script can be executed independently for each different model.

[0113] Application of alignment and / or normalization algorithms (e.g., for...) Figure 3C (as described) removed or reduced such Figure 3B The spectral biases of the modified Raman-based spectral datasets (e.g., including Raman-based spectral datasets 312a, 312b, and 312c) depicted in the text are described. Figure 3B Visualization 312 and Figure 3C The comparison of visualization 322 is illustrated graphically, where Raman-based spectral datasets (e.g., including Raman-based spectral datasets 312a, 312b, and 312c, such as...) are used. Figure 3B Spectral biases (e.g., vertical and / or horizontal biases) are removed or reduced to produce a result as shown. Figure 3C The depicted data are aligned and / or normalized Raman-based spectral datasets with fewer variations (e.g., Raman-based spectral datasets 322a, 322b, and 322c).

[0114] Figure 4 An example visualization 402 of Raman-based spectral datasets of mAb 1 DP 410e and mAb 2 DP 410a, scanned by a handheld bioanalyzer (e.g., configurable handheld bioanalyzer 102) according to various embodiments disclosed herein, is provided. The Raman-based spectral datasets (e.g., including Raman-based spectral datasets 410a and 410e) are visualized using trans-Raman intensity values ​​(on Raman intensity axis 404) and light wavelength / frequency values ​​(on Raman shift axis 406). Raman intensity axis 404 indicates the intensity of scattered light at a given wavelength across Raman shift axis 406. Raman intensity axis 404 can show a number of photons scattered by the biological product sample as scanned by the analyzer (e.g., configurable handheld bioanalyzer 102) (e.g., where data / value 3 is a relative measure of photon intensity measured / scanned by first scanner 106). Raman shift axis 406 indicates the wavenumber of the scattered light (e.g., inverse wavelength). The unit of wavenumber (i.e., wavenumber per centimeter (cm)) -1 This provides an indication of the frequency or wavelength difference between the incident and scattered light. Figure 4 In visualization 402, the displacement axis 406 includes 350 to 2000 cm. -1 The range. Raman intensity axis 404 includes a Raman intensity range of 0 to 3. For example... Figure 4 As shown, each Raman-based spectral dataset (e.g., including Raman-based spectral datasets 410a and 410e) pairs at 600 to 1500 cm⁻¹ -1 The Raman intensity values ​​measured within the spectral range are visualized.

[0115] Figure 4 Visual representations based on Raman spectra have similar datasets and may therefore be difficult to distinguish. That is, in Figure 4 In the example visualization, the Raman spectra (Raman intensities across Raman shifts) of mAb 1 DP 410e and mAb 2 DP 410a make them difficult to distinguish with the configurable handheld bioanalyst 102. However, it should be noted that the Raman spectra produced by different product types are not entirely identical (but they can be similar). For example, as... Figure 4 The different (but similar) Raman spectra of mAb 1 DP 410e and mAb 2 DP 410a shown may have tracking and / or overlapping Raman spectral features, but at least some of these Raman spectral features are different. Therefore, in various aspects of this paper, a first biological product type among a given one or more biological product types and a second biological product type among a given one or more biological product types may have similar Raman-based spectra (e.g., as shown in the original text). Figure 4 The mAb 1 DP 410e and mAb 2 DP 410a are shown, but these spectra have different features that can be discovered by training a model (e.g., a supervised model or an unsupervised model).

[0116] Figure 5A This demonstrates when a Raman-based spectral dataset (including...) is provided. Figure 4 Example visualization of the Q-residuals of a non-ensemble unsupervised biological classification model when those datasets are used as input (502). At least in some respects, Figure 5A This represents a non-integrated unsupervised model, demonstrating the model with the best case and showing that it is impossible to obtain true positives (e.g., true negatives) with 100% predictive power or other accuracy. Specifically, as shown in key 509, when mAb 1 DP 410e is scanned by a scanner designated TM5137, mAb 1 DP 410e is represented as mAb 1 - TM5137 (510mAb15137 in visualization 502). When mAb 2 DP 410a is scanned by a scanner designated TM5137, mAb 2 DB 410a is represented as mAb 2 - TM5137 (510mAb25137 in visualization 502). Additional Raman-based spectral datasets scanned by different configurable handheld bioanalyzers are also shown. Figure 4 (Not shown in the image). Figure 5A The three configurable handheld bioanalysts in the example are identified or otherwise indicated as follows: TM4164, TM5133, and TM5137. Therefore, the respective Raman-based spectral datasets scanned by the respective scanners further include mAb 1 datasets mAb1 - TM4164 (510mAb14164 for visualization 502) and mAb 1 - TM5133 (510mAb15133 for visualization 502), and mAb 2 datasets mAb 2 - TM4164 (510mAb24164 for visualization 502) and mAb 2 - TM5133 (510mAb25133 for visualization 502). Key 509 also indicates the Q residual threshold 508, which is... Figure 5A The threshold used to distinguish mAb 1 from mAb 2.

[0117] For example, visualizing 502 could include classification results using only the PCA model, where mAb 1 DP 410e is considered the target product. Figure 5A In the example, visualization 502 includes a Q-residual axis 504 and a linear index axis 506. The linear index 506 represents the ordering of the individual spectra used in the dataset to evaluate model performance. The Q-residuals (e.g., the Q-residuals of the Q-residual axis 504) represent error values ​​in the range of 0 to 1.5 and provide misfit statistics calculated as the sum of squares for each product sample. The Q-residuals represent the magnitude of the remaining variation in each sample after projection through a given model (e.g., a PCA model of a biological ensemble classification model as described in this paper). More generally, as... Figure 5A As shown in the embodiments, the Q-residual values ​​(along the Q-residual axis 504) are used as discriminant statistics. The Q-residual is a measure of the portion of a given model that is not explained. For example, in embodiments using PCA models (e.g., projecting the spectrum onto the first principal component), Figure 5A The value will show the remaining portion (residual) after the scan data (e.g., 410a and 410e) are projected onto the first principal component.

[0118] Visualization 502 illustrates the limitations of using only a PCA model based on simplified Q-residuals as the discriminative feature. For example, as Figure 5A As shown, PCA models alone (or other classification models) may be inaccurate, for example, Figure 5A The Q residuals for a single PCA model calibrated for mAb 1 DP are shown. Figure 5A In the example, this model is not sensitive enough to distinguish between mAb1 DP and mAb2 DP.

[0119] More specifically, in Figure 5A In the examples, a single PCA model was trained or otherwise configured using mAb 1 DP spectra of two samples / batches acquired on two Raman instruments (TM5133 and TM5137). The model was built using standard preprocessing parameters (e.g., truncating the spectral range to 350–1801 cm⁻¹). -1 Then, the first derivative was taken for Savitsky-Golay smoothing, standard normal variables were applied, and the values ​​were mean-centered. Model validation was performed using three batches of independent spectra from each of mAb 1 DP and mAb 2 DP, acquired on three Raman instruments (e.g., configurable handheld bioanalyzers (TM4164, TM5133, and TM5137)). In some respects, higher residuals were often observed on instruments not included in the model building (i.e., network instruments), thus the pass / fail criteria were adjusted to compensate for this. Figure 5A In the example, the Raman curves of mAb 2 DP are similar to those of mAb 1 DP enough that for any configurable handheld bioanalyzer, it is impossible to distinguish between the two products based on their Q-residual values ​​within normal / expected instrument variability.

[0120] Figure 5B This demonstrates various embodiments of the present paper when Raman-based spectral datasets (including...) are provided. Figure 4 Example visualizations of the predicted output of a supervised biological classification ensemble model (e.g., biological classification ensemble model 200) when those datasets are used as input. 552. At least in some respects, Figure 5B This indicates a non-supervised model that provides a true positive rate (i.e., true negative rate) for distinguishing between mAb 1 DP and mAb 2 DP with 100% prediction accuracy or other precision. This model allows for fine-tuning of the unsupervised model (e.g., as...). Figure 5A As described), this makes the true positive rate 100% (or its appropriate value). Additionally, as... Figure 5B The described second supervised model provides better performance than the single supervised model. Figure 5A The model achieves better performance to achieve a true negative rate (e.g., 100% or an approximation thereof). Figure 5B The examples used ensemble models as described in this paper, such as the biological classification ensemble model 200, which employs each of the unsupervised and supervised models. As an example, Figure 5B The use of the PLSDA model is demonstrated, where mAb 1 DP is considered the target product. Figure 5B In the example, visualization 552 includes a Y-prediction axis 554 and a linear index axis 556. The linear index 556 represents the order of individual spectra within the dataset in the range of 0 to 275. The Y-prediction axis 554 represents a series of values ​​that represent the label output or other dependent variable output of the PLSDA model with respect to mAb 1 DP 70 mg / mL. In each aspect, Figure 5B The output of the second PLSDA model was demonstrated (e.g., Figure 2A The model shown is 204m). For precursor PCA models (e.g., Figure 2A In model 202m, conditions can be adjusted (e.g., setting a threshold for whether the Q residual is acceptable) so that the true positive rate can be 100%. Although PCA models may generate false positive results for, for example, mAb 2, supervised biological taxonomy ensemble models can avoid false positives, i.e., their true positive rate is 100%.

