Biopharmaceutical laboratory automation data acquisition and processing method and system, and storage medium

Abstract

The present disclosure provides a biopharmaceutical laboratory automation data acquisition and processing method, system and related equipment. The method comprises: acquiring an experimental process signal in a biopharmaceutical batch experiment process to form a batch time sequence, and calculating a derived feature for event detection based on the batch time sequence; based on the batch time sequence and the derived feature, identifying a plurality of biological milestone anchors that meet a preset judgment condition, and generating corresponding anchor point confidence for the plurality of biological milestone anchors; constructing a segmented mapping function from physical time to pseudo time based on the plurality of biological milestone anchors, and converting the batch time sequence into a pseudo time sequence using the segmented mapping function; calling a preset reference trajectory, and performing dynamic time warping alignment on the pseudo time sequence and the reference trajectory in the pseudo time domain with the plurality of biological milestone anchors and their anchor point related information as alignment constraints to output the aligned reparameterization data.

Description

Technical Field

[0001] This disclosure relates to the field of financial big data technology, specifically to an automated data acquisition and processing method for biopharmaceutical laboratories, an automated data acquisition and processing system for biopharmaceutical laboratories, electronic equipment, and storage media. Background Technology

[0002] Batch experiments in biopharmaceutical laboratories typically require automated processing of experimental process signals collected during culture, and based on this, batch-to-batch process comparison and consistency analysis. Existing automated processing methods often organize batch time series using physical clock time, that is, directly mapping sampling points from different batches at the same culture duration and then comparing curves.

[0003] However, due to factors such as inoculation status, cell viability, subtle differences in culture medium, and process disturbances, the metabolic progress of different batches often varies, and the process stages corresponding to the same culture duration may not be consistent. In this case, directly corresponding sampling points to physical clock time can easily lead to mismatches of key stage fragments between batches, resulting in unstable differences in batch comparisons based on this correspondence, making it difficult to establish a consistent basis for stage correspondence.

[0004] Furthermore, existing batch alignment methods often employ general time warping or sequence matching to find correspondences between curves from different batches. When there is a lack of explicit constraints on the location of key stages, the alignment path is easily affected by local fluctuations, noise disturbances, and similar curve segments, resulting in multiple solutions or non-unique matching results. This leads to changes in the correspondence between the same batch and the reference trajectory under different operating conditions or parameter settings, resulting in insufficient repeatability and consistency of the alignment results, and affecting the stability of subsequent comparative analysis conclusions.

[0005] It should be noted that the statements in the background section above are only for providing background information related to this disclosure and do not necessarily constitute prior art. Summary of the Invention

[0006] To at least partially overcome the problems existing in the related technologies, embodiments of this disclosure provide a method for automated data acquisition and processing in a biopharmaceutical laboratory, a system for automated data acquisition and processing in a biopharmaceutical laboratory, an electronic device, and a storage medium.

[0007] According to one aspect of this disclosure, an automated data acquisition and processing method for a biopharmaceutical laboratory is provided, comprising: The experimental process signals during the batch experiment of biopharmaceuticals are collected to form a batch time series, and derived features for event detection are calculated based on the batch time series. Based on the batch time series and the derived features, multiple biological milestone anchors that meet preset judgment conditions are identified, and corresponding anchor confidence scores are generated for the multiple biological milestone anchors. Based on the multiple biological milestone anchors, a piecewise mapping function from physical time to pseudo time is constructed, and the piecewise mapping function is used to convert the batch time series into a pseudo time series. A preset reference trajectory is retrieved, and in the pseudo-time domain, the multiple biological milestone anchors and their related information are used as alignment constraints to perform dynamic time warping alignment between the pseudo-time series and the reference trajectory, so as to output the aligned reparameterized data.

[0008] In one exemplary embodiment of this disclosure, the calculation of derived features for event detection includes: Denoising filtering is performed on the batch time series to obtain the filtered sequence; Based on the filtered sequence, calculate the first-order difference sequence and / or the second-order difference sequence to obtain the difference features; The derived features are formed by calculating the sliding window statistics based on the filtered sequence and the differential features.

