Storing data from processes used to produce chemical, pharmaceutical, biopharmaceutical, and / or biological products.

Abstract

This application relates to a computer-implemented method, computer-readable medium, and system for storing heterogeneous sequences of discrete-time data determined from processes producing chemicals, pharmaceuticals, biopharmaceuticals, and / or biological products. The method includes receiving discrete-time data, which includes first data from a first scientific instrument. The method also includes receiving second data from a second scientific instrument. The first and second scientific instruments each have different corresponding observation types, the first data having a first observation type, and the second data having a second observation type. The method further includes: storing the first data and first metadata in a first record of a database, the first data being stored according to the function of stored data mapped by the first observation type; and storing the second data and second metadata in a second record of the database, the second data being stored according to the function of stored data mapped by the second observation type.

Description

Technical Field

[0001] This disclosure relates to controlling and monitoring processes used in the production of chemicals, pharmaceuticals, biological drugs, and / or biological products. More specifically, this disclosure relates to storing heterogeneous sequences of discrete-time data determined from said processes. This disclosure also relates to querying the stored data through a common interface (i.e., a single interface capable of querying all stored data). Background Technology

[0002] Continuous process validation (CPV) can refer to the collection and analysis of production components and process data to ensure that process outputs are within specified quality limits or key performance indicators. More specifically, CPV can be used for process validation in the pharmaceutical and / or biopharmaceutical industries. CPV ensures process control throughout its execution and can help identify and eliminate process inconsistencies. CPV can include multivariate statistical analysis, outlier event identification, and other techniques to promote consistent process performance. Therefore, a large amount of data can be collected from each process performed. The data can be diverse, especially since it can be collected from various scientific instruments (e.g., devices or machines).

[0003] It is expected that not only single measurements, such as temperature or pH measurements associated with a specific timestamp (i.e., a point in time), can be stored, but also thousands, tens of thousands, or even hundreds of thousands of data points associated with a given timestamp. Large numbers of measurements associated with a single timestamp can occur in the context of spectroscopic measurements (e.g., Raman spectroscopy or mass spectrometry) or biological sequence data (e.g., gene expression measurements). Specifically, in the context of Raman spectroscopy, thousands of data points are acquired at high frequencies. Furthermore, high-resolution mass spectrometry (e.g., using liquid chromatography coupled with an orbital trap mass spectrometer) can generate hundreds of thousands of data points per acquisition. Additionally, surface-enhanced Raman spectroscopy (SERS) can be particularly suitable for analytes at concentrations below 0.5%. However, these and other sensitive spectroscopic techniques can be difficult to replicate and may require further calibration or chemometric evaluation. Therefore, sampling inconsistencies can cause difficulties when storing data, especially since corresponding intensities may appear at different wavelengths in multiple observations.

[0004] It is also desirable to store media information associated with the process, such as images, videos, or charts. For example, it is desirable to store microscope images, i.e., images taken using a microscope. Furthermore, machine learning (e.g., deep learning) can be used for image processing to monitor cells in bioreactors (e.g., fermenters). Therefore, it is desirable to provide a database system or database management system capable of processing data generated during process monitoring and control. In particular, the system is capable of processing diverse measurement data provided by various scientific instruments. Furthermore, it is desirable to provide a common query interface for querying all data originating from the process.

[0005] Typically, time-series databases or multiple time-series databases are used to store process data. Time-series databases can be optimized to handle time-series data, i.e., sequences of data points indexed by time. Time-series databases can be implemented using specialized algorithms (e.g., Facebook Gorilla) or relational databases. Time-series databases can support Structured Query Language (SQL) queries. Furthermore, time-series databases can be implemented using predefined schemas. Additional examples of time-series databases are extremeDB, Apache Druid, and IBM Informix Time Series. Alternatively, a common approach may include an operational history database. Operational history databases are also known as data history databases or process history databases. Operational history databases efficiently store and retrieve data from time-series databases. Operational history database records can include sensor data (such as temperature, flow rate, fluid or pressure levels, readings from discrete-level sensors, and whether motors are on or off), as well as quality assurance data, alarms, and aggregated data (such as moving averages or standard deviations). Data can be timestamped and cataloged in a machine-readable format.

[0006] An example of operating a historical database is a Process Information (PI) system. PI systems support various data types, including float16, float32, float64 (16, 32, and 64-bit floating-point numbers), int16, int32 (16 and 32-bit integers), numbers (defined states), strings (text), timestamps, and blobs (binary large objects). Data streams in a PI system are called tags. Each process parameter can have one data stream. Therefore, each tag can be associated with multiple values ​​and the time each of these values ​​was recorded. For example, a process might include temperature data streams, pH data streams, and dissolved oxygen data streams, each with a corresponding tag. Typically, tags must be defined in advance (i.e., before data collection). Data stream values ​​can be stored in columns of the database. In particular, each data stream can have its own columns.

[0007] Typically, PI systems are already used to store spectral data. For example, the PI interface for Siemens Spectroscopy is commonly used to store data from the Siemens Spectral Energy Management System (EMS). Typically, when using an operational history database system (such as a PI system) to store spectral data, each wavelength requires a separate tag. Furthermore, each value is usually stored along with a corresponding timestamp, and multiple redundant timestamps are stored (e.g., for times when no values ​​were collected). Moreover, querying spectral data in such a system can be challenging because each spectral observation (e.g., emission spectrum) does not correspond to a single tag (e.g., as would be the case with temperature or pressure), but rather to a (potentially large) set of tags.

[0008] Furthermore, as discussed above, inconsistencies or the need for calibration in spectroscopic observations can lead to corresponding data points from two different observations being classified under different labels. This classification can make it difficult to draw conclusions from spectroscopic observations. Inconsistencies in spectroscopic data can arise for various reasons, particularly from digitizing continuous variables (i.e., digitizing information or converting it into a computer-readable format) into discrete values. Spectroscopic observations can correspond to continuous physical phenomena, such as molecular mass, wavelength, or chemical shift. Because labels in a PI system must be declared before use, numerical calculations such as interpolation or statistical interpolation may be required to populate the label values.

[0009] For example, Raman spectral signals are variations in fluorescence intensity as a continuous function of wavelength. Digitization of signals received from a Raman spectrometer can lead to misalignment. In particular, for different runs of a process, corresponding intensities may appear at different wavelengths. Differences in digitization between different processes can be caused by signal processing or calibration artifacts. Therefore, a collection of spectroscopic observations can each have a different set of descriptors for their corresponding spectral intensities. These differences can cause problems when attempting to store spectroscopic observations in a PI system or another database with a predefined mode (i.e., a write-time mode). For example, when storing spectral data in a PI system or another system that relies on a time-series database, a new tag may have to be created each time the spectrum is stored in the database.

[0010] Alternatively, it is necessary to perform interpolation according to a predefined scale and store the interpolated spectrum, rather than the actual spectrum collected from the process. Determining the predefined scale based on specifications that predict future needs can be very challenging. Therefore, it is expected that data corresponding to the digital signal (i.e., the raw digital signal data) can be stored without interpolation or predefined digitization rules.

[0011] Storing image or video data in a PI system can be even more problematic. In particular, it is necessary to create a label for each pixel of a color image.

[0012] Another common approach is to store each observation in its own numerical array, which is then stored in a database. An example of this approach is Extruec. Thus, the spectral data from the observations can be stored in a single array rather than as a collection of labels. This approach avoids the label alignment problems that occur in PI systems. Furthermore, arrays of spectral data are more likely to be aligned compared to multiple collections of labels corresponding to different observations. However, the size of the array can be problematic, especially in the context of modern sensitive mass spectrometers or surface-enhanced Raman spectroscopy discussed above. In such cases, the size and dimensionality of the captured data can increase beyond what would be expected in a single array, particularly if the array is limited to one dimension. Moreover, queries involving different types of observational data (e.g., spectra, biological sequence data, temperature) can lead to poor computer performance or even sudden failures in the computer executing the queries.

[0013] Therefore, conventional relational database management systems may not be suitable for storing heterogeneous process data, especially the large amounts of data required for biological processes (e.g., biopharmaceutical processes). Furthermore, storing data according to a predefined data model or schema is not desirable. Additionally, conventional databases tend to be limited to similar types of data, such as genomic data or sensor data (e.g., temperature or pH). The diversity of data collected in processes used to produce chemical, pharmaceutical, biopharmaceutical, and / or biological products may not fit into conventional storage models. Moreover, for data about a specific process, it is not desirable to use different data models to query multiple databases. Process data storage across multiple databases with different data models can lead to complex SQL queries and enormous computational and / or storage requirements.