[0121] Key 560 includes the same values ​​for Raman-based spectral datasets and associated configurable handheld bioanalysts, including when mAb 1 DP 410e is scanned by a scanner designated TM5137, mAb 1 DP 410e is represented as mAb1-TM5137 (i.e., 510mab15137 for visualization 502), and so on, such as Figure 5A As stated above. However, key 560 also includes a Y discrimination threshold 558, which is... Figure 5B The threshold used to distinguish mAb 1 from mAb 2. Figure 5A The model does not use the Q residual threshold (e.g. Figure 5A (as shown), instead, it uses the output of PLSDA, which is the predicted value, in Figure 5B In the example, for mAb 1 DP 70 mg / mL, the range can be from -0.1 to 1.2. In this example, a nominal value of 1 corresponds to a perfect fit for the mAb 1 DP 70 mg / mL class, and a nominal value of 0 corresponds to a perfect fit for the mAb 2 DP 100 mg / mL class.

[0122] Figure 5B Visualization 552 illustrates the novel, accurate output of a biological classification ensemble model (e.g., biological classification ensemble model 200), showing predictions for identifying mAb 1 when the output value is above a threshold (e.g., a Y-discrimination threshold 558) and predictions for identifying mAb 2 when the output value is below a threshold (e.g., a Y-discrimination threshold 558), thus allowing for accurate classification of mAb 1 DP and mAb 2 DP. In some aspects, for example, as Figure 2A As also described herein, the output value can be used for "PASS" products (e.g., when the output value is higher than approximately 0.5, such as...). Figure 5B (As shown). Similarly, the output value can be used for "FAIL" products (e.g., when the output value is below approximately 0.5, such as...). Figure 5B (As shown).

[0123] exist Figure 5B In the example, the model was trained or otherwise configured (e.g., a PLSDA model) using mAb 1 DP and mAb 2 DP spectra (two samples / batches, each using two Raman instruments, TM5133 and TM5137). This model is used for... Figure 5A The standard preprocessing parameters for the PCA model are established (e.g., truncating the spectral range to 350–1801 cm⁻¹). -1 Then, the first derivative was taken for Savitsky-Golay smoothing, standard normal variables were applied, and the values ​​were then mean-centered. Model validation was performed using three batches of independent spectra from each of mAb 1 DP and mAb 2 DP, acquired on three Raman instruments (TM4164, TM5133, and TM5137). Figure 5B As shown, the ensemble model including the supervised PLSDA algorithm (compared to the standalone PCA model) can better distinguish mAb 1 DP and mAb 2 DP based on their unique Raman spectra, thus mitigating the effects of instrument-to-instrument bias. In this way, the supervised model of the biological classification ensemble model (e.g., biological classification ensemble model 200) can be configured to distinguish between biological product samples with a given biological product type and different biological product samples with different (but similar) biological product types (e.g., in the case of mAb 1 DP and mAb 2 DP). In various cases, the first biological product type and the different biological product types may each have different local features within similar Raman spectral ranges, and the biological classification ensemble model 200 can use these features to distinguish between the first biological product type and the different biological product types.

[0124] In this way, ensemble models (e.g., a biological classification ensemble model 200) can leverage the advantages of classification and prediction models (e.g., unsupervised and supervised models) to improve the accuracy of identification and detection by the configurable handheld biological analyzer 102, thereby distinguishing product types. In contrast, single models with broad specificity (e.g., PCA models) may struggle to detect product types with similar Raman spectral datasets or other Raman-related features for other products with dissimilar or moderately similar Raman curves. One such approach is described in publication WO 2021 / 081263, filed October 23, 2020, entitled "Configurable Handheld Biological Analyzers for Identification of Biological Products based on Raman Spectroscopy," application number PCT / US2020 / 056961. While single models offer improvements over traditional detection methods, the bioclassification ensemble models described herein (e.g., bioclassification ensemble model 200) allow for further accurate identification and detection of product types. Typically, bioclassification ensemble models are configured to enhance accuracy and product type identification performance by linking or otherwise combining predictions from two or more artificial intelligence models. For example, the first model may include, for instance, […]. Figure 2A The unsupervised model described. As a non-limiting example, the first model may include a PCA model. A PCA model provides a first layer of specificity and identification between different product types. A PCA model is also unsupervised, meaning it can distinguish product samples without training. However, in some cases, a PCA model may not be able to distinguish product types with sufficient certainty based on Raman spectroscopy.

[0125] To improve PCA models, ensemble models as described in this paper (e.g., biological classification ensemble model 200) can be developed, in which a supervised model or other second model is added to address product types with similar Raman spectra. The second model may include, for example, those described in this paper... Figure 2A The PLSDA model is described. The PLSDA model can be trained to have high specificity with respect to product type (e.g., different product types with similar Raman spectra, such as mAb 1 and mAb 2).

[0126] It should be noted that when attempting to identify, measure, or classify such biological product types, typical analyzers (without implementing or executing the integrated biological classification model configuration 103 as described herein) often produce a large number of Type I errors (e.g., false positives) and Type II errors (e.g., false negatives).

[0127] However, a configurable handheld bioanalyst (e.g., configurable handheld bioanalyst 102) loaded and executing a biological integrated classification model configuration as described herein (e.g., biological integrated classification model configuration 103) can be used to accurately identify, classify, measure, or otherwise distinguish biological product types such as mAb 2 DS / DP, mAb 1 DP, mAb 3 DP, etc. Specifically, each product type may have local features of varying Raman intensity values ​​(with different shapes, peaks, or other unique / different relative intensities) within the Raman spectral range, specific to each biological product type, such as mAb 2 DS / DP, mAb 1 DP, mAb 3 DP, etc. Therefore, these different local features provide a source of product-specific information that the configurable handheld bioanalyst 102 can use to identify, classify, or otherwise distinguish biological products as described herein. However, it should be understood that these biological product types are merely examples, and other biological product types or biological products can be identified, classified, measured, or otherwise distinguished in the same or similar manner as described with respect to the various embodiments herein.

[0128] As described herein, a bio-integrated classification model (e.g., bio-integrated classification model 200) can be configured to identify, classify, measure, or otherwise distinguish a given biological product sample having a given biological product type (e.g., mAb 2 DS / DP) from different or second biological product samples having different or second biological product types (e.g., mAb 2 DS / DP). For example, as described herein, once the bio-integrated classification model configuration 103 is configured, a handheld bioanalyzer 102 can be configured to perform spectral preprocessing algorithms (e.g., as described herein) on a Raman-based spectral dataset such as that received by a first scanner 106. Figures 3A to 3C (As described). Once the Raman-based spectral dataset has been preprocessed using a spectral preprocessing algorithm, the configurable handheld bioanalyzer 102 can use an integrated model to identify or classify biological products (e.g., as described in this paper). Figure 2A and Figure 2B (as described).

[0129] Additional examples

[0130] The following additional examples provide further support based on the various embodiments described herein. In particular, the following additional examples demonstrate Raman spectroscopy for rapid identification (ID) verification of biological therapeutic protein products in solution. These examples demonstrate a unique combination of Raman features associated with both the therapeutic agent and the excipient, serving as the basis for product differentiation. As described herein, product ID methods (e.g., bioanalytical methods) involve acquiring Raman spectra of (multiple) target products on multiple Raman analyzers (e.g., configurable handheld bioanalyzers as described herein). The spectra can then be dimensionality-reduced using principal component analysis (PCA) and / or partial least squares discriminant analysis (PLSDA) to define product-specific models (e.g., bioclassification ensemble models), which will serve as the basis for product ID determination using configurable handheld bioanalyzers and bioanalytical methods for Raman spectroscopy-based identification of biological products as described herein. Product-specific models (e.g., bioclassification ensemble models) can be transferred to individual instruments (e.g., configurable handheld bioanalyzers) validated for product testing. These models can be used for a variety of purposes, including quality control, incoming quality assurance, and manufacturing. Such analyzers and methods can be used in various Raman devices (e.g., configurable handheld bioanalysts) from different manufacturers. In this way, the additional examples further demonstrate that the Raman ID analyzers and methods described herein (e.g., configurable handheld bioanalysts and related methods) offer a wide range of uses and testing options for solution-based protein products in the biopharmaceutical industry.

[0131] Additional examples—Raman instruments (e.g., configurable handheld bioanalyzers) and measurements

[0132] As an additional example, as described herein, Raman spectroscopy is measured using a configurable handheld bioanalyzer. For instance, in some embodiments, the configurable handheld bioanalyzer may be a Raman-based handheld analyzer, such as the TruScan™ RM handheld Raman analyzer supplied by Thermo Fisher Scientific Inc. In such embodiments, the configurable handheld bioanalyzer may implement TruTools. TM Chemometric software packages. However, it should be understood that, based on the disclosure herein, other brands or types of Raman analyzers using additional and / or different software packages may be utilized. In some embodiments, the configurable handheld bioanalyzer may be configured with a 785 nm grating-stabilized laser source (maximum output 250 mW) coupled to a focusing optics (e.g., 0.33 NA, 18 mm working distance, > 0.2 mm spot size) for sample review. For an additional example, the product solution contained in a glass vial is fixed in front of the focusing optics using the vial adapter of the configurable handheld bioanalyzer. All spectra are collected using the same spectral acquisition settings (but other settings may also be used), e.g., laser power = 250 mW, integration time = 1000 ms, spectral co-additions = 70. For an additional example, the spectra of the product are collected over a period of time using three different configurable handheld bioanalyzers (hereinafter referred to as configurable handheld bioanalyzers 1-3) and / or instruments dedicated to configuring and / or developing (multiple) bioanalytical methods for the identification of biological products based on Raman spectroscopy as described herein. It should be understood that more or fewer analyzers using the same or different settings can be used to set up, configure or otherwise initialize configurable handheld bioanalyzers and associated (multiple) bioanalytical methods as described herein.