[0009] In one exemplary embodiment of this disclosure, the identification of multiple biological milestone anchors that satisfy preset judgment conditions includes: Local extreme points are extracted from the batch time series to obtain the feature points to be tested; At the feature points to be measured, the slope reversal features of the pH channel and the metabolic rate change features of the lactate channel are extracted, and the intersection position within the same time window is determined based on the slope reversal features and the metabolic rate change features to obtain the metabolic shift features. The metabolic shift characteristics are matched with the dissolved oxygen rebound characteristics of the dissolved oxygen channel within a time window to update the anchor candidate points, and the multiple biological milestone anchor points are determined based on the positions of the updated anchor candidate points in the batch time series. The anchor confidence level corresponding to each of the multiple biological milestone anchor points is calculated based on the verification results of the time window matching.

[0010] In one exemplary embodiment of this disclosure, constructing the piecewise mapping function from physical time to pseudo-time includes: Each of the multiple biological milestone anchor points is assigned a preset pseudo-time node; Based on the anchor point positions of the multiple biological milestone anchor points on the physical time axis and the pseudo-time nodes, the mapping segments are divided. For each of the aforementioned mapping segments, a monotonically continuous piecewise mapping parameter is fitted, and the piecewise mapping parameters are summarized to generate the piecewise mapping function.

[0011] In one exemplary embodiment of this disclosure, performing dynamic time warping alignment includes: An alignment constraint set is generated based on the multiple biological milestone anchors and their related information. Calculate the alignment cost matrix based on the aforementioned alignment constraint set; Under the constraints of the alignment constraint set, the minimum cumulative cost path is searched based on the alignment cost matrix to obtain the alignment mapping; The pseudo-time series is reparameterized based on the alignment mapping to output aligned and reparameterized data.

[0012] In one exemplary embodiment of this disclosure, the calculation of the alignment cost matrix includes: The signal channel weights are determined based on the anchor point confidence level, and a weighted parameter set is generated. Calculate the weighted distance between the pseudo-time series and the reference trajectory at each pseudo-time sampling point based on the weighted parameter set; The weighted distance is filled into the alignment cost matrix for use in the search of the minimum cumulative cost path.

[0013] In one exemplary embodiment of this disclosure, the output aligned reparameterized data includes: Identify missing sampling points in the reparameterized data and determine the pseudo-time coordinates corresponding to the missing sampling points; Retrieve the pseudo-time conditional distribution of historical batches and obtain the distribution parameters at the pseudo-time coordinates; Based on the distribution parameters and combined with the pseudo-temporal neighborhood observations of the missing sampling points, a posterior correction is performed to obtain the compensation value; The uncertainty value corresponding to the compensation value is written as a metadata tag into the reparameterized data.

[0014] According to one aspect of this disclosure, an automated data acquisition and processing system for a biopharmaceutical laboratory is provided, comprising: The signal acquisition module is configured to acquire experimental process signals during the batch experiment of biopharmaceuticals to form a batch time series, and to calculate derived features for event detection based on the batch time series. An anchor point identification module is configured to identify multiple biological milestone anchor points that meet preset judgment conditions based on the batch time series and the derived features, and generate corresponding anchor point confidence scores for the multiple biological milestone anchor points. The sequence conversion module is configured to construct a segmented mapping function from physical time to pseudo time based on the multiple biological milestone anchors, and to use the segmented mapping function to convert the batch time series into a pseudo time series. The normalization and alignment module is configured to retrieve a preset reference trajectory and perform dynamic time normalization and alignment on the pseudo-time series and the reference trajectory in the pseudo-time domain using the multiple biological milestone anchors and their related information as alignment constraints, so as to output the aligned reparameterized data.

[0015] According to one aspect of this disclosure, an electronic device is provided, comprising: Processor; and Memory for storing the executable instructions of the processor; The processor is configured to execute any of the methods described above by executing the executable instructions.

[0016] According to one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the preceding claims.