[0014] For example, one thousand spectra might be collected during the process. Each spectrum contains 100,000 data points, each stored as a double-precision floating-point number. Therefore, for mass spectrometry data, 100,000 mass-to-charge ratios and 100,000 intensities could be stored. Thus, for a single observation, the total storage space required to store the spectrum in a time-series database (such as a PI system) could be 200,000 × 1,000 × 8 bytes = approximately 1.5 gigabytes. In some cases, it may be desirable to compare subsets of spectra collected between two different time stamps during multiple runs of the process, for example, spectra from 20 different observations. Therefore, the system might need to load 20 different time series (i.e., all data points from 20 observations) into memory to perform the comparison. Thus, when using a time-series database, such a comparison of spectra could require approximately 20 observations × 1.5 GB per observation = 30 GB of memory.

[0015] Furthermore, the amount of memory required may be independent of the actual number of spectra that need to be loaded (e.g., only a few wavelengths or wavelength ranges may need to be compared). In particular, when using time series, it may be necessary to load 1,000 spectra for the entire series for a process in order to compare a subset of the spectra (e.g., a range of 10 wavelengths) with a corresponding subset of the spectra from another process. More efficient use of memory may be desired for performing such comparisons.

[0016] Furthermore, according to some conventional solutions, the first data management system can store spectral data, including Raman spectral data acquired during the process. Another separate data management system can be used to store data from various sensors, such as temperature, pH, and dissolved oxygen. Additionally, yet another data management system can be used to store, for example, cell image data obtained using a microscope. Therefore, it is necessary to query all three data management systems to obtain data for a single process. A third data management system may be needed to store biological sequence data. Each system may have different data models and require different query attributes.

[0017] You might want to store data from multiple data sources in the same database, allowing you to retrieve data from each data source using a single query. This makes it possible to search for data from multiple sources with a single query and provides a simple and consistent interface for querying data from each data source. Summary of the Invention

[0018] According to one aspect, a computer-implemented method is provided for storing heterogeneous sequences of discrete-time data determined from processes producing chemical, pharmaceutical, biopharmaceutical, and / or biological products. The method includes receiving discrete-time data, which includes: first data from a first scientific instrument, the first data including a first timestamp corresponding to a first digital signal; and second data from a second scientific instrument, the second data including a second timestamp corresponding to a second digital signal. The first scientific instrument is different from the second scientific instrument. The method further includes storing the first data and first metadata in a first record of a database. The first record includes a first intensity field having a first data type and a first descriptor field having a second data type. The method further includes storing the second data and second metadata in a second record of the database. The second record includes a second intensity field having a first data type and a second descriptor field having a second data type. The first metadata includes a first identifier and the second metadata includes a second identifier. When the first data includes a first intensity and a first descriptor of a first digital signal, storing the second data further includes storing the first intensity in the first intensity field and the first descriptor in the first descriptor field. When the second data includes the second intensity and the second descriptor of the second digital signal, storing the first data further includes storing the second intensity in the second intensity field and storing the second descriptor in the second descriptor field.

[0019] In some examples, the method may further include: determining whether the first data includes a first intensity and a first descriptor. Additionally, the method may also include: determining whether the second data includes a second intensity and a second descriptor.

[0020] Heterogeneous data can be compared with similar data. Similar data can be obtained from a single type of scientific instrument (e.g., one or more Raman spectrometers) and / or similar data can all have the same observation type (e.g., all are spectroscopic data or all are microscopic data). Heterogeneous data can be obtained from different scientific instruments and can have different characteristics. Therefore, heterogeneous data can have different observation types. For example, heterogeneous data can include (Raman) spectroscopic data from a spectrometer and images from a microscope.

[0021] Discrete-time data sequences can be a series of observations indexed in chronological order. Specifically, the sequence can be obtained at consecutive, equally spaced time points. Each observation can be associated with a timestamp.

[0022] In some examples, the process can be a biological process. More specifically, the process can be an upstream biological process or a downstream biological process. In the case of an upstream biological process, the process can be carried out in a bioreactor. The bioreactor can be part of a process control device. Thus, bacterial or mammalian cell lines can be grown in the bioreactor. In the case of a downstream biological process, the output from the upstream biological process can be further processed (particularly through cell disruption, purification, and refinement) to meet purity and quality requirements.

[0023] In some examples, a scientific instrument (e.g., a first or second scientific instrument) can be a device for scientific purposes, such as an analytical apparatus. A scientific instrument can be a substance or molecular identification instrument, or a sensor attached to a container of a process control device. In particular, a scientific instrument can be a structural (or spectral) fingerprinting device. A scientific instrument can be a chemical or fermentation analyzer, or a measuring instrument (e.g., for measuring nutrients and / or metabolites).

[0024] In some examples, digital signals represent data as a discrete sequence of values; at any given time, the signal can only take one of a finite number of values. This contrasts with analog signals, which represent continuous values; at any given time, an analog signal represents a real number that may fall within a specified range.

[0025] Each data record may include a timestamp field for storing a timestamp. Specifically, the first data record may include a first timestamp field for storing a first timestamp, and the second data record may include a second timestamp field for storing a second timestamp.

[0026] In some examples, each timestamp (i.e., a digital timestamp) can be a sequence of characters identifying when an event occurred, possibly giving a day, a time within a day, and accurate to a fraction of a second. A timestamp corresponding to a digital signal can indicate when the signal was received or recorded. In particular, a timestamp can indicate when the signal was observed (i.e., the observation time). The observation time can differ from the acquisition or recording time (e.g., in the context of offline spectral analysis).

[0027] For example, if the image is acquired from a quenched biological sample (fluorescence has decreased), the quenching time can be recorded as a timestamp. As another example, since the observation time and data acquisition time of spectral data in an online spectral system overlap, either can be recorded as the corresponding timestamp. However, when a biological sample is acquired and sent to the laboratory for analysis (offline spectroscopy), the time of sample acquisition is the observation time, and this time can be recorded as a timestamp; the data acquisition time is different.

[0028] In the context of this disclosure, an observation result can refer to the output of a scientific instrument (e.g., a spectrometer, a biological sequencing device, or a sensor) associated with a given timestamp. Therefore, an observation result can include substantially all data from a digital signal corresponding to a timestamp. Data can correspond to a timestamp in the sense that sample processing was completed at that timestamp. Thus, an observation result can include many data points. Alternatively, for example, in the context of temperature or pH, an observation result can include a single data point.

[0029] In some examples, intensity can be a measurement result or can be derived from a measurement result. Intensity can describe the strength of the corresponding digital signal. A descriptor can describe the digital signal or a characteristic of the digital signal. For example, in the context of a spectral signal, the descriptor can be the wavelength. Data received from a scientific instrument (e.g., first data) can include additional elements. In particular, the data can include the mass-to-charge ratio, for example, in the context of a mass spectrometer. For example, the first data can include a first intensity, a first descriptor, and a first mass-to-charge ratio.

[0030] In some examples, each record (i.e., a data record) can be a set of fields in a database associated with an entity. Specifically, each record can correspond to a single observation. Each field can be used to store data. Fields can be placeholders for data or data structures, and may have variable lengths (i.e., not all fields of the same data type have the same physical size or length).

[0031] Therefore, the first strength and the second strength can each be stored in their respective strength fields, such that the first strength field and the second strength field are each strength fields with the same data type and the same structure. In other words, the first strength field is the strength field of the first data, and the second strength field is the strength field of the second data. Similarly, the first descriptor field is the descriptor field of the first data, and the second descriptor field is the descriptor field of the second data.

[0032] In the context of this disclosure, the strength field can be a field suitable for storing strength. Similarly, the descriptor field can be a field suitable for storing descriptors.

[0033] In some examples, each record may encode or encapsulate data according to a standard format or standard encoding. Exemplary encodings include Extensible Markup Language (XML), JavaScript Object Notation (JSON), and Binary JSON (BSON). Each record may have a unique key (i.e., a unique identifier). For example, a unique key may be an identifier included in the metadata such that a first identifier is the unique key for a first record, and a second identifier is the unique key for a second record.

[0034] In some examples, metadata can include the same attributes regardless of which scientific instrument the data is received from. Metadata attributes can include one or more of the following: unique identifier, operator, protocol, instrument identifier, and project identifier. Even when the data is not human-readable, including common elements in metadata attributes can facilitate metadata retrieval and finding the corresponding stored data.

[0035] Each data type can constrain the values ​​that can be stored in a record. Furthermore, each data type can define operations that can be performed on data of that data type.

[0036] The following description provides examples that can be combined with aspects of methods, systems, and computer programs. Therefore, even though method steps are described or the term "method" is used, the specific implementation described can be used in the context of a system or computer program.

[0037] In some cases, a database can be "not just SQL" (NoSQL). A database can be read-time schema or free schema. Specifically, a database can be a key-value store, preferably a document store. Read-time schema can refer to a method of creating a schema when data is read from the database. Therefore, a certain structure is applied to the data only when it is read, allowing data to be stored in the database without first specifying a schema. Read-time schema contrasts with write-time schema, which requires creating a schema for the data before it is written to the database. Write-time schema is commonly used in the context of relational databases and Structured Query Language (SQL).