[0133] Additional Example – Development of Multivariate Raman ID Bioanalytical Methods

[0134] Raman spectral models (e.g., ensemble biological classification models) can be generated, developed, or loaded as described herein. For example, in some embodiments, SOLO software equipped with the ModelExporter add-on (Solo+Model_Exporter version 8.2.1; EigenvectorResearch, Inc.) can be used to generate, develop, or load Raman spectral models (e.g., ensemble biological classification models). However, it should be understood that other software can be used to generate, develop, or load Raman spectral models (e.g., ensemble biological classification models). Typically, spectra for model construction can be collected using configurable handheld bioanalysts (e.g., three configurable handheld bioanalysts) as repeated scans of two or more different batches of material. Spectra are typically acquired over multiple days to account for instrument drift. In some embodiments, the spectral range can be reduced to exclude detector noise > 1800 cm⁻¹ and background variability caused by Rayleigh lines < 400 cm⁻¹ before incorporating into the model (e.g., ensemble biological classification model). As described herein, the spectra can be further preprocessed and mean-centered for each model. The model can also be refined by using cross-validation through a random subset procedure, referring to the Raman spectra of the target and challenge products shown in Table 1.

[0135] Biological ensemble classification model configurations (e.g., PCA and PLSDA ensemble model configurations) and Raman spectroscopy acquisition parameters can be configured or loaded onto a configurable handheld bioanalyzer, and / or multiple bioanalytical methods can be used for Raman spectroscopy-based identification of biological products as described herein. Acceptance (e.g., pass / fail) criteria for each method can also be specified.

[0136] As described in this paper, the qualification criteria for unsupervised models can be based on a simplified Hotelling T. 2 (T r 2 ) and Q residual (Q r The threshold is one of two summary statistics that typically describe the goodness of Raman spectra as described by biological ensemble classification models (e.g., PCA models). Equations (1) through (4) below provide user-selectable example decision logic options for positive identification or determination (e.g., pass / fail criteria) using an unsupervised model (e.g., PCA model) of a biological ensemble classification model (e.g., biological ensemble classification model 200):

[0137] (1)

[0138] (2)

[0139] (3)

[0140] (4)

[0141] In the example equation above, Hotelling T is obtained by dividing the original value by the corresponding confidence interval. 2 The values ​​and Q residuals are normalized (i.e., simplified to T respectively). r 2 and Q r This sets the upper limit value to 1. It should be understood that configurable handheld bioanalyzers can use different and / or additional thresholds, different and / or additional equations, and / or use (multiple) bioanalytical methods without departing from the disclosure herein.

[0142] Additional description

[0143] The above description describes various devices, components, parts, subsystems, and methods used in connection with drug delivery devices. Devices, components, parts, subsystems, methods, or drug delivery devices may further include or be used with drugs, including but not limited to those drugs referred to below and their class and biosimilar counterparts. As used herein, the term "drug" is used interchangeably with other similar terms and can refer to any type of drug or therapeutic material, including traditional and non-traditional drugs, nutritional supplements, tonics, biologics, bioactive agents and compositions, macromolecules, biosimilars, bioequivalents, therapeutic antibodies, peptides, proteins, small molecules, and class of substances. Non-therapeutic injectable materials are also included. Drugs may be in liquid form, lyophilized form, or in a form reconstructable from lyophilized form. The following exemplary list of drugs should not be considered as all-encompassing or restrictive.

[0144] The medication will be contained in a reservoir. In some cases, the reservoir is a master container that is filled or pre-filled with the medication for treatment. This master container can be a vial, cartridge, or pre-filled syringe.

[0145] In some embodiments, the reservoir of the drug delivery device may be filled with colony-stimulating factors (such as granulocyte colony-stimulating factor (G-CSF)), or the device may be used in conjunction with colony-stimulating factors. Such G-CSF agents include, but are not limited to, Neulasta® (pefilgrastim, PEGylated filgrastim, PEGylated G-CSF, PEGylated hu-Met-G-CSF) and Neupogen® (filgrastim, G-CSF, hu-MetG-CSF).

[0146] In other embodiments, the drug delivery device may include or be used with an erythropoiesis stimulant (ESA), which may be in liquid or lyophilized form. An ESA is any molecule that stimulates erythropoiesis. In some embodiments, the ESA is an erythropoiesis-stimulating protein. As used herein, "erythropoiesis-stimulating protein" means any protein that directly or indirectly causes activation of the erythropoietin receptor (e.g., by binding to and causing dimerization of the receptor). Erythropoiesis-stimulating proteins include erythropoietin and its variants, analogs, or derivatives that bind to and activate the erythropoietin receptor; antibodies that bind to and activate the erythropoietin receptor; or peptides that bind to and activate the erythropoietin receptor. Erythropoietin-stimulating proteins include, but are not limited to, Epogen® (epogen α), Aranesp® (dabepoetin α), Dynepo® (epogen δ), Mircera® (methoxy-polyethylene glycol-epogen β), Hematide®, MRK-2578, INS-22, Retacrit® (epogen ζ), Neorecormon® (epogen β), Silapo® (epogen ζ), Binocrit® (epogen α), epogen α Hexal, Abseamed® (epogen α), Ratioepo® (epogen θ), Eporatio® (epogen θ), Biopoin® (epogen θ), epogen α, epogen β, epogen ι, epogen ω, epogen δ, epogen ζ, epogen θ and epogen δ, pegylated erythropoietin, carbamylated erythropoietin, and their molecules or variants or analogues.