[0017] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: In the embodiments of this disclosure, the batch time series processing no longer relies solely on the correspondence of physical clock time at the same scale as the sole basis for comparison. Instead, it first determines a set of event positions with stage indication significance based on the morphological changes and derived characteristics of the batch time series, and establishes a segmented correspondence between physical time and pseudo-time on this basis. This allows different stage segments within the same batch to have a recalcible positioning method in the pseudo-time domain. Thus, the correspondence between different batches is transformed from the same cultivation duration into a segmented correspondence in the pseudo-time domain, making it less likely that the stage segments on which cross-batch comparisons depend will be misaligned due to differences in batch progress speed, thereby making the correspondence basis of key stage segments between batches more consistent. Furthermore, a reference trajectory is introduced in the pseudo-time domain, and the event positions and their related information are used to constrain the dynamic time warping alignment process. This limits the search range and correspondence of the alignment path, reducing the occurrence of multiple or non-unique matches between morphologically similar segments during the alignment process, thus making the alignment mapping of the same batch relative to the reference trajectory less prone to drift under different operating conditions. The reparameterized data output by this alignment mapping is more stable in terms of the correspondence between stage boundaries, thus providing a more reliable data foundation for the consistency of conclusions in subsequent batch comparisons based on the reparameterized data.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this disclosure, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] Figure 1 A system architecture diagram is shown for an automated data acquisition and processing method for biopharmaceutical laboratories that can be applied to embodiments of this disclosure.

[0021] Figure 2 A flowchart illustrating an automated data acquisition and processing method for a biopharmaceutical laboratory, as described in an embodiment of this disclosure, is shown.

[0022] Figure 3 A flowchart illustrating the sub-steps of an automated data acquisition and processing method for a biopharmaceutical laboratory, as described in an embodiment of this disclosure, is shown.

[0023] Figure 4 A flowchart illustrating the sub-steps of an automated data acquisition and processing method for a biopharmaceutical laboratory, as described in an embodiment of this disclosure, is shown.

[0024] Figure 5 A flowchart illustrating the sub-steps of an automated data acquisition and processing method for a biopharmaceutical laboratory, as described in an embodiment of this disclosure, is shown.

[0025] Figure 6 A flowchart illustrating the sub-steps of an automated data acquisition and processing method for a biopharmaceutical laboratory, as described in an embodiment of this disclosure, is shown.

[0026] Figure 7 A schematic diagram of an automated data acquisition and processing system for a biopharmaceutical laboratory is shown in an embodiment of this disclosure.

[0027] Figure 8 A schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present disclosure is shown.

[0028] Figure 9 A schematic diagram of a computer-readable storage medium according to an embodiment of the present disclosure is shown.

[0029] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts. Detailed Implementation

[0030] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0031] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0032] Figure 1 A schematic diagram of a system architecture for an automated data acquisition and processing method for a biopharmaceutical laboratory, which can be applied to embodiments of this disclosure, is shown.

[0033] like Figure 1 As shown, system architecture 100 may include one or more terminal devices such as desktop computers 101 or portable computers, a biopharmaceutical laboratory 102, a network 104, and a server 105. Network 104 serves as the medium for providing a communication link between the terminal devices and server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0034] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of terminal devices, networks, and servers. For example, server 105 could be a server cluster composed of multiple servers.

[0035] The automated data acquisition and processing method for biopharmaceutical laboratories provided in this embodiment can be executed by a terminal device, and correspondingly, the automated data acquisition and processing system for biopharmaceutical laboratories can be installed in the terminal device. However, it is readily understood by those skilled in the art that the automated data acquisition and processing method for biopharmaceutical laboratories provided in this embodiment can also be executed by a server 105, and correspondingly, the automated data acquisition and processing system for biopharmaceutical laboratories can also be installed in the server 105. This exemplary embodiment does not impose any special limitations on this.

[0036] This disclosure first provides a method for automated data acquisition and processing in a biopharmaceutical laboratory. The method is then described below using server execution as an example.

[0037] Figure 2 A flowchart illustrating an embodiment of an automated data acquisition and processing method for a biopharmaceutical laboratory is shown schematically. (Reference) Figure 2 As shown, the method may include steps S210 to S240: Step S210: Collect experimental process signals during the batch experiment of biopharmaceutical to form a batch time series, and calculate derived features for event detection based on the batch time series; Step S220: Based on the batch time series and the derived features, identify multiple biological milestone anchors that meet preset judgment conditions, and generate corresponding anchor confidence scores for the multiple biological milestone anchors; Step S230: Construct a piecewise mapping function from physical time to pseudo time based on the multiple biological milestone anchors, and use the piecewise mapping function to convert the batch time series into a pseudo time series; Step S240: Retrieve the preset reference trajectory, and in the pseudo-time domain, use the multiple biological milestone anchors and their related information as alignment constraints to perform dynamic time warping alignment on the pseudo-time series and the reference trajectory, so as to output the aligned reparameterized data.