[0038] Write-time consistency ensures complete consistency of data stored in a database. In contrast, read-time consistency allows for some degree of inconsistency in the data stored in the database. Consistency can be applied as data is used. Specifically, a database can provide eventual consistency (i.e., optimistic replication), meaning that if no new updates are made to a stored data item, eventually all accesses to that item will return the most recently updated value.

[0039] Using NoSQL to implement a database can make the implementation of metadata more efficient. For example, storing metadata implemented through key-value pairs in an RDBMS may be inefficient.

[0040] In some cases, the first and second scientific instruments can be used for chemical and / or material analysis.

[0041] Each scientific instrument may include one of the following (or consist of): a spectrometer, a flow cytometer, a physiological data collection device, a biological sequence data collection device, or a microscope.

[0042] When the scientific instrument includes a spectrometer, the corresponding data of the scientific instrument may include spectral data; the spectral data may include wavelengths and spectral values ​​corresponding to the wavelengths. A corresponding descriptor may be a wavelength and / or a corresponding intensity may be a spectral value corresponding to the wavelength. When the scientific instrument includes a biological sequence data collection device, the corresponding data may include biological sequence data (e.g., gene expression data). A corresponding descriptor may be a gene identifier and a corresponding intensity may reflect protein expression (e.g., specifying the amount of protein).

[0043] In some examples, a spectrometer can be a device used to separate and measure the spectral components of a physical phenomenon. Spectrometers can use various techniques to analyze materials (e.g., cell cultures, cell culture media, raw materials). Specifically, a spectrometer can use one or more of the following: near-infrared (NIR), mid-infrared (MIR), Raman fluorescence, and nuclear magnetic resonance (NMR) spectroscopy. A spectrometer can perform vibrational spectroscopy, optical spectroscopy, or mass spectrometry. Other techniques may also be used.

[0044] In some examples, biological sequence data can be nucleotide sequences in DNA or RNA that encode functional molecules. Genes in the sequence data can be expressed by transcription into RNA, and that transcript can then be translated into proteins.

[0045] In some cases, data from each scientific instrument may have different corresponding observation types. Therefore, first data may have a first observation type and second data may have a second observation type. Each observation type can be associated with a function used to process data from the corresponding scientific instrument. Specifically, each observation type can be associated with a function used to extract, translate, transform, analyze, interpret, transfer, and / or display data of that type.

[0046] Each record in the database (e.g., the first record and / or the second record) may include a type field that specifies the observation type of the corresponding scientific instrument from which data in the record received from it is stored. The type field can be used to identify the scientific instrument from which data in the record storing the type field is received.

[0047] For example, data from a spectrometer may have a first observation type, data from a microscope may have a second observation type, and data from a biological sequence data collection device may have a third observation type.

[0048] Therefore, by retrieving the corresponding display function using the observation type, the corresponding data from a spectrometer (e.g., a Raman spectrometer) can be displayed in an xy-coordinate graph (e.g., where intensity is the y-value and descriptor is the x-value). Similarly, corresponding data from biological sequence data collection devices can be mapped onto existing biological networks. In particular, the intensity of gene expression can be mapped to nodes in the network.

[0049] The first data may consist of all data provided by a first scientific instrument corresponding to a first timestamp. The second data may consist of all data provided by a second scientific instrument corresponding to a second timestamp. The first data may represent a first digital signal at the first timestamp. The second data may represent a second digital signal at the second timestamp.

[0050] The terms "first" and "second" refer only to the order in which the data is received. Alternatively, data from a second scientific instrument can be processed as first data, and data from a first scientific instrument can be processed as second data. Therefore, it can be advantageous for the database to store data from any scientific instrument in a space-saving manner.

[0051] Furthermore, data provided by a second scientific instrument can also be provided in the form of descriptors and intensities. For example, image data from a microscope can be provided in the form of descriptors and intensities, for instance, if the image from the microscope is uncompressed. In this case, the descriptor can indicate the intensity corresponding to the image, and the intensity can include pixel information, such as in the form of a multidimensional array of color values.

[0052] The corresponding scientific instrument in scientific instruments refers to either the first scientific instrument or the second scientific instrument. The term "corresponding descriptor" refers to either the first descriptor or the second descriptor. Similarly, the term "corresponding intensity" refers to either the first intensity or the second intensity.

[0053] Each record may also include a reference field. When the corresponding data from one of the scientific instruments does not include the corresponding intensity or corresponding descriptor, the method may further include storing the corresponding data in a cluster file system, preferably a distributed file system, more preferably Apache Hadoop. The method may also include storing the location of the corresponding data in the reference field. The format of the corresponding data may have one or more of the following characteristics: the format is not human-readable, the format is binary, the format is an encoded content format, and the format is a compressed format (e.g., an image or video compression format). When the corresponding data from one of the scientific instruments includes the corresponding intensity and corresponding descriptor as well as additional data, the method may further include storing the additional data in a cluster file system and storing the location of the additional data in the reference field.

[0054] When the format is not human-readable, the corresponding data can include content that is not a Unicode code point. Therefore, reference fields can be used to integrate non-human-readable data (e.g., coded data of compressed images as discussed above) with human-readable data (e.g., spectral data or gene sequence data) in a database. Non-human-readable data is typically part of observations that do not include intensity and descriptors. Therefore, reference fields can make data queries more efficient because data in non-human-readable formats can be queried simultaneously with human-readable data (i.e., through the same user interface).

[0055] The data referenced by the reference field can also be partially human-readable, such as a Portable Document Format (PDF) file or a Microsoft Word document. Reference fields can also be used to temporarily reference data for which parsing functionality is not available (e.g., data without associated observation types). Once parsing functionality is available, the data can be imported as descriptors and strengths.

[0056] Therefore, not all data from a first scientific instrument has a first data type or a second data type. For example, a first scientific instrument may also provide images, and the location of the image may be stored in a reference field.

[0057] In some examples, the location stored in the reference field can be represented using a Uniform Resource Identifier (URI) (i.e., a string that explicitly identifies the resource). In particular, the location can uniquely identify the file's network location on the cluster file system.

[0058] In some examples, a clustered file system can be shared by mounting it on multiple servers simultaneously. A clustered file system can provide location-independent addressing and redundancy. The file system can include master and worker nodes. The file system can provide job tracking and / or scheduling.

[0059] Additional data may be formatted documents, such as the manufacturer's specifications for the corresponding scientific instrument in scientific instruments (e.g., documents for spectrometers or microscopes).

[0060] When the corresponding data from one of the scientific instruments does not include the corresponding intensity or corresponding descriptor, the corresponding data may include formatted text documents or media (such as images, videos, or charts). When the corresponding data includes images, the images may come from a microscope. Images or videos may be in various formats and may be compressed (e.g., portable web graphics) or uncompressed (e.g., bitmap). More specifically, images may be in raster graphics format. Formatted text documents may be styled text documents or rich text documents. Formatted text may contrast with plain text and may refer to including style information such as color, text style (e.g., bold, italic), size, highlighting, etc.

[0061] In some examples, a first identifier can identify the first sample from which the first intensity originates. A second identifier can identify the second sample from which the second intensity originates. The first and second samples can be different samples. Each sample can be biological. For example, the first and / or second sample can be derived from cell cultures.

[0062] The first metadata and / or the second metadata may also include at least one of the following:

[0063] -User or operator identifier,

[0064] -Project identifier,

[0065] - Sample collection time

[0066] - Cell line identifier

[0067] - Growth medium identifier,

[0068] -Protocol identifier,

[0069] -Note.

[0070] The first metadata and / or the second metadata can be characterized as descriptive metadata, that is, metadata that can describe the corresponding data and / or provide the instrument with the corresponding data.

[0071] The method may also include providing a user interface to query first data, second data, first metadata, and second metadata. The user interface can be implemented using schema and data manipulation tools.

[0072] In some cases, it may not be possible to query the data itself through the user interface, for example, if the data is not in a human-readable format. In such cases, the user interface may be limited to querying the metadata corresponding to the data. When the data includes corresponding strengths and descriptors, the data itself can be queried in addition to the corresponding metadata. In particular, when the data is in a human-readable format (e.g., encoded in U.S. Standard Code for Information Interchange - ASCII - or Unicode text), the user interface can be used to query the data.

[0073] Schema and data manipulation tools can include Object Relational Mappers (ORMs), such as the Django ORM. Alternatively, a SQL query engine, such as Apache Drill, can be used to implement the user interface.

[0074] Although ORMs are not typically used without an SQL database, they can be modified to provide a user interface for retrieving relevant data from one of the scientific instruments. The necessary modifications can be known to a technician.