[0147] The specific illustrative proteins are those described below, including their fusions, fragments, analogs, variants, or derivatives: OPGL-specific antibodies, peptides, related proteins, etc. (also known as RANKL-specific antibodies, peptides, etc.), including fully humanized OPGL-specific antibodies and human OPGL-specific antibodies, especially fully humanized monoclonal antibodies; myostatin-binding proteins, peptides, related proteins, etc., including myostatin-specific peptides; IL-4 receptor-specific antibodies, peptides, related proteins, etc., particularly those inhibiting the activity of IL-4 and / or IL-4 receptors. -13-mediated activities involving receptor binding; interleukin-1 receptor 1 ("IL1-R1") specific antibodies, peptides, and related proteins; Ang2 specific antibodies, peptides, and related proteins; NGF specific antibodies, peptides, and related proteins; CD22 specific antibodies, peptides, and related proteins, especially human CD22 specific antibodies, such as, but not limited to, humanized and fully human antibodies, including but not limited to humanized and fully human monoclonal antibodies, particularly including but not limited to human CD22 specific IgG antibodies, such as human-mouse monoclonal hLL2. Dimers of the γ-chain linked to the human-mouse monoclonal hLL2 κ chain by disulfide, such as the fully humanized human CD22-specific antibody in epazuzumab, CAS Registry No. 501423-23-0; IGF-1 receptor-specific antibodies, peptides, and related proteins, including but not limited to anti-IGF-1R antibodies; B-7-related protein 1-specific antibodies, peptides, and related proteins ("B7RP-1", also known as B7H2, ICOSL, B7h, and CD275), including but not limited to B7RP-specific fully human monoclonal IgG2 antibodies, including but not limited to fully human IgG2 monoclonal antibodies binding to epitopes in the first immunoglobulin-like domain of B7RP-1, including but not limited to those inhibiting the interaction of B7RP-1 with its native receptor ICOS on activated T cells; IL-15-specific antibodies, peptides, and related proteins, such as, in particular, humanized monoclonal antibodies, including but not limited to HuMaxIL-15 antibodies and related proteins, such as, for example, 146B7; IFN γ-specific antibodies, peptides, and related proteins, including but not limited to human IFN γ-specific antibodies, and including but not limited to fully human anti-IFN γ antibodies; TALL-1 specific antibodies, peptides, and related proteins, as well as other TALL-specific binding proteins; parathyroid hormone ("PTH") specific antibodies, peptides, and related proteins; thrombopoietin receptor ("TPO-R") specific antibodies, peptides, and related proteins;Hepatocyte growth factor ("HGF") specific antibodies, peptides, and related proteins, including those targeting the HGF / SF:cMet axis (HGF / SF:c-Met), such as fully human monoclonal antibodies that neutralize hepatocyte growth factor / dispersant (HGF / SF); TRAIL-R2 specific antibodies, peptides, and related proteins; activin A specific antibodies, peptides, and proteins; TGF-β specific antibodies, peptides, and related proteins; amyloid-β protein specific antibodies, peptides, and related proteins; c-Kit specific antibodies, peptides, and related proteins, including but not limited to proteins that bind to c-Kit and / or other stem cell factor receptors; OX40L specific... Antibodies, peptides, and related proteins, including but not limited to proteins that bind to other ligands of OX40L and / or OX40 receptors; Activase® (alteplase, tPA); Aranesp® (dabepoetin α); Epogen® (epogenetin α, or erythropoietin); GLP-1, Avonex® (interferon β-1a); Bexxar® (tosimomab, anti-CD22 monoclonal antibody); Betaseron® (interferon-β); Camppath® (alemumab, anti-CD52 monoclonal antibody); Dynepo® (epogenetin δ); Velcade® (bortezomib); MLN0002 (anti-α4β7) mAb); MLN1202 (anti-CCR2 chemokine receptor mAb); Enbrel® (etanercept, TNF receptor / Fc fusion protein, TNF blocker); Eprex® (ebertin α); Erbitux® (cetuximab, anti-EGFR / HER1 / c-ErbB-1); Genotropin® (growth hormone, human growth hormone); Herceptin® (trastuzumab, anti-HER2 / neu(erbB2) receptor mAb); Humatrope® (growth hormone, human growth hormone); Humira® (adalimumab); Vectibix® (panitumab), Xgeva® (dinosumab), Prolia® (dinosumab) Susemab, Enbrel® (etanercept, TNF-receptor / Fc fusion protein, TNF blocker), Nplate® (romistine), rilotumumab, ganitumab, conatumumab, brodalumab, insulin in solution; Infergen® (alfacon-1 interferon); Natrecor® (nesiritide; recombinant human B-type natriuretic peptide (hBNP)); Kineret® (anaspirin); Leukine® (saxaglastine, rhuGM-CSF); LymphoCide® (epazolizumab, anti-CD22 mAb);Benlysta™ (lymphostat B, belimumab, anti-BlyS mAb); Metalyse® (tenectase, t-PA analog); Mircera® (methoxy-PEG-ebertheline beta); Mylotarg® (gem-tuzumab-ozomicin); Raptiva® (efalizumab); Cimzia® (sertozumab, CDP 870); Soliris™ (eculizumab); Pexazumab (anti-C5 complement); Numax® (MEDI-524); Lucentis® (ranibumab); Panorex® (17-1A, ezolomide); Trabio® (lerdelimumab); TheraCim hR3 (Nimotuzumab); Omnitarg (Pertuzumab, 2C4); Osidem® (IDM-1); OvaRex® (B43.13); Nuvion® (Vemcizumab); Cantuzumab mertansine (huC242-DM1); NeoRecormon® (Ibertin β); Neumega® (Interleukin-11); Orthoclone OKT3® (Moromab-CD3, anti-CD3 monoclonal antibody); Procrit® (Ibertin α); Remicade® (Infliximab, anti-TNFα monoclonal antibody); Reopro® (Abciximab, anti-GP) IL-1 receptor monoclonal antibody); Actemra® (anti-IL6 receptor mAb); Avastin® (bevacizumab); HuMax-CD4 (zanolimumab); Rituxan® (rituximab, anti-CD20 mAb); Tarceva® (erlotinib); Roferon-A® (interferon α-2a); Simulect® (baliximab); Prexige® (romecoxib); Synagis® (palizumab); 146B7-CHO (anti-IL15 antibody, see US Patent No. 7,153,507); Tysabri® (nateliximab, anti-α4 integrin mAb); Valortim® (MDX-1303, anti-anthrax protective antigen mAb); ABthrax™; Xolair® (omaliximab); ETI211 (anti-MRSA mAb); IL-1 trap (extracellular domain of the Fc portion of human IgG1 and IL-1 receptor components (type I receptor and receptor accessory protein); VEGF trap (Ig domain of VEGFR1 fused to IgG1 Fc); Zenapax® (dalizumab); Zenapax® (dalizumab, anti-IL-2Rα mAb).Zevalin® (Teimomab); Zetia® (Ezetimibe); Orencia® (Acecip, TACI-Ig); Anti-CD80 Monoclonal Antibody (Galiximab); Anti-CD23 mAb (Luximab); BR2-Fc (huBR3 / huFc fusion protein, soluble BAFF antagonist); CNTO 148 (Golimubab, anti-TNFα mAb); HGS-ETR1 (Mapatumumab, human anti-TRAIL receptor-1 mAb); HuMax-CD20 (Ocrelizumab, anti-CD20 human mAb); HuMax-EGFR (Zalumumab); M200 (Volociximab, anti-α5β1 integrin mAb); MDX-010 (Ipilimumab, anti-CTLA-4) mAb and VEGFR-1 (IMC-18F1); anti-BR3 mAb; anti-Clostridium difficile toxin A and toxin BC mAb MDX-066 (CDA-1 and MDX-1388); anti-CD22 dsFv-PE38 conjugates (CAT-3888 and CAT-8015); anti-CD25 mAb (HuMax-TAC); anti-CD3 mAb (NI-0401); adecatumumab; anti-CD30 mAb (MDX-060); MDX-1333 (anti-IFNAR); anti-CD38 mAb (HuMax CD38); anti-CD40L mAb; anti-Cripto mAb; anti-CTGF fibrinogen for stage I idiopathic pulmonary fibrosis (FG-3019); anti-CTLA4 mAb; anti-eotaxin1 mAb (CAT-213); anti-FGF8 mAb; anti-ganglioside GD2 mAb; Antiganglioside GM2 mAb; Anti-GDF-8 human mAb (MYO-029); Anti-GM-CSF receptor mAb (CAM-3001); Anti-HepC mAb (HuMax HepC); Anti-IFNα mAb (MEDI-545, MDX-1103); Anti-IGF1R mAb; Anti-IGF-1R mAb (HuMax-Inflam); Anti-IL12 mAb (ABT-874); Anti-IL12 / IL23 mAb (CNTO 1275); Anti-IL13 mAb (CAT-354); Anti-IL2Ra mAb (HuMax-TAC); Anti-IL5 receptor mAb; Anti-integrin receptor mAb (MDX-018, CNTO 95); Anti-IP10 ulcerative colitis mAb (MDX-1100); BMS-66513;Anti-mannose receptor / hCGβ mAb (MDX-1307); anti-mesothelin dsFv-PE38 conjugate (CAT-5001); anti-PD1 mAb (MDX-1106 (ONO-4538)); anti-PDGFRα antibody (IMC-3G3); anti-TGFβ mAb (GC-1008); anti-TRAIL receptor-2 human mAb (HGS-ETR2); anti-TWEAK mAb; anti-VEGFR / Flt-1 mAb; and anti-ZP3 mAb (HuMax-ZP3).

[0148] In some embodiments, the drug delivery device may comprise or be used with sclerosing protein antibodies, such as, but not limited to, romosozumab, blosozumab, or BPS 804 (Novartis), and in other embodiments, a monoclonal antibody (IgG) binding to the human proprotein convertase subtilisin / Kexin 9 (PCSK9). Such PCSK9-specific antibodies include, but are not limited to, Repatha® (evolocumab) and Praluent® (alirocumab). In other embodiments, the drug delivery device may comprise or be used with rituximab, bixalomer, trebananib, ganitamumab, kanamumab, motesanib diphosphate, brodanumab, vidupiprant, or panitumumab. In some embodiments, the reservoir of the drug delivery device may be filled with IMLYGIC® (talimogenelaherparepvec) or another oncolytic HSV for the treatment of melanoma or other cancers, or the device may be used with such other oncolytic HSVs, including but not limited to OncoVEXGALV / CD; OrienX010; G207; 1716; NV1020; NV12023; NV1034; and NV1042. In some embodiments, the drug delivery device may comprise or be used with an endogenous tissue metalloproteinase inhibitor (TIMP), such as, but not limited to, TIMP-3. Antagonistic antibodies against the human calcitonin gene-related peptide (CGRP) receptor (such as, but not limited to, mAb 1) and bispecific antibody molecules targeting the CGRP receptor and other headache targets may also be delivered using the drug delivery device of this disclosure. Additionally, bispecific T-cell conjugates (BiTE) may also be used. ® ) molecules (such as, but not limited to, BLINCYTO) ® (Bonatumab) can be used in or with the drug delivery device of this disclosure. In some embodiments, the drug delivery device may contain or be used with an APJ macromolecular agonist, such as, but not limited to, apelinide or an analogue thereof. In some embodiments, a therapeutically effective amount of anti-thymocyte stromal lymphopoietin (TSLP) or TSLP receptor antibody is used in or with the drug delivery device of this disclosure.

[0149] Although drug delivery devices, components, parts, subsystems, and methods have been described with reference to exemplary embodiments, they are not limited thereto. This detailed description is to be interpreted as exemplary only and does not describe every possible embodiment of this disclosure. Many alternative embodiments can be implemented using current technology or technology developed after the date of this patent application, and these embodiments still fall within the scope of the claims defining the invention disclosed herein.

[0150] Those skilled in the art will understand that various modifications, alterations, and combinations can be made to the embodiments described above without departing from the spirit and scope of the invention disclosed herein, and such modifications, alterations, and combinations are considered to be within the scope of the inventive concept.

[0151] Additional considerations

[0152] While this disclosure sets forth detailed descriptions of many different embodiments, it should be understood that the legal scope of this specification is defined by the words of the claims set forth at the end of this patent and their equivalents. This detailed description is to be construed as exemplary only and does not describe every possible embodiment, as describing every possible embodiment would be impractical. Many alternative embodiments may be implemented using current technology or technology developed after the filing date of this patent, which still fall within the scope of these claims.

[0153] The following additional considerations apply to the foregoing discussion. Throughout this specification, multiple instances can implement components, operations, or structures described as single instances. Although the various operations of one or more methods are shown and described as separate operations, one or more of the various operations can be performed simultaneously, and the operations do not need to be performed in the order shown. Structures and functions presented as separate components in the example configuration can be implemented as composite structures or components. Similarly, structures and functions presented as single components can be implemented as separate components. These and other variations, modifications, additions, and improvements all fall within the scope of this document.