[0038] The following will provide a detailed description of the automated data acquisition and processing method for biopharmaceutical laboratories in this example embodiment.

[0039] In step S210, experimental process signals during the biopharmaceutical batch experiment are collected to form a batch time series, and derived features for event detection are calculated based on the batch time series. In this example implementation, the experimental process signals are preferably considered as a set of raw characterizing quantities of the experimental process state of the same biopharmaceutical batch, acquired by collecting data from the reactor body and its surrounding pipelines / exhaust gas pipelines. The experimental process signals can be derived from two sources: one is measurement channel signals, generated by deployed sensors and characterizing the physicochemical state within the reactor; the other is operational channel signals, directly provided by the control system or fed back by the actuators, used to characterize changes in process operating conditions. Measurement channel signals may include, for example, pH electrode output, dissolved oxygen sensor output, temperature sensor output, and exhaust gas O2 / CO2 analyzer output; operational channel signals may include, for example, stirring speed, aeration flow rate, feed pump drive frequency, or opening status quantities. To ensure consistent data semantics when forming batch time series, it is preferable to solidify the channel identifier, dimensions / units, and acquisition source for each channel, and bind a timestamp and data quality identifier to each channel sampling record during acquisition, wherein the data quality identifier is used to distinguish at least three states: valid values, invalid values, and missing values.

[0040] In the specific data acquisition chain, for analog outputs generated by sensors, it is preferable to first perform isolation, amplification, and anti-aliasing processing through a signal conditioning circuit before entering the analog-to-digital conversion process. Signal conditioning may include range matching and common-mode rejection for 4–20mA current loops or 0–5V voltage signals, and configuring an analog low-pass filter before entering the analog-to-digital converter to suppress high-frequency interference; the digital quantity obtained after analog-to-digital conversion is written into the acquisition record as the measurement value of the channel at the corresponding timestamp. For digital / bus outputs provided by the control system or actuator, it is preferable to read and form acquisition records through fieldbus or industrial communication protocols, such as periodically reading the operating parameter register or status frame in the controller via CAN, Modbus, etc., and writing the reading time as the timestamp into the acquisition record. For offline sampling quantities (such as substrate / product concentration, cell density VCD / OD, etc.) that cannot be continuously obtained online but still belong to the experimental process signals, offline detection can be triggered by an automated sampling device, and the detection completion time or sampling time can be written into the acquisition record when the detection result is returned. When the sampling time is used as the timestamp, it is preferable to record the detection delay at the same time to make the temporal semantics of the same channel consistent.

[0041] When calculating derived features for event detection based on the batch time series, the derived features are limited to a set of features that can be obtained from the batch time series through deterministic operations, and the derived features are associated with the batch time series at time indices, so that the derived features at any time index can be traced back to the batch time series values ​​in the neighborhood of that time index.

[0042] refer to Figure 3As shown, in this example embodiment, derived features for event detection can be calculated through the following steps S301 to S303. Wherein: Step S301: Perform denoising filtering on the batch time series to obtain a filtered sequence.

[0043] In this example implementation, the filtering object is preferably a scale-normalized channel sequence, and the filtered output is a filtered sequence that corresponds one-to-one with the input channel and is associated with the original time index. The denoising filter preferably uses the suppression of measurement noise, communication jitter, and instantaneous spikes as boundary conditions, while preserving the slope changes and inflection point shapes of the process signal near the stage boundaries. In implementation, a discrete low-pass filter can be used, with the current filtered value obtained by recursively updating the current input and the previous output. The filter coefficients can be determined by a set time constant and sampling interval. When the sampling interval is not constant, the difference between adjacent timestamps can be used as the coefficient update input to maintain a stable equivalent smoothing strength. In some exemplary embodiments, Kalman filtering or Savitzky-Golay smoothing can also be used; this exemplary embodiment does not impose any special limitations on these methods.

[0044] Step S302: Calculate the first-order difference sequence and / or the second-order difference sequence based on the filtered sequence to obtain the difference features.

[0045] In this example implementation, the first-order difference sequence is used to characterize the local rate of change of the filtered sequence, and the second-order difference sequence is used to characterize the trend of the rate of change. Preferably, the timestamp difference is explicitly used in the difference calculation to avoid distortion of the difference amplitude due to changes in the sampling interval: when the adjacent sampling interval is Δt, the first-order difference can be obtained by dividing the difference between