[0075] In some examples, schema and data manipulation tools can provide mechanisms for creating, reading, updating, and deleting data and data schemas. SQL query engines can support queries without specifying the schema in advance (e.g., free schema or read-time schema). SQL query engines can provide distributed query optimization, columnar execution, optimistic execution, and pipelining execution. Schema and data manipulation tools can also support querying and schema management within clustered file systems.

[0076] When the data from a scientific instrument includes the intensity and descriptor, the data can be in a human-readable format. Specifically, the data can include ASCII, Unicode, or markup language text.

[0077] Discrete-time data may also include third-party data. Third-party data may include a single value determined by a process control device used to control the process. This single value may have a corresponding timestamp.

[0078] The process control device can also be configured to monitor the process. A single value can be determined by a sensor of the process control device. For example, a single value could be temperature determined by a temperature sensor or pH value determined by a pH sensor. A single value can be measured. A descriptor can be determined based on data received from the process control device. For example, when the single value is temperature, the process control device can provide an indication that the single value is temperature and can determine the corresponding descriptor.

[0079] Process control devices (e.g., automated bioreactor systems) can be first-scale process control devices, serving as platforms for early process development at a first scale (e.g., small or microscale, up to 1 liter). More specifically, process control devices can be used to develop biotechnology pharmaceuticals. After the process is executed at the first scale, it may be desirable to transfer the process to a second (e.g., larger or macroscopic) scale. At both the first and second scales, the process can follow common protocols and maintain the Quality Target Product Profile (QTPP).

[0080] The process control device may include a plurality of first-scale containers, each containing a fluid for producing a product. The fluid may include starting materials for the process. More specifically, the fluid may include culture media and / or biological materials (e.g., cell cultures). The containers may be controlled by the process control device, and more specifically, the containers may be contained by the process control device.

[0081] The process performed in each first-scale container can be designed to produce the same product. There can be at least 10, 20, or at least 40 containers, preferably 12, 24, or 48 containers.

[0082] The method may further include determining a descriptor for the third data. The method may also include storing the third data and third metadata in a third record, the third record including a third strength field having a first data type and a third descriptor field having a second data type. Storage may include storing a single value in the third strength field and storing the descriptor in the third descriptor field. The third metadata may include a third identifier.

[0083] The third data can be scale-independent and / or sampling-independent. The third data can be one or more of the following: pH, temperature, optical density, oxygen, and partial pressure.

[0084] The term "sampling-related" can refer to data measured through sampling. Sampling-related data can include offline analysis. In particular, determining sampling-related data may require sampling fluid from a container in a process control device, for example, for analysis by a spectrometer. Sampling-irrelevant data may not require sampling or offline analysis. For example, determining sampling-irrelevant data may involve remotely interrogating sensor points on a container.

[0085] The term "scale-independent" can refer to data that is independent of the process scale. Therefore, scale-independent data can be expected to have the same (or similar) values ​​at a first scale (e.g., 100 ml) and a second scale (e.g., 50 L), with a difference at least an order of magnitude from the first scale. Scale-dependent data can vary across different scales. Therefore, the values ​​of scale-dependent data can differ between the first and second scales. For a given process, the stirring rate or hydrostatic pressure of the vessel can be scale-dependent. For a given process, temperature can be scale-independent.

[0086] Being able to store both sample-related and sample-irrelevant data in the same database can be advantageous. In particular, data obtained using a spectrometer can be sample-related. Similarly, biological sequence data (e.g., gene expression data) can be sample-related (e.g., offline analysis of samples).

[0087] The first and / or second data types can include (are) composite data types. Composite (non-primitive) data types are sometimes referred to as composite data types, structured data types, or aggregate data types. Composite data types can be composed of a variety of other data types. In particular, composite data types can be composed of primitive data types and / or other composite data types. Composite data types can also be distinguished from basic types or SQL built-in types.

[0088] In some cases, the first and / or second data types may include (are) linear data structures. Therefore, the elements of the first and / or second data types may include sequences of elements. Linear data structures (such as arrays and lists) can be contrasted with non-linear data structures (such as trees and heaps). The first and / or second data types may include arrays. Each data type may specify the same number of dimensions. For example, if the first data type includes a two-dimensional array, then the second data type may include a two-dimensional array.

[0089] According to another aspect, a computer program comprising computer-readable instructions is provided. When loaded and executed on a computer system, the instructions cause the computer system to perform operations according to the methods described above. The computer program can be implemented in the form of a computer program product. In some cases, the computer program can be tangibly embodied in a computer-readable medium.

[0090] According to another aspect, a system is provided for storing heterogeneous sequences of discrete-time data determined from processes for producing chemical, pharmaceutical, biopharmaceutical, and / or biological products. The system includes a first interface to a first scientific instrument, a second interface to a second scientific instrument, and a database. The database is configured to receive discrete-time data via the first and second interfaces, the discrete-time data including first data from the first scientific instrument. The first data includes a first timestamp corresponding to a first data signal. The discrete-time data also includes second data from the second scientific instrument. The second data includes a second timestamp corresponding to a second digital signal. The database is further configured to store the first data and first metadata in a first record. The first record includes a first intensity field having a first data type and a first descriptor field having a second data type. The database is further configured to store the second data and second metadata in a second record. The second record includes a second intensity field having a first data type and a second descriptor field having a second data type. The first metadata includes a first identifier and the second metadata includes a second identifier. When the first data includes a first intensity and a first descriptor of a first digital signal, the database is configured to store the first data by storing the first intensity in the first intensity field and the first descriptor in the first descriptor field. When the second data includes the second intensity and the second descriptor of the second digital signal, the database is configured to store the second data by storing the second intensity in the second intensity field and the second descriptor in the second descriptor field.

[0091] The aspects discussed above can provide data structures that reflect the characteristics of biological process data. In particular, data determined from different scientific instruments used to monitor processes can include intensity and descriptors. By providing a flexible data structure for storing intensity and descriptors from various scientific instruments, data from diverse sources can be stored in the same data structure. In other words, when the first record can store an image from a microscope and the second record can store an emission spectrum from a spectrometer (and vice versa), data from both devices can be stored more efficiently (e.g., requiring less space to store it) and more easily queried (through a single interface). Thus, a field with a first data type can be used to store intensity, and a field with a second data type can be used to store descriptors, regardless of which scientific instrument produced the intensity and descriptor.

[0092] The conditional clauses in the independent claims (e.g., "when...") reflect the fact that data from various scientific instruments can be stored in a data structure.

[0093] The superior efficiency of this storage is particularly evident compared to storage in multiple databases with different data models, or compared to operating historical database systems. Furthermore, raw data can be stored, eliminating the need for prior processing with statistical techniques such as interpolation, thus avoiding the loss or distortion of the original data.

[0094] The subject matter described in this disclosure can be implemented as a method or on an apparatus, and may take the form of one or more computer programs (computer program products). Such computer programs can cause data processing equipment to perform one or more operations described in this disclosure.

[0095] The subject matter described in this disclosure can be implemented in data signals or on a machine-readable medium, wherein the medium is embodied in one or more information carriers, such as CD-ROM, DVD-ROM, semiconductor memory, or hard disk. In particular, the disclosed subject matter can be tangibly embodied in a machine (computer)-readable medium.

[0096] Furthermore, the subject matter described in this disclosure can be implemented as a system including a processor and memory coupled to the processor. The memory can encode one or more programs to cause the processor to execute one or more of the methods described in this application. Moreover, the subject matter described in this disclosure can be implemented using various machines.

[0097] Details of one or more implementations are set forth in the exemplary drawings and the following description. Other features will be apparent from the description, drawings, and claims. Attached Figure Description

[0098] Figure 1The data model of the database records is shown.

[0099] Figure 2 A record is shown based on a data model that stores data received from a Raman spectrometer.

[0100] Figure 3 A record is shown based on a data model that stores data received from a microscope.

[0101] Figure 4 Another record is shown based on a data model that stores data received from a Raman spectrometer.

[0102] Figure 5 The storage is depicted graphically. Figure 4 The data in the records.

[0103] Figure 6 The diagram shows a record storing data received from a nuclear magnetic resonance spectrometer.

[0104] Figure 7 The storage is depicted graphically. Figure 6 The data in the records.

[0105] Figure 8 The record storing 2D data received from the nuclear magnetic resonance spectrometer is shown.

[0106] Figure 9 The storage is depicted graphically. Figure 8 The data in the records.

[0107] Figure 10 A record is shown storing data received from a liquid chromatography-mass spectrometry system.

[0108] Figure 11 The storage is depicted graphically. Figure 10 The data in the records.

[0109] Figure 12 This shows a record storing a subset of data received from a liquid chromatography-mass spectrometry system.

[0110] Figure 13 A visual representation of the intensity and descriptor is shown.

[0111] Figure 14 This illustrates a specific implementation of a data model in a computer system.

[0112] Figure 15 A view of the process control device is shown.