[0154] Additionally, certain embodiments herein are described as including logic or a plurality of routines, subroutines, applications, or instructions. These may constitute software (e.g., code embodied on a machine-readable medium or in transmitted signals) or hardware. In hardware, routines, etc., are tangible units capable of performing certain operations and can be configured or arranged in a certain manner. In example embodiments, one or more computer systems (e.g., standalone client or server computer systems) or one or more hardware modules (e.g., processors or a group of processors) of a computer system may be configured by software (e.g., an application or an application portion) to operate as hardware modules to perform certain operations as described herein.

[0155] In various embodiments, hardware modules can be implemented mechanically or electronically. For example, a hardware module may include a dedicated circuit system or logic that is permanently configured (e.g., as a dedicated processor, such as a field-programmable gate array (FPGA), or as an application-specific integrated circuit (ASIC)) to perform specific operations. A hardware module may also include programmable logic or circuit systems that are temporarily configured by software to perform specific operations (e.g., as contained within a general-purpose processor or other programmable processor). It will be understood that cost and time considerations can drive the decision to implement the hardware module mechanically in a dedicated and permanently configured circuit system or in a temporarily configured circuit system (e.g., configured by software).

[0156] Accordingly, the term "hardware module" should be understood to include tangible entities, meaning entities that are physically constructed, permanently configured (e.g., hardwired), or temporarily configured (e.g., programmed) to operate in a particular manner or perform the specific operations described herein. Consider embodiments where hardware modules are temporarily configured (e.g., programmed), and each hardware module does not need to be configured or instantiated at any given time. For example, in cases where the hardware modules include a general-purpose processor configured using software, the general-purpose processor can be configured as distinct hardware modules at different times. The software can accordingly configure the processor, for example, to constitute a specific hardware module at one time and different hardware modules at different times.

[0157] The term "coupled to" as used in this article does not require direct coupling or connection, but rather allows two items to be "coupled" to each other through one or more intermediate components or other elements (such as electronic buses, wires, mechanical parts, or other such indirect connections).

[0158] Hardware modules can provide and receive information from other hardware modules. Therefore, the described hardware modules can be considered communicatively coupled. When multiple such hardware modules exist simultaneously, communication can be achieved through signal transmission connecting the hardware modules (e.g., via appropriate circuitry and buses). In embodiments where multiple hardware modules are configured or instantiated at different times, such communication between hardware modules can be achieved, for example, by storing and retrieving information in a memory structure accessible to the multiple hardware modules. For example, one hardware module can perform an operation and store the output of that operation in a storage device communicatively coupled to it. Another hardware module can then access the storage device at a later time to retrieve and process the stored output. Hardware modules can also initiate communication with input or output devices and can operate on resources (e.g., collections of information).

[0159] The various operations of the example methods described herein can be performed, at least in part, by one or more processors that are temporarily configured (e.g., by software) or permanently configured to perform the relevant operations. Whether temporarily or permanently configured, such processors can constitute processor-implemented modules that operate to perform one or more operations or functions. In some example embodiments, the modules mentioned herein may include processor-implemented modules.

[0160] Similarly, the methods or routines described herein can be implemented at least in part by a processor. For example, at least some operations of the method can be performed by one or more processors or hardware modules implemented by processors. The execution of certain operations can be distributed across one or more processors, residing not only within a single machine but also deployed across multiple machines. In some example embodiments, one or more processors may reside in a single location, while in other embodiments, processors may be distributed across multiple locations.

[0161] The execution of certain operations in an operation can be distributed across one or more processors, residing not only within a single machine but also deployed across multiple machines. In some example embodiments, one or more processors or processor-implemented modules may reside in a single geographic location (e.g., within a home environment, office environment, or server cluster). In other embodiments, one or more processors or processor-implemented modules may be distributed across multiple geographic locations.

[0162] This detailed description is to be interpreted as exemplary only and does not describe every possible embodiment, as describing every possible embodiment would be impractical if not impossible. Many alternative embodiments can be implemented by those skilled in the art using current technology or technology developed after the filing date of this application.

[0163] Those skilled in the art will understand that various modifications, alterations, and combinations can be made to the embodiments described above without departing from the scope of the invention, and such modifications, alterations, and combinations can be considered to be within the scope of the inventive concept.

[0164] Unless conventional device-plus-function language is explicitly enumerated, such as the language of "means for..." or "steps for..." as explicitly enumerated in the (multiple) claims, the patent claims at the end of this patent application are not intended to be based on 35 USC. This will be explained by 112(f). The systems and methods described herein relate to improving computer functionality and enhancing the operation of conventional computers.

[0165] Aspects of this disclosure

[0166] Aspect 1. A configurable handheld bioanalyzer for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, the configurable handheld bioanalyzer comprising: a first housing adapted for handheld operation; a first scanner carried by the first housing; a first processor communicatively coupled to the first scanner; and a first computer memory communicatively coupled to the first processor, wherein the first computer memory is configured to load a biological ensemble classification model configuration, the biological ensemble classification model configuration including a biological classification ensemble model comprising an unsupervised model and a supervised model, wherein training is performed using Raman-based spectral training data. The unsupervised model is trained to configure the unsupervised model to output a first indicator of one or more biological product types, wherein the supervised model is trained using Raman-based spectral training data to configure the supervised model to output a second indicator of one or more biological product types, wherein the biological classification ensemble model configuration further includes one or more spectral preprocessing algorithms, the first processor is configured to execute the one or more spectral preprocessing algorithms to reduce the spectral bias of the first Raman-based spectral dataset when the first processor receives a first Raman-based spectral dataset, and wherein the biological classification ensemble model is configured to execute on the first processor, the first processor being configured to (1) receive a first Raman-based spectral dataset defining a first biological product sample as scanned by the first scanner, and (2) use the biological classification ensemble model to identify the biological product type among the one or more biological product types based on the first Raman-based spectral dataset.

[0167] Aspect 2. A configurable handheld bioanalyzer as described in Aspect 1, wherein the biointegrated classification model configuration is electronically transferable to a second configurable handheld bioanalyzer, the second configurable handheld bioanalyzer comprising: a second housing adapted for handheld operation; a second scanner coupled to the second housing; a second processor communicatively coupled to the second scanner; and a second computer memory communicatively coupled to the second processor, wherein the second computer memory is configured to load the biointegrated classification model configuration, the biointegrated classification model configuration including the biointegrated classification model, wherein the biointegrated classification model is configured to execute on the second processor, the second processor being configured to (1) receive a second Raman-based spectral dataset defining a second biological product sample as scanned by the second scanner, and (2) identify the biological product type using the biointegrated classification model based on the second Raman-based spectral dataset, wherein the second biological product sample is a new sample of the biological product type.

[0168] Aspect 3. The configurable handheld bioanalyst as described in Aspect 1, wherein the spectral bias is an analyzer-to-analyst spectral bias between the first Raman-based spectral dataset and one or more other Raman-based spectral datasets of one or more corresponding other handheld bioanalysts, each of the one or more other Raman-based spectral datasets representing the biological product type, and wherein the one or more spectral preprocessing algorithms are configured to mitigate the analyzer-to-analyst spectral bias between the first Raman-based spectral dataset and the one or more other Raman-based spectral datasets.

[0169] Aspect 4. The configurable handheld bioanalyzer as described in Aspect 3, wherein the one or more spectral preprocessing algorithms include: applying a derivative transformation to the first Raman-based spectral dataset to generate a modified Raman-based spectral dataset, aligning the modified Raman-based spectral dataset on the Raman shift axis, and normalizing the modified Raman-based spectral dataset on the Raman intensity axis.

[0170] Aspect 5. A configurable handheld bioanalyzer as described in aspect 4, wherein the modified Raman-based spectral dataset is centered.

[0171] Aspect 6. A configurable handheld bioanalyzer as described in aspect 4, wherein the derivative transformation is applied to consecutive groupings of 5 to 15 Raman intensity values ​​across the Raman displacement axis.

[0172] Aspect 7. A configurable handheld bioanalyzer as described in aspect 5, wherein corresponding derivatives for consecutive groups of 5 to 15 Raman intensity values ​​are determined across these Raman displacement axes.

[0173] Aspect 8. A configurable handheld bioanalyzer as described in any one of Aspects 1 to 7, wherein the unsupervised model is configured to detect and identify variability associated with the one or more types of biological products.

[0174] Aspect 9. The configurable handheld bioanalyzer as described in aspect 8, wherein the variability includes instrument variability or batch-to-batch variability.

[0175] Aspect 10. The configurable handheld bioanalyzer of any one of Aspects 1 to 9, wherein the bioclassification ensemble model identifies the type of bioproduct when it determines that the first indicator passes a first pass / fail threshold and the second indicator passes a second pass / fail threshold.

[0176] Aspect 11. A configurable handheld bioanalyzer as described in any one of Aspects 1 to 9, wherein the first indicator output by the unsupervised model is based on whether the one or more types of biological products meet a threshold.

[0177] Aspect 12. The configurable handheld bioanalyzer as described in aspect 11, wherein the unsupervised model determines whether the output is qualified based on the threshold.