[0113] Figure 16 Another view of the process control device is shown. Detailed Implementation

[0114] The examples will be described in detail below with reference to the accompanying drawings. Various modifications can be made to the examples. In particular, one or more elements of one example can be combined and used in other examples to form new examples.

[0115] Figure 1 A data model based on some specific implementations is illustrated. A data model may include a database model specifying how the database is constructed and used. A data model may organize data records and the data within those records, specifying how they are related to each other. A data model may specify data structures. Data structures may specify the storage format used to collect data values, the relationships among the data values, and the functions or operations that can be applied to the data values. Data values ​​may correspond to first data and second data received from a first scientific instrument and a second scientific instrument, respectively.

[0116] Scientific instruments may include any instruments that can be used in conjunction with a bioreactor for monitoring cell growth during antibody production. The bioreactor may be located in a process development laboratory or used in a regulated production environment (e.g., to produce a specific drug). Scientific instruments may include sensors attached to the bioreactor or another form of process control device. Sensors can be used to determine values ​​for various process parameters, including pH, temperature, optical density, and oxygen partial pressure. Furthermore, scientific instruments may include Raman spectrometers (such as those used in...). Figure 2 (described in the context) and microscopes (as described in) Figure 3 (Described in the context of...)

[0117] Figure 1 The described elements can be included in every record of the database. Specifically, each element can correspond to all or part of a record's fields. Each field can be a placeholder in the data model for data of the corresponding data type.

[0118] The type field 101 may include (i.e., store or contain) a description of the type of the corresponding scientific instrument. This type may be referred to as an observation type and can be mapped to functions used for processing data received from the corresponding scientific instrument. In some cases, multiple observation types may be assigned to the same scientific instrument and can be used to store data received from the corresponding scientific instrument in different ways (described in more detail below). Specific observation types include RAMAN, Proton NMRNOESY microscope, etc. 13 C- 13 C COSY NMR, LC-MS, LC-MS pairs. The type corresponding to the scientific instrument can be specified before receiving data from the corresponding scientific instrument.

[0119] The timestamp field 103 can correspond to a digital signal. The timestamp field 103 can specify when the digital signal was observed. Therefore, the timestamp field 103 can specify when a file corresponding to or including data from the digital signal (e.g., a microscope image) was created. In the case of online spectroscopy (e.g., analysis performed at the location of a process control device), the timestamp can specify when data was acquired from the sample. In the case of offline spectroscopy (e.g., analysis performed in a remote laboratory), the timestamp can specify the time of sampling.

[0120] Data 105 may include multiple components. Specifically, data 105 may include tabular component 107 and unstructured (non-tabular) component 109. Tabular component 107 may include multiple fields. Specifically, tabular component 107 may include a descriptor field 111 for at least a one-dimensional vector and an intensity field 113 for arrays (e.g., multidimensional arrays). Dimensional vectors can be implemented as arrays.

[0121] Based on the depicted example, the data type of descriptor field 111 is an array. Similarly, the data type of intensity field 113 is also an array. More specifically, descriptor field 111 may include a dimension vector of the descriptor for each dimension of the multidimensional array in intensity field 113. Unstructured component 109 may include fields for non-tabular data (such as formatted text or media (e.g., multimedia)).

[0122] The value of type field 101 can be mapped to one or more of the following: the number of dimension vectors in descriptor field 111, the number of dimensions of the multidimensional array in intensity field 113, and the key (attribute) of metadata 115. Metadata attributes can include one or more of the following: acquisition time, sample identifier, operator, annotation, column identifier, instrument, elution method, and item.

[0123] In other words, the functionality associated with each type can be used to construct data records for data of that type. Although the structures may differ, all structures conform to the data model.

[0124] Additionally or alternatively, type field 101 can specify the processing to be performed on the data before it is stored, such as in combination with... Figure 12 Let's discuss this in more detail.

[0125] The data model can also include metadata115. Metadata can be implemented as a set of key-value pairs.

[0126] Advantageously, data received from any scientific instrument used in the production of chemical, pharmaceutical, biopharmaceutical and / or biological products can be stored in a data record according to the depicted data model.

[0127] Furthermore, metadata 115 can include common elements regardless of which scientific instrument the data is received from. For example, when the received data originates from a sample, metadata 115 can include a (unique) sample identifier. When the data does not originate from a sample, metadata 115 can include a corresponding identifier with a similar (or identical) format to the sample identifier. Therefore, the metadata can be used to query the data regardless of which scientific instrument it is received from.

[0128] Unless otherwise stated, the data records described below conform to Figure 1 The data model. Specifically, Figure 2 , Figure 3 , Figure 4 , Figure 6 , Figure 8 , Figure 10 and Figure 12 It shows the conformity Figure 1 Data records of the data model.

[0129] Timestamp field 103 may include the following Figures 2 to 12 Any timestamp discussed in the context of [the following text]. Each timestamp mentioned below can be a first timestamp or a second timestamp. Similarly, the strength mentioned below can be a first strength or a second strength, and the descriptor mentioned below can be a first descriptor or a second descriptor. Therefore, according to [the following text] Figure 1 The data record of the data model can store any data described in the context of the following diagram (e.g., first data or second data).

[0130] Figure 2 A data record 200 storing data (e.g., first data) received from a Raman spectrometer is shown. Data record 200 includes a type field 101. In this case, type field 101 has a Raman type 201 to indicate that the data was received from a Raman spectrometer. Data record 200 also includes a timestamp field 103. In this case, timestamp field 103 includes a timestamp 203 indicating the time when the Raman spectrometer completed data collection. Data 105, including table component 107, corresponds to... Figure 1 The data shown is 105 and fields 111 and 113. In this case, fields 111 and 113 are each implemented using one-dimensional arrays.

[0131] Data from a Raman spectrometer can be received in the form of digital signals. Descriptor field 111 can store descriptors 205 for the digital signals. In the depicted example, a single-dimensional vector storing the wavelengths is present.

[0132] Example output data received from a Raman spectrometer (e.g., provided as a text file) is as follows:

[0133] #Timestamp: October 22, 2018 - 15:23

[0134] #Data collection time: 2 min

[0135] #Sample ID: an_id_001

[0136] #Instrument ID: instrument_02

[0137] # Operator: John Smith

[0138] #Note: 'No comments'

[0139] #Protocol: SOP_x34 400 0.024 400.1 0.032 400.2 0.056

[0143] ... 2000 2.987

[0145] Output data is stored in data record 200. First data from the first scientific instrument or second data from the second scientific instrument may include the output data described above. The output data may correspond to the first data from the first scientific instrument or the second data from the second scientific instrument. Therefore, descriptor field 111 stores the wavelength value "400" and wavelength value "400.1" from the exemplary output data. Due to space limitations, not all values ​​are shown. Intensity field 113 includes the values ​​"0.024" and "0.032" from the exemplary output data. Therefore, 0.024 may correspond to the intensity of wavelength 400. Similarly, 0.032 may correspond to the intensity of wavelength 400.1. Therefore, wavelengths 400 and 400.1 describe intensities 0.024 and 0.032, respectively.

[0146] exist Figure 2 In this example, unstructured component 109 does not include data. Metadata 115 includes acquisition time 209. Acquisition time 209 may reflect the time required for the Raman spectrometer to collect descriptor 205 and intensity 207. Metadata 115 also includes sample identifier 211. Sample identifier 211 is an example of a unique identifier (i.e., the first identifier or the second identifier). Metadata 115 also includes operator 213. In this case, operator 213 has the value "John Smith". Metadata 115 also includes annotation 215. In addition, metadata 115 includes protocol 217.

[0147] Figure 3 Another example is shown of a data record 300 that stores data received from scientific instruments (e.g., second data). Figure 3 In this context, the scientific instrument is the microscope. Images can be acquired periodically using a microscope during the process. In particular, visual images of cells can be used to monitor cell development.

[0148] Data record 300 can correspond to data record 200, the difference being that, compared to data record 200, data record 300 stores different data (values) from different scientific instruments.

[0149] The following are example output data provided by the microscope:

[0150] -Image files

[0151] - File creation date: October 22, 2018, 15:23

[0152] - Sample ID name in the filename: an_id_001.png

[0153] - The project name contained in the directory name of the file: clone234

[0154] - Text files containing comments

[0155] -Image files

[0156] Operator: John Smith

[0157] -Observation results: 'Poorly differentiated cells'

[0158] The output data from the microscope shown above includes image files (“picture files”) and text files containing annotations from the operator (“John Smith”). Figure 3 A data record 300 storing exemplary output is shown. The output data may correspond to first data from a first scientific instrument or second data from a second scientific instrument.

[0159] First data from the first scientific instrument or second data from the second scientific instrument may include the output data shown above.

[0160] exist Figure 3 In the image data, the type field 101 includes the microscope type 301, indicating that the output data was received from the microscope. The timestamp field 103 includes a timestamp 303 indicating when the file storing the microscope image data was created or last modified. Figure 3In the depicted example, the corresponding data received from the microscope does not include the corresponding intensity or the corresponding descriptor. Specifically, the descriptor field 111 and the intensity field 113 are empty. Instead, data 105 includes file byte content. More specifically, unstructured component 109 may include a reference field that may specify the location 305 of the corresponding data (e.g., image data obtained from the microscope).

[0161] Image data can be stored in a clustered file system. Specifically, image data can be stored within the Apache Hadoop framework, which is used for distributed storage and processing. The Apache Hadoop framework can include a clustered file system.

[0162] Similar to Figure 2 Metadata 115 includes sample identifier 307, operator 309, and comment 311. Additionally, metadata 115 includes item identifier 313. In some cases, each sample identifier included in metadata 115 can be unique. Sample identifier 307 differs from sample identifier 211. Although other metadata fields... Figure 2 and Figure 3 They are shown as the same, but these fields can be different.

[0163] Figures 1 to 3 The described data model can be used not only for process data in the context of process development and manufacturing, but also for research or analytical laboratories. In particular, the data model can be used in the context of genomics, transcriptomics, proteomics, metabolomics, lipidomics, and so on.

[0164] Figure 4 A data record 400 storing data from a scientific instrument is shown. Similar to data record 200, data record 400 stores data received from a Raman spectrometer. Unless otherwise stated, data record 400 corresponds to data record 200.

[0165] Data record 400 stores data different from data record 200. Specifically, data record 400 stores descriptor 405 in descriptor field 111 and intensity 407 in intensity field 113. The timestamp 403 in timestamp field 103 is different from values ​​303 and 203. Furthermore, sample identifier 411 is different from sample identifier 211 and sample identifier 307. Metadata elements 409, 413, 415, and 417 can correspond to a combination Figure 2 and Figure 3 Metadata elements under discussion.

[0166] Figure 5The data from the Raman spectrometer, stored in data record 400, is depicted graphically. The x-axis shows the Raman shift, the value of which is stored as descriptor 405 in descriptor field 111. The y-axis shows the relative intensity, which is stored as intensity 407 in intensity field 113.

[0167] Figure 6 A data record 600 storing data received from a scientific instrument is shown. Similar to data records 200, 300, and 400, data record 600 includes... Figure 1 All elements of the depicted data model. However, unlike data records 200, 300, and 400, data record 600 stores data received from a nuclear magnetic resonance (NMR) spectrometer (i.e., data obtained via magnetic resonance spectroscopy). The type field 101, type 601, records the type "Proton NMR NOESY". The timestamp field 103, timestamp 603, records the observation time (e.g., when the sample was collected via a process control device) and differs from values ​​403, 303, and 203.

[0168] The data in data record 600 may have been collected using proton NMR (i.e., hydrogen-1 NMR) utilizing nuclear Overhauser effect spectroscopy (NOESY) to determine the molecular structure of the substance. Descriptor field 111 includes a one-dimensional vector for storing descriptors 605. The intensity field includes a one-dimensional array for storing intensities 607.

[0169] Sample identifier 611 identifies the sample. Other metadata elements 609, 613, and 615 correspond to information about... Figures 2 to 4 Metadata elements under discussion. Metadata elements 617 and 619 can be specific to NMR spectrometers.

[0170] Figure 7 A subset of the data in data record 600 is depicted graphically. For example... Figure 5 As shown, the x-axis represents the descriptor, in Figure 7 In this case, the descriptor is measured in parts per million (ppm). 1 H chemical shift. The y-axis coordinate (not shown) represents the intensity, such as... Figure 5 As shown.

[0171] Figure 8 A data record 800 storing data received from a scientific instrument (i.e., a nuclear magnetic resonance spectrometer) is shown. In this case, the data was generated by two-dimensional correlation spectroscopy (COSY).

[0172] Type field 101 stores type 801, which indicates that the function of using two-dimensional spectral data should be used to store the data in data record 800 (e.g., type 801 can be parsed from the data stream and used to process the rest of the data). As described above, the timestamp 803 of timestamp field 103 reflects the observation time.

[0173] Unlike the data stored in data records 200, 300, 400, and 600 (single-dimensional vectors and one-dimensional arrays), the data stored in data record 800 is stored using two-dimensional vectors and two-dimensional arrays. Therefore, descriptor field 111 stores two-dimensional vector 805. Furthermore, the intensity field stores two-dimensional array 807. In two-dimensional array 807, the two dimensions represent ppm values; additionally, the value of each point in array 807 specifies the intensity.

[0174] Metadata elements 809 to 819 (key-value pairs) correspond to metadata elements 609 to 619.

[0175] Similar to data record 800, 3D NMR data can be identified by the corresponding observation type and stored using a three-dimensional vector in descriptor field 111 and a three-dimensional array in intensity field 113. Furthermore, 4D NMR data can also be identified by the corresponding observation type and stored using a four-dimensional vector in descriptor field 111 and a four-dimensional array in intensity field 113.

[0176] Figure 9 The data in data record 800 is depicted graphically. The x-axis and y-axis show descriptors and values, which reflect the intensity in the chart.

[0177] Figure 10 A data log 1000 is shown storing data received from a scientific instrument (in this case, a liquid chromatography-mass spectrometry (LC-MS) system). LC-MS is an analytical chemistry technique that combines the physical separation capabilities of liquid chromatography with the mass analysis capabilities of mass spectrometry.

[0178] Data record 1000 has a type field 101 with LC-MS type 1001. Type 1001 specifies a particular observation type and describes how the data received from the LC-MS system should be stored. Specifically, descriptors 1005 received from the LC-MS system are stored in a two-dimensional vector in descriptor field 111 (the two-dimensional vector can also be implemented as a two-dimensional array); intensities 1007 received from the LC-MS system are stored in a two-dimensional array (e.g., a matrix) in intensity field 113.

[0179] The timestamp field 103 has a timestamp of 1003.

[0180] Elements 1009 to 1015 and 1019 of metadata 115 can correspond to metadata elements 809 to 815 (the difference lies in the different values ​​of sample identifier 1011 and instrument 1019). Metadata elements 1017 and 1021 can be specific to the LC-MS system.

[0181] Figure 11 A subset of the data in data record 1000 is depicted graphically.

[0182] Figure 12 A data record 1200 storing data received from a scientific instrument is shown. Data record 1200 includes a timestamp field 103 with a timestamp 1203. Similar to... Figure 10 In this case, the scientific instrument is also an LC-MS system. Unless otherwise stated, elements of data record 1200 correspond to elements of data record 1000.

[0183] However, unlike type 1001, type 1201 of type field 101 specifies that data received from the LC-MS system will be stored as time-quality pairs. Therefore, descriptor 1205 includes a first time-quality pair with time "0.0" and quality "50.0103". Descriptor 1205 also includes a second time-quality pair with time "0.0" and quality "51.0234".

[0184] Each of the intensities 1205 corresponds to one of the time-mass pairs. For example, "125" corresponds to the first time-mass pair, and "465" corresponds to the second time-mass pair.

[0185] The advantage of both storage methods in the LC-MS system is flexibility. Specifically, data record 1000 can store all data from the LC-MS system, including redundant data. Therefore, data can be processed and stored quickly, but it requires more space.

[0186] Unlike data record 1000, data record 1200 can store a subset of data from the LC-MS system. Data can be processed before storage to eliminate redundant data. Processing may take more time, but it can also lead to more efficient storage (i.e., requiring less storage space).

[0187] Both storage methods are made possible by including observation types in the data records and mapping the observation types to functions that specify how the data is stored.

[0188] Figure 13A visual representation of data that can be stored in descriptor field 111 and intensity field 113 is shown. For example, descriptor field 111 can store a single word as a descriptor. In the depicted example, the single word is "pH". Therefore, intensity field 113 can store a single descriptor 1301, such as the pH value corresponding to descriptor field 111.

[0189] In another example, the intensity field 113 can store the Raman spectral intensity as an intensity vector 1303, for example, as Figure 2 and Figure 4 As described in the context. In this case, intensity field 113 may have a single dimension. Intensity field 113 may have a corresponding descriptor field 111. Descriptor field 111 may store the wavelength corresponding to any intensity stored in intensity field 113 (as described above).

[0190] The intensity vector 1303 can correspond to 5,000 data points. Each of these data points can be numerical. Therefore, the numerical data points of the Raman spectrum can be stored in the intensity field 113, and the intensity field 113 can be a one-dimensional array with 5,000 entries. The intensity vector 113 can be implemented as a single vector comprising 5,000 wavelengths corresponding to the data points.