[0178] Aspect 13. A configurable handheld bioanalyzer as described in aspect 11 or 12, wherein the threshold is based on one or more of the following: a simplified Q residual, a Hotelling T-squared value, a Mahalanobis distance value, or a specific range of principal component scores.

[0179] Aspect 14. The configurable handheld bioanalyzer of any one of aspects 1 to 13, wherein the first bioproduct type and the second bioproduct type of the one or more bioproduct types have similar Raman-based spectra.

[0180] Aspect 15. The configurable handheld bioanalyzer as described in any one of Aspects 1 to 14, wherein the second indicator output by the supervised model is based on whether the one or more bioproduct types meet a bioproduct type prediction threshold.

[0181] Aspect 16. The configurable handheld bioanalyzer as described in aspect 15, wherein the supervisory model predicts a threshold output for qualification or non-qualification based on the type of biological product.

[0182] Aspect 17. The configurable handheld bioanalyzer of any one of Aspects 1 to 16, wherein the computer memory is configured to load a new biological classification ensemble model, the new biological classification ensemble model comprising an updated unsupervised model and / or an updated supervised model.

[0183] Aspect 18. A configurable handheld bioanalyzer as described in any one of Aspects 1 to 17, wherein the integrated biological classification model is configured in Extensible Markup Language (XML) format.

[0184] Aspect 19. The configurable handheld bioanalyzer as described in any one of Aspects 1 to 18, wherein the bioproduct type is a therapeutic product.

[0185] Aspect 20. A configurable handheld bioanalyzer as described in any one of aspects 1 to 19, wherein the type of bioproduct is identified by the bioclassification integration model during the manufacture of a bioproduct having that type of bioproduct.

[0186] Aspect 21. The configurable handheld bioanalyzer as described in any one of Aspects 1 to 20, wherein the supervised model of the bioclassification ensemble model is configured to distinguish the first bioproduct sample having the bioproduct type from different bioproduct samples having different bioproduct types.

[0187] Aspect 22. The configurable handheld bioanalyzer as described in aspect 21, wherein the bioproduct type and the different bioproduct types each have different local characteristics within similar Raman spectral ranges.

[0188] Aspect 23. The configurable handheld bioanalyzer as described in any one of Aspects 1 to 22, wherein the integrated biological classification model is generated by a remote processor located away from the configurable handheld bioanalyzer.

[0189] Aspect 24. A configurable handheld bioanalyzer as described in any one of Aspects 1 to 23, wherein the configuration of the unsupervised model is based on: principal component analysis (PCA), Euclidean distance or correlation; neighborhood-based algorithms, K-means algorithms, quality threshold (QT) algorithms, centroid algorithms, Ward algorithms or fuzzy C-means clustering algorithms.

[0190] Aspect 25. The configurable handheld bioanalyzer as described in aspect 24, wherein the unsupervised model is a PCA model, and wherein the PCA model comprises a set of simplified principal components.

[0191] Aspect 26. A configurable handheld bioanalyzer as described in any one of Aspects 1 to 25, wherein the supervised model is trained using Partial Least Squares Discriminant Analysis (PLSDA), Linear Discriminant Analysis (LDA), K Nearest Neighbor (KNN) algorithm, Soft Independent Modeling with Analogous Comparison (SIMCA) or Logistic Regression Discriminant Analysis (LREGDA) algorithm.

[0192] Aspect 27. The configurable handheld bioanalyzer as described in aspect 26, wherein the supervised model is a PLSDA model, and wherein the PLSDA model includes a simplified set of latent variables.

[0193] Aspect 28. A configurable handheld bioanalyzer as described in any one of Aspects 1 to 27, wherein the unsupervised model is configured based on principal component analysis (PCA) and the supervised model is configured based on partial least squares discriminant analysis (PLSDA).

[0194] Aspect 29. The configurable handheld bioanalyzer of any one of Aspects 1 to 28, wherein the one or more spectral preprocessing algorithms are executed to modify at least one of: (a) training data used to train the supervised model or one or both of the unsupervised model; or (b) production data used to generate outputs from the supervised model or one or both of the unsupervised model.

[0195] Aspect 30. A bioanalytical method for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, the bioanalytical method comprising: loading a biological integrated classification model configuration into a first computer memory of a first configurable handheld bioanalyzer having a first processor and a first scanner, the biological integrated classification model configuration including an unsupervised model and a supervised model, wherein the unsupervised model is trained using Raman-based spectral training data to configure the unsupervised model to output a first indicator of one or more biological product types, and wherein the supervised model is trained using Raman-based spectral training data to configure the supervised model to output a second indicator of the one or more biological product types; receiving at the first processor a first Raman-based spectral dataset, such as that of a first biological product sample defined by the first scanner; executing by the first processor one or more spectral preprocessing algorithms, such as those specified by the biological integrated classification model configuration, to reduce spectral bias in the first Raman-based spectral dataset; and identifying biological product types using the biological integrated classification model based on the first Raman-based spectral dataset.

[0196] Aspect 31. The bioanalysis method of aspect 30, further comprising: transferring the biointegrated classification model configuration to a second configurable handheld bioanalyzer; loading the biointegrated classification model configuration into a second computer memory, the biointegrated classification model configuration including the biointegrated classification model; a second processor of the second configurable handheld bioanalyzer receiving a second Raman-based spectral dataset defining a second biological product sample as scanned by the second scanner; and identifying the biological product type by the second processor implementing the biointegrated classification model based on the second Raman-based spectral dataset, wherein the second biological product sample is a new sample of the biological product type.

[0197] Aspect 32. The bioanalytical method of aspect 30, wherein the spectral bias is an analyzer-to-analyr spectral bias between the first Raman-based spectral dataset and one or more other Raman-based spectral datasets of one or more corresponding other handheld bioanalyzers, each of the one or more other Raman-based spectral datasets representing the type of biological product, and wherein the one or more spectral preprocessing algorithms are configured to mitigate the analyzer-to-analyr spectral bias between the first Raman-based spectral dataset and the one or more other Raman-based spectral datasets.

[0198] Aspect 33. The bioanalysis method as described in aspect 32, wherein the one or more spectral preprocessing algorithms include: applying a derivative transformation to the first Raman-based spectral dataset to generate a modified Raman-based spectral dataset, aligning the modified Raman-based spectral dataset on the Raman shift axis, and normalizing the modified Raman-based spectral dataset on the Raman intensity axis.

[0199] Aspect 34. The bioanalytical method as described in aspect 33, wherein the modified Raman-based spectral dataset is centered.

[0200] Aspect 35. The bioanalytical method as described in aspect 33, wherein the derivative transformation is applied to consecutive groupings of 5 to 15 Raman intensity values ​​across the Raman displacement axis.

[0201] Aspect 36. The bioanalytical method as described in aspect 35, wherein the corresponding derivatives of consecutive groups of 5 to 15 Raman intensity values ​​are determined across these Raman displacement axes.

[0202] Aspect 37. The bioanalytical method of any one of Aspects 30 to 36, wherein the unsupervised model is configured to detect variability associated with the identification of one or more types of biological products.

[0203] Aspect 38. The bioanalytical method as described in aspect 37, wherein the variability includes instrument variability or batch-to-batch variability.

[0204] Aspect 39. The bioanalytical method of any one of Aspects 30 to 38, wherein the bioclassification ensemble model identifies the type of bioproduct when the first indicator passes a first pass / fail threshold and the second indicator passes a second pass / fail threshold.

[0205] Aspect 40. The bioanalytical method as described in any one of Aspects 30 to 38, wherein the first indicator output by the unsupervised model is based on whether the one or more types of biological products meet a threshold.

[0206] Aspect 41. The bioanalytical method as described in aspect 40, wherein the unsupervised model outputs a pass / fail determination based on the threshold.

[0207] Aspect 42. The bioanalytical method as described in aspect 40 or 41, wherein the threshold is based on one or more of the following: a simplified Q residual, a Hotelling T-squared value, a Mahalanobis distance value, or a specific range of principal component scores.

[0208] Aspect 43. The bioanalytical method of any one of aspects 30 to 42, wherein the first bioproduct type and the second bioproduct type of the one or more bioproduct types have similar Raman-based spectra.

[0209] Aspect 44. The bioanalytical method of any one of Aspects 30 to 43, wherein the second indicator output by the supervised model is based on whether the one or more bioproduct types meet a bioproduct type prediction threshold.

[0210] Aspect 45. The bioanalytical method as described in aspect 44, wherein the supervised model predicts a threshold output for qualification or non-qualification based on the type of the bioproduct.

[0211] Aspect 46. The bioanalytical method of any one of Aspects 30 to 45, wherein the computer memory is configured to load a new biological classification ensemble model, the new biological ensemble model comprising an updated unsupervised model and / or an updated supervised model.

[0212] Aspect 47. The biological analysis method as described in any one of Aspects 30 to 46, wherein the biological classification integration model is configured to be implemented in Extensible Markup Language (XML) format.

[0213] Aspect 48. The bioanalytical method as described in any one of Aspects 30 to 47, wherein the type of bioproduct is a therapeutic product.

[0214] Aspect 49. The bioanalytical method of any one of Aspects 30 to 48, wherein the type of bioproduct is identified by the bioclassification integration model during the manufacture of a bioproduct having the type of bioproduct.