[0191] In some cases, when descriptor field 111 stores descriptors (e.g., descriptor 205) and intensity field 113 stores intensity (e.g., intensity 207), unstructured component 109 may refer to additional data. For example, a reference field may store a reference to a manufacturer's specification for a cell culture medium. Therefore, descriptor field 111 may include descriptor 205 of the cell culture medium, intensity field 113 may include the spectral intensity of the cell culture medium, and unstructured component 109 may include a reference to a location storing a manufacturer's specification describing the cell culture medium. The manufacturer's specification may be provided as a formatted text document, such as a Portable Document Format file. The manufacturer's specification may be stored in a clustered file system. Therefore, the location mentioned in the reference field may refer to a clustered file system. For example, the referenced system may include a URI referencing the clustered file system.

[0192] In yet another example, intensity field 113 may store media (e.g., an image) data as a matrix 1305 or a cube (not shown) in an uncompressed format. In this case, the image can be represented as a cube of intensity and stored in intensity field 113. Descriptor field 111 can indicate that media information is stored in intensity field 113.

[0193] For a cube, a 300×200 pixel color image can be represented by numerical values ​​(i.e., a 3×300×200 multidimensional array). Therefore, the dimensions of the array can be divided into different color components, such that one dimension describes red, the second describes green, and the third describes blue. Other ways of constructing the dimensions of the array are also possible.

[0194] The corresponding descriptor field 111 may include a single word indicating that the uncompressed image data is stored in the intensity field 113. Storing uncompressed image data can be advantageous in some cases. For example, it is possible to query uncompressed data of image regions with similar values. Such queries may not be possible (or may be more difficult) using compressed image data.

[0195] In some cases, images or videos can be stored as compressed files. In such cases, the location of the compressed file can be stored in the reference field of unstructured component 109. Specifically, location 305 can be a Uniform Resource Identifier and can be stored in the reference field.

[0196] Figure 14 A system that can be used to implement aspects of this disclosure is shown.

[0197] Figure 14 The depicted system includes a data storage device 1401, a data management system 1403, and a scientific instrument 1405. Although the data management system 1403 is shown as a relational database management system (RDBMS) 1407, a NoSQL database 1409, and a file system 1411, the data management system may also use a single database or be implemented in a file system 1411 that does not include one of the databases 1407 or 1409. The file system 1411 may be a clustered file system implemented within a distributed storage and processing framework such as Apache Hadoop, as described above. The scientific instrument 1405 may include a connector 1413 for facilitating communication with the data management system 1403. The connector 1413 may be implemented using a first interface and a second interface. The data provided by the scientific instrument 1405 through the connector 1413 may be in the form of raw data units 1415 (i.e., observations). The observations may be processed and / or parsed to obtain the data described above. Figure 1 The data structure defined in the context of the data model.

[0198] Data management system 1403 can exist prior to the implementation of data structures defined by the data model. Data can be imported from data management system 1403 into a database that supports data structures defined by the data model. In some cases, database views of RDBMS 1407 can be used to import data into a database that supports the data structures. Database views can simplify the import process or ensure that unnecessary data is not imported.

[0199] Parsing data received from connector 1413 and storing it in a table according to a predefined schema (e.g., a set of predefined attributes or columns) performed as in conventional methods may result in data loss. In particular, it may be necessary to transform the data to fit into an RDBMS table or conform to the defined labels of an operating historical database system.

[0200] The drawback of storing data according to a predefined pattern is that it locks the digital definition of a continuous (i.e., analog) signal to that predefined pattern. Therefore, any improvement in signal resolution or a shift in the calibration of scientific instruments will often result in a different set of properties. In such cases, significant changes or reconfigurations of the database may be necessary.

[0201] Compared to RDBMS tables or operational history database systems, storing data in a format that supports… Figure 1 Before the data model described is stored in the database, it may not be necessary to truncate or interpolate the observation data. In other words, the observation results do not need to be adjusted according to a predefined pattern for storage. Therefore, the observation results received from connector 1413 do not need to be adjusted before being stored in the database as raw data units 1415.

[0202] Storing observations as raw data units 1415 offers the advantages of using a read-time mode approach and makes it possible to store observations without data loss. In other words, storing data as raw data units 1415 makes it possible to retain the raw data from the scientific instrument 1405 and perform additional processing based on that raw data. The raw data can be provided in the form of digital signals.

[0203] File system 1411 can reference fields (in combination) Figure 3 (Refer to) When stored in file system 1411, observations can be saved in formats such as JSON, BSON, or XML. They can also be combined with... Figure 1 Other formats corresponding to the data model discussed.

[0204] If you are saving observations to an RDBMS 1407, you may need a data schema to account for the diversity of the data, such as the diversity of spectral data and the possibility that the intensities of different groups may not map to the same wavelength. For example, when using MySQL as an RDBMS 1407, you can use MySQL's JSON data type.

[0205] Once the data received from instrument 1405 via connector 1413 is stored in data management system 1403, a dataset can be created. The dataset can be created with or without preprocessing steps. Exemplary preprocessing steps include curve smoothing, image feature extraction, and other data manipulation techniques.

[0206] In some implementations, sensor measurements (i.e., measurements where individual values ​​are associated with timestamps) can be stored in an operational history database system such as an OSI PI historical database. Spectral data can be recorded in a database supporting the data model (e.g., implemented as a MongoDB server). Spectral data can be recorded along with corresponding metadata and timestamps. Microscope images can be saved to a file system (possibly implemented as part of the Apache Hadoop distributed storage and processing framework). Images from the microscope can be saved along with corresponding metadata and timestamps.

[0207] Metadata (e.g., metadata for spectral or image data) may include process-related information such as cell lines, growth media, research, projects, and operators. Media (e.g., images or videos) stored on file system 1411 may be compressed. As discussed above, uncompressed images may be stored in a database that supports the data model.

[0208] After a large amount of process data has been collected, and after running multiple processes using different cell lines and different culture media, the stored data (including the different types of data discussed above) can be queried. A single user interface can be used to access all the data stored in a database that supports the data model. Queries to the data management system 1403 can be handled through the user interface. The user interface can be implemented using schema and data manipulation tools such as object-relational mappers (e.g., Django ORM) or SQL query engines (e.g., Apache drill).

[0209] The specific implementations discussed above can offer various advantages. In particular, images taken at any time during any process operation can be easily matched to a specific experimental or production environment. Furthermore, metadata can be used to match observations received from microscopes or other scientific instruments with other observations received from spectrometers or various sensors. Additionally, all spectral data from a range of pH values ​​derived from sensors can be collected, processed to estimate glucose concentration, and plotted as a function of glucose concentration and pH. Such queries can be facilitated through a shared user interface and records of timestamps associated with the spectral data and data collected from individual sensors.

[0210] Furthermore, a shared metadata format (e.g., using identifiers to match all data belonging to the same process) facilitates querying. Additionally, updates to how spectral data is processed can improve analysis months after the initial analysis of the spectral data, without requiring changes to the data contained in the database supporting the data model. This is the benefit of storing data unchanged in a data management system (e.g., without truncating or interpolating the data). When attempting to store spectral data in a PI system or a conventional relational database management system, it may be impossible to store unchanged data.

[0211] In the example depicted, the data is stored in three different systems (i.e., RDBMS 1407, NoSQL database 1409, and file system 1411). However, data can be imported from a single database.

[0212] The data management system 1403 can be part of an existing infrastructure. Therefore, the desire to use the existing infrastructure may be a reason to continue using the data management system 1403.

[0213] Figure 15 A process control device 10 (possibly implemented as a bioreactor system) comprising a series of containers (e.g., first-scale or microscale bioreactors) is shown. The containers are configured to contain fluids (other types of fluids are also possible) for producing biopharmaceutical products. The containers may be located in a container station 11 (also called a receiving station). The container station 11 is configured to receive a specified number of containers, such as 6, 12, or 24. The process control device 10 is operable to control and / or monitor the containers in the container station 11 at least partially in parallel, and possibly completely in parallel. The process control device 10 is operable to determine or set process parameters to be controlled (i.e., process control parameters). Examples of process parameters to be controlled include stirring speed and gas supply rate.

[0214] Process control device 10 is operable to periodically determine process parameter values ​​for a process parameter (e.g., the process parameter to be measured). Process parameter values ​​can be determined directly from the container (e.g., via a sensor point) or from a sample obtained from the container. More specifically, analysis module 12 can be used to process fluid (e.g., a sample) from the container to determine process parameter values. Thus, analysis module 12 can deliver fluid from the container along a specific route to scientific instruments (e.g., analytical instruments) to determine values ​​for process parameters (such as pH, cell count, metabolite levels, nutrient levels). The pH value determined by the analysis module can be used for sensor calibration. Analysis module 12 can also support sample preparation and cleaning and rinsing after sample collection.