[0215] Aspect 50. The bioanalytical method of any one of Aspects 30 to 49, wherein the supervised model of the bioclassification ensemble model is configured to distinguish the first bioproduct sample having the bioproduct type from different bioproduct samples having different bioproduct types.

[0216] Aspect 51. The bioanalytical method as described in aspect 50, wherein the type of bioproduct and the different types of bioproducts each have different local characteristics within similar Raman spectral ranges.

[0217] Aspect 52. The bioanalytical method of any one of Aspects 30 to 51, wherein the bioclassification ensemble model is generated by a remote processor located away from the configurable handheld bioanalyzer.

[0218] Aspect 53. The bioanalytical method as described in any one of Aspects 30 to 52, wherein the configuration of the unsupervised model is based on: principal component analysis (PCA), Euclidean distance or correlation; neighborhood-based algorithm, K-means algorithm, quality threshold (QT) algorithm, centroid algorithm, Ward algorithm or fuzzy C-means clustering algorithm.

[0219] Aspect 54. The bioanalytical method as described in aspect 53, wherein the unsupervised model is a PCA model, and wherein the PCA model comprises a set of simplified principal components.

[0220] Aspect 55. The biological analysis method as described in any one of Aspects 30 to 54, wherein the supervised model is trained using Partial Least Squares Discriminant Analysis (PLSDA), Linear Discriminant Analysis (LDA), K-Nearest Neighbor (KNN) algorithm, Soft Independent Modeling with Analogous Comparison (SIMCA) or Logistic Regression Discriminant Analysis (LREGDA) algorithm.

[0221] Aspect 56. The bioanalytical method as described in aspect 55, wherein the supervised model is a PLSDA model, and wherein the PLSDA model includes a simplified set of latent variables.

[0222] Aspect 57. The bioanalytical method as described in any one of Aspects 30 to 56, wherein the unsupervised model is configured based on principal component analysis (PCA) and the supervised model is configured based on partial least squares discriminant analysis (PLSDA).

[0223] Aspect 58. The bioanalytical method of any one of Aspects 30 to 57, wherein the one or more spectral preprocessing algorithms are performed to modify at least one of: (a) training data used to train the supervised model or one or both of the unsupervised model; or (b) production data used to generate outputs from the supervised model or one or both of the unsupervised model.

[0224] Aspect 59. A tangible, non-transitory, computer-readable medium storing instructions for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, the instructions, when executed by one or more processors of a configurable handheld bioanalyzer, causing the one or more processors of the configurable handheld bioanalyzer to perform the following operations: loading a biological integrated classification model configuration into a first computer memory of a first configurable handheld bioanalyzer having a first processor and a first scanner, the biological integrated classification model configuration comprising a biological classification integrated model including an unsupervised model and a supervised model, wherein the unsupervised model is trained using Raman-based spectral training data to identify the unsupervised product based on Raman spectroscopy. The supervised model is configured to output a first indicator of one or more biological product types, wherein the supervised model is trained using Raman-based spectral training data to configure the supervised model to output a second indicator of the one or more biological product types; a first Raman-based spectral dataset defining a first biological product sample as scanned by the first scanner is received at the first processor; one or more spectral preprocessing algorithms as specified by the biological ensemble classification model are executed by the first processor to reduce spectral bias in the first Raman-based spectral dataset; and the biological product type is identified using the biological ensemble model based on the first Raman-based spectral dataset.< / model> < / model> < / model> < / model>

Claims

1. The foregoing aspects of this disclosure are merely exemplary and are not intended to limit the scope of this disclosure. A configurable handheld bioanalyzer for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, the configurable handheld bioanalyzer comprising: A first housing, which is adapted for handheld operation; A first scanner, which is carried by the first housing; A first processor, communicatively coupled to the first scanner; and A first computer memory, communicatively coupled to the first processor. The first computer memory is configured to load a biological ensemble classification model configuration, which includes a biological ensemble model comprising unsupervised and supervised models. Specifically, the unsupervised model is trained using Raman-based spectral training data to configure it to output a first indicator for one or more types of biological products. Specifically, the supervised model is trained using Raman-based spectral training data to configure it to output a second indicator for one or more types of biological products. The biological classification ensemble model configuration further includes one or more spectral preprocessing algorithms. The first processor is configured to execute these one or more spectral preprocessing algorithms when it receives a first Raman-based spectral dataset to reduce the spectral bias of the first Raman-based spectral dataset. The biological classification ensemble model is configured to execute on the first processor, which is configured to (1) receive a first Raman-based spectral dataset of a first biological product sample as defined by the first scanner, and (2) use the biological classification ensemble model to identify the biological product type of one or more biological product types based on the first Raman-based spectral dataset.

2. The configurable handheld bioanalyzer as described in claim 1, wherein, The configuration of this integrated biological classification model can be electronically transferred to a second configurable handheld bioanalyzer, which includes: A second housing adapted for handheld operation; A second scanner, which is coupled to the second housing; A second processor, communicatively coupled to the second scanner; and A second computer memory, communicatively coupled to the second processor. The second computer memory is configured to load the biological classification ensemble model configuration, which includes the biological classification ensemble model, wherein the biological classification ensemble model is configured to execute on the second processor, which is configured to (1) receive a second Raman-based spectral dataset defining a second biological product sample as scanned by the second scanner, and (2) identify the biological product type using the biological classification ensemble model based on the second Raman-based spectral dataset. The second biological product sample is a new sample of this type of biological product.

3. The configurable handheld bioanalyzer as described in claim 1, wherein, The spectral bias is the analyzer-to-analyst spectral bias between the first Raman-based spectral dataset and one or more other corresponding Raman-based spectral datasets from one or more other handheld bioanalysts, each of which represents the biological product type. The one or more spectral preprocessing algorithms are configured to mitigate analyzer-to-analyzer spectral bias between the first Raman-based spectral dataset and the one or more other Raman-based spectral datasets.

4. The configurable handheld bioanalyzer as described in claim 3, wherein, The one or more spectral preprocessing algorithms include: The derivative transformation is applied to the first Raman-based spectral dataset to generate a modified Raman-based spectral dataset; Align the modified Raman-based spectral dataset along the Raman displacement axis; and The modified Raman-based spectral dataset is normalized along the Raman intensity axis.

5. The configurable handheld bioanalyzer as described in claim 4, wherein, The modified Raman-based spectral dataset is centered.

6. The configurable handheld bioanalyzer as described in claim 4, wherein, The derivative transformation is applied to consecutive groupings of 5 to 15 Raman intensity values ​​across these Raman displacement axes.

7. The configurable handheld bioanalyzer as described in claim 5, wherein, Determine the corresponding derivatives for consecutive groups of 5 to 15 Raman intensity values ​​across these Raman displacement axes.

8. The configurable handheld bioanalyzer as described in any one of claims 1 to 7, wherein, The unsupervised model is configured to detect and identify variability associated with one or more types of biological products.

9. The configurable handheld bioanalyzer as described in claim 8, wherein, This variability includes instrument variability or batch-to-batch variability.

10. The configurable handheld bioanalyzer as described in any one of claims 1 to 9, wherein, The biological classification ensemble model identifies the type of biological product when the first indicator passes a first pass / fail threshold and the second indicator passes a second pass / fail threshold.

11. The configurable handheld bioanalyzer as described in any one of claims 1 to 9, wherein, The first indicator output by the unsupervised model is based on whether one or more types of biological products meet a threshold.

12. The configurable handheld bioanalyzer as described in claim 11, wherein, The unsupervised model outputs a pass / fail judgment based on this threshold.

13. The configurable handheld bioanalyzer as described in claim 11 or 12, wherein, The threshold is based on one or more of the following: simplified Q residuals, Hotelling T-squared values, Mahalanobis distance values, or a specific range of principal component scores.

14. The configurable handheld bioanalyzer as described in any one of claims 1 to 13, wherein, The first biological product type and the second biological product type in the one or more biological product types have similar Raman-based spectra.

15. The configurable handheld bioanalyzer as described in any one of claims 1 to 14, wherein, The second indicator output by the supervised model is based on whether one or more biological product types meet the biological product type prediction threshold.

16. The configurable handheld bioanalyzer as described in claim 15, wherein, The monitoring model predicts a threshold based on the type of biological product and outputs a pass / fail judgment.

17. The configurable handheld bioanalyzer as described in any one of claims 1 to 16, wherein, The computer memory is configured to load a new biological classification ensemble model, which includes an updated unsupervised model and / or an updated supervised model.

18. The configurable handheld bioanalyzer as described in any one of claims 1 to 17, wherein, The biological classification ensemble model is configured and implemented in Extensible Markup Language (XML) format.

19. The configurable handheld bioanalyzer as described in any one of claims 1 to 18, wherein, This biological product is a therapeutic product.

20. The configurable handheld bioanalyzer as described in any one of claims 1 to 19, wherein, During the manufacturing of biological products of this type, the biological product type is identified using this biological classification ensemble model.

21. The configurable handheld bioanalyzer as described in any one of claims 1 to 20, wherein, The supervised model of the biological classification ensemble model is configured to distinguish the first biological product sample with the biological product type from different biological product samples with different biological product types.

22. The configurable handheld bioanalyzer as described in claim 21, wherein, The biological product type and the different biological product types each have different local characteristics within similar Raman spectral ranges.

23. The configurable handheld bioanalyzer as described in any one of claims 1 to 22, wherein, The biological classification ensemble model was generated by a remote processor located away from the configurable handheld bioanalyzer.

24. The configurable handheld bioanalyzer as described in any one of claims 1 to 23, wherein, The configuration of this unsupervised model is based on: Principal Component Analysis (PCA), Euclidean distance or correlation; neighborhood-based algorithms, K-means algorithm, quality threshold (QT) algorithm, centroid algorithm, Ward algorithm or fuzzy C-means clustering algorithm.

25. The configurable handheld bioanalyzer as described in claim 24, wherein, The unsupervised model is a PCA model, and the PCA model includes a set of simplified principal components.

26. The configurable handheld bioanalyzer as described in any one of claims 1 to 25, wherein, The supervised model is trained using Partial Least Squares Discriminant Analysis (PLSDA), Linear Discriminant Analysis (LDA), K Nearest Neighbor (KNN) algorithm, Soft Independent Modeling with Analogy (SIMCA) or Logistic Regression Discriminant Analysis (LREGDA) algorithm.

27. The configurable handheld bioanalyzer as described in claim 26, wherein, The supervised model is a PLSDA model, which includes a simplified set of latent variables.

28. The configurable handheld bioanalyzer as described in any one of claims 1 to 27, wherein, The unsupervised model is configured based on principal component analysis (PCA), while the supervised model is configured based on partial least squares discriminant analysis (PLSDA).

29. The configurable handheld bioanalyzer as described in any one of claims 1 to 28, wherein, The one or more spectral preprocessing algorithms are executed to modify at least one of the following: (a) training data used to train the supervised model or the unsupervised model or both; or (b) production data used to generate outputs from the supervised model or the unsupervised model or both.

30. A bioanalytical method for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, the bioanalytical method comprising: A biological ensemble classification model configuration is loaded into the first computer memory of a first configurable handheld biological analyzer having a first processor and a first scanner. This biological ensemble classification model configuration includes a biological classification ensemble model comprising unsupervised and supervised models. Specifically, the unsupervised model is trained using Raman-based spectral training data to configure it to output a first indicator for one or more types of biological products. Specifically, the supervised model is trained using Raman-based spectral training data to configure the supervised model to output a second indicator for one or more types of biological products. The first processor receives a first Raman-based spectral dataset of a first biological product sample as defined by the first scanner. The first processor executes one or more spectral preprocessing algorithms, as specified by the biological ensemble classification model configuration, to reduce the spectral bias of the first Raman-based spectral dataset; and This biological classification ensemble model is used to identify biological product types based on the first Raman-based spectral dataset.

31. The bioanalytical method of claim 30, further comprising: The configuration of the integrated biological classification model was transferred to a second configurable handheld biological analyzer; The biological classification ensemble model configuration is loaded into a second computer memory, the biological classification ensemble model configuration including the biological classification ensemble model; The second processor of the second configurable handheld bioanalyzer receives a second Raman-based spectral dataset, as defined by the second scanner, of a second biological product sample. as well as The second processor, which implements the biological classification ensemble model, identifies the type of biological product based on the second Raman-based spectral dataset. The second biological product sample is a new sample of this type of biological product.

32. The bioanalytical method as described in claim 30, wherein, The spectral bias is the analyzer-to-analyst spectral bias between the first Raman-based spectral dataset and one or more other corresponding Raman-based spectral datasets from one or more other handheld bioanalysts, each of which represents the biological product type. The one or more spectral preprocessing algorithms are configured to mitigate analyzer-to-analyzer spectral bias between the first Raman-based spectral dataset and the one or more other Raman-based spectral datasets.

33. The bioanalytical method as described in claim 32, wherein, The one or more spectral preprocessing algorithms include: The derivative transformation is applied to the first Raman-based spectral dataset to generate a modified Raman-based spectral dataset; Align the modified Raman-based spectral dataset along the Raman displacement axis; and The modified Raman-based spectral dataset is normalized along the Raman intensity axis.

34. The bioanalytical method as described in claim 33, wherein, The modified Raman-based spectral dataset is centered.

35. The bioanalytical method as described in claim 33, wherein, The derivative transformation is applied to consecutive groupings of 5 to 15 Raman intensity values ​​across these Raman displacement axes.

36. The bioanalytical method as described in claim 34, wherein, Determine the corresponding derivatives for consecutive groups of 5 to 15 Raman intensity values ​​across these Raman displacement axes.

37. The bioanalytical method according to any one of claims 30 to 36, wherein, The unsupervised model is configured to detect and identify variability associated with one or more types of biological products.

38. The bioanalytical method as described in claim 37, wherein, This variability includes instrument variability or batch-to-batch variability.

39. The bioanalytical method according to any one of claims 30 to 38, wherein, The biological classification ensemble model identifies the type of biological product when the first indicator passes a first pass / fail threshold and the second indicator passes a second pass / fail threshold.

40. The bioanalytical method according to any one of claims 30 to 38, wherein, The first indicator output by the unsupervised model is based on whether one or more types of biological products meet a threshold.

41. The bioanalytical method as described in claim 40, wherein, The unsupervised model outputs a pass / fail judgment based on this threshold.

42. The bioanalytical method as described in claim 40 or 41, wherein, The threshold is based on one or more of the following: simplified Q residuals, Hotelling T-squared values, Mahalanobis distance values, or a specific range of principal component scores.

43. The bioanalytical method according to any one of claims 30 to 42, wherein, The first biological product type and the second biological product type in the one or more biological product types have similar Raman-based spectra.

44. The bioanalytical method according to any one of claims 30 to 43, wherein, The second indicator output by the supervised model is based on whether one or more biological product types meet the biological product type prediction threshold.

45. The bioanalytical method as described in claim 44, wherein, The monitoring model predicts a threshold based on the type of biological product and outputs a pass / fail judgment.

46. ​​The bioanalytical method according to any one of claims 30 to 45, wherein, The computer memory is configured to load a new biological classification ensemble model, which includes an updated unsupervised model and / or an updated supervised model.

47. The bioanalytical method according to any one of claims 30 to 46, wherein, The biological classification ensemble model is configured and implemented in Extensible Markup Language (XML) format.

48. The bioanalytical method according to any one of claims 30 to 47, wherein, This biological product is a therapeutic product.

49. The bioanalytical method according to any one of claims 30 to 48, wherein, During the manufacturing of biological products of this type, the biological product type is identified using this biological classification ensemble model.

50. The bioanalytical method according to any one of claims 30 to 49, wherein, The supervised model of the biological classification ensemble model is configured to distinguish the first biological product sample with the biological product type from different biological product samples with different biological product types.

51. The bioanalytical method as described in claim 50, wherein, The biological product type and the different biological product types each have different local characteristics within similar Raman spectral ranges.

52. The bioanalytical method according to any one of claims 30 to 51, wherein, The biological classification ensemble model was generated by a remote processor located away from the configurable handheld bioanalyzer.

53. The bioanalytical method according to any one of claims 30 to 52, wherein, The configuration of this unsupervised model is based on: Principal Component Analysis (PCA), Euclidean distance or correlation; neighborhood-based algorithms, K-means algorithm, quality threshold (QT) algorithm, centroid algorithm, Ward algorithm or fuzzy C-means clustering algorithm.

54. The bioanalytical method as described in claim 53, wherein, The unsupervised model is a PCA model, and the PCA model includes a set of simplified principal components.

55. The bioanalytical method according to any one of claims 30 to 54, wherein, The supervised model is trained using Partial Least Squares Discriminant Analysis (PLSDA), Linear Discriminant Analysis (LDA), K Nearest Neighbor (KNN) algorithm, Soft Independent Modeling with Analogy (SIMCA) or Logistic Regression Discriminant Analysis (LREGDA) algorithm.

56. The bioanalytical method as described in claim 55, wherein, The supervised model is a PLSDA model, which includes a simplified set of latent variables.

57. The bioanalytical method according to any one of claims 30 to 56, wherein, The unsupervised model is configured based on principal component analysis (PCA), while the supervised model is configured based on partial least squares discriminant analysis (PLSDA).

58. The bioanalytical method according to any one of claims 30 to 57, wherein, The one or more spectral preprocessing algorithms are executed to modify at least one of the following: (a) training data used to train the supervised model or the unsupervised model or both; or (b) production data used to generate outputs from the supervised model or the unsupervised model or both.

59. A tangible, non-transitory, computer-readable medium storing instructions for identifying biological products using integrated artificial intelligence (AI) based on Raman spectroscopy, which, when executed by one or more processors of a configurable handheld bioanalyzer, cause the one or more processors of the configurable handheld bioanalyzer to perform the following operations: A biological ensemble classification model configuration is loaded into the first computer memory of a first configurable handheld biological analyzer having a first processor and a first scanner. This biological ensemble classification model configuration includes a biological classification ensemble model comprising unsupervised and supervised models. in, The unsupervised model was trained using Raman-based spectral training data to configure it to output a first indicator for one or more types of biological products. Specifically, the supervised model is trained using Raman-based spectral training data to configure the supervised model to output a second indicator for one or more types of biological products. The first processor receives a first Raman-based spectral dataset of a first biological product sample as defined by the first scanner. The first processor executes one or more spectral preprocessing algorithms, as specified by the biological ensemble classification model configuration, to reduce the spectral bias of the first Raman-based spectral dataset; and This biological classification ensemble model is used to identify biological product types based on the first Raman-based spectral dataset.

Images