[0215] The process control device 10 includes a robot, which may be implemented as a liquid processor 13. The robot is capable of handling each first-scale container and dispensing and extracting fluid from the containers. The liquid processor 13 performs automated process control and sampling. The liquid processor 13 collects (or extracts) samples from each individual container in the container station 11 and supplies nutrients or cleaning agents (e.g., acids, alkalis, defoamers, etc.) to each individual container. In a specific implementation, these tasks may also be performed by a robot instead of the liquid processor 13.

[0216] The process control device 10 may include a process control module 14 (also referred to as a workstation). The process control module 14 includes a user interface (e.g., an input device such as a keyboard, an output device such as a display), processing devices, and storage devices. The process control module may store process control strategies to control the process control device 10, and more specifically, to control the liquid processor 13 and the analysis module 12. In particular, the process control device may store values ​​of process parameters to be controlled (i.e., control setpoints). Furthermore, the process control device may store process recipes.

[0217] The process control device 10 may include a sampling device 15. More specifically, the liquid processor 13 may include a sampling device 15. The sampling device 15 may enable an automated pipetting system and / or carry pipette tips.

[0218] The process control device 10 may include a liquid 16 to supply to the analysis module 12. The liquid 16 may include cleaning and rinsing agents, pH buffer solutions, calibration solutions, etc.

[0219] The analysis module 12 and the process control module 14 can be integrated into the controller.

[0220] Storage container 17 can be used to store liquids to be supplied to the container. Liquids from storage container 17 can be supplied by process control device 10, particularly liquid processor 13. Liquids may include glucose feed, acids, alkalis, defoamer solutions, etc.

[0221] The process control device 10 may include a sample holder or container, possibly implemented as a sample cup 18. More specifically, the sample cup 18 may be part of the analysis module 12. The sample cup 18 may be configured to receive samples obtained by the liquid processor 13 and / or the sampling device 15, and to supply the samples to the analysis module 12 and further analytical devices.

[0222] Process control device 10 may include scientific instruments, possibly in the form of analytical device 20. Analytical device 20 may be implemented as a Raman measurement system (i.e., a spectrometer), a high-performance liquid chromatography (HPLC) device, or a mass spectrometry device. Multiple analytical devices (not shown) may be present. Analytical device 20 may be configured to receive samples from analytical module 12 and perform analytical measurements to determine process parameter values ​​or process outputs. Process outputs may include product quality attributes, such as glycosylation.

[0223] One or more heaters or coolers (not shown) may be located adjacent to container station 11 to control the temperature of the container.

[0224] Figure 16 Shown from the top angle Figure 15 The process control device.

Claims

1. A computer-implemented method for storing heterogeneous sequences of discrete-time data determined from processes for producing chemical, pharmaceutical, biopharmaceutical, and / or biological products, the method comprising: Receive the discrete-time data, which includes: First data from a first scientific instrument, the first data including a first timestamp (203) corresponding to a first digital signal; and Second data from a second scientific instrument, the second data including a second timestamp (303) corresponding to a second digital signal; The first scientific instrument is different from the second scientific instrument; The first scientific instrument and the second scientific instrument each have different corresponding observation types. The first data has a first observation type (201), and the second data has a second observation type (301). Each observation type is associated with a function for extracting, parsing, and interpreting the data of that observation type. The first data and the first metadata are stored in a first record (200) of the database, wherein the first data is stored according to the function of the stored data mapped by the first observation type (201), and the first record (200) includes: The first strength field has a first data type, and A first descriptor field having a second data type; The second data and second metadata are stored in a second record (300) of the database, wherein the second data is stored according to the function of stored data mapped by the second observation type (301), and the second record (300) includes: A second strength field having the first data type, and A second descriptor field having the second data type; The first metadata includes a first identifier (211) and the second metadata includes a second identifier (307). When the first data includes the first intensity (207) and the first descriptor (205) of the first digital signal, storing the first data further includes: Store the first intensity (207) in the first intensity field, and Store the first descriptor (205) in the first descriptor field; When the second data includes the second intensity and the second descriptor of the second digital signal, storing the second data further includes: Store the second intensity in the second intensity field, and Store the second descriptor in the second descriptor field.

2. The method according to claim 1, wherein, The first and / or second scientific instruments in the scientific instruments include one of the following: a spectrometer, a flow cytometer, a physiological data collection device, a biological sequence data collection device, and a microscope.

3. The method according to claim 2, wherein, When the scientific instrument in the scientific instrumentation includes the spectrometer, the corresponding data of the corresponding scientific instrument includes spectral data, which includes wavelength and spectral determination values ​​corresponding to the wavelength. The corresponding descriptor is the wavelength, and the corresponding intensity is the spectral value corresponding to the wavelength; and When the scientific instrument in the scientific instrument includes the biological sequence data collection device, the corresponding data includes gene sequence data, wherein the corresponding descriptor is a gene identifier, and the corresponding intensity reflects protein expression.

4. The method according to any one of the preceding claims, The first data consists of all data provided by the first scientific instrument corresponding to the first timestamp (203); and / or The second data consists of all the data provided by the second scientific instrument corresponding to the second timestamp (303); Wherein the first data represents the first digital signal at the first timestamp (203); The second data represents the second digital signal at the second timestamp (303).

5. The method according to claim 1, Each of the records mentioned therein includes a reference field (109); When the corresponding data from one of the scientific instruments does not include the corresponding intensity or the corresponding descriptor, the method further includes: The corresponding data is stored in the cluster file system; The position (305) of the corresponding data is stored in the reference field (109); The format of the corresponding data has one or more of the following characteristics: the format is not human-readable, the format is binary, the format is an encoded content format, and the format is a compressed format; When the corresponding data from one of the scientific instruments includes the corresponding intensity and corresponding descriptor, as well as additional data, the method further includes: The additional data is stored in the cluster file system; and The location of the additional data is stored in the reference field (109).

6. The method according to claim 1, When the corresponding data from one of the scientific instruments does not include the corresponding intensity or the corresponding descriptor: The corresponding data includes formatted text documents, charts, images, or videos. When the corresponding data includes the image, the image is from a microscope.

7. The method according to claim 1, further comprising: Provide a user interface to query the first data, the second data, the first metadata, and the second metadata; The user interface is implemented using pattern and data manipulation tools.

8. The method of claim 1, wherein when the corresponding data from one of the scientific instruments includes a corresponding intensity and a corresponding descriptor, the corresponding data is in a human-readable format.

9. The method according to claim 1, wherein, The discrete-time data also includes third data, which includes a single value determined by a process control device for controlling the process, the single value having a corresponding timestamp.

10. The method of claim 9, further comprising: Determine the descriptor of the third data; Storing the third data and third metadata in a third record, the third record including a third strength field having the first data type and a third descriptor field having the second data type, and storing the third data and third metadata in the third record includes: Store the single value in the third strength field; and The descriptor is stored in the third descriptor field; The third metadata includes a third identifier.

11. The method according to claim 9, wherein, The third data is scale-independent and / or sampling-independent.

12. The method according to claim 1, wherein, The first data type and / or the second data type include at least one of the following: composite data type and linear data structure.

13. The method according to claim 1, wherein, The first data type and / or the second data type include arrays, wherein each of the data types specifies the same number of dimensions.

14. A computer-readable medium comprising computer-readable instructions that, when loaded onto and executed on a computer system, cause the computer system to perform operations according to any one of the preceding claims.

15. A system for storing heterogeneous sequences of discrete-time data determined from processes for producing chemical, pharmaceutical, biopharmaceutical, and / or biological products, the system comprising: To the first interface of the first scientific instrument; To the second interface of the second scientific instrument; The database is configured as follows: The discrete-time data is received through the first interface and the second interface, and the discrete-time data includes: First data from the first scientific instrument, the first data including a first timestamp (203) corresponding to a first digital signal; and Second data from the second scientific instrument, the second data including a second timestamp (303) corresponding to the second digital signal; The first scientific instrument and the second scientific instrument each have different corresponding observation types. The first data has a first observation type (201), and the second data has a second observation type (301). Each observation type is associated with a function for extracting, parsing, and interpreting the data of that observation type. The first data and the first metadata are stored in a first record (200), wherein the first data is stored according to the function of the stored data mapped by the first observation type (201), and the first record (200) includes: The first strength field has a first data type, and A first descriptor field having a second data type; The second data and second metadata are stored in a second record (300), wherein the second data is stored according to the function of the stored data mapped by the second observation type (301), and the second record (300) includes: A second strength field having the first data type, and A second descriptor field having the second data type; The first metadata includes a first identifier (211) and the second metadata includes a second identifier (307). When the first data includes a first intensity (207) and a first descriptor (205) of the first digital signal, the database is configured to store the first data in the following manner: Store the first intensity (207) in the first intensity field, and Store the first descriptor (205) in the first descriptor field; When the second data includes the second strength and the second descriptor of the second digital signal, the database is configured to store the second data in the following manner: Store the second intensity in the second intensity field, and Store the second descriptor in the second descriptor field.

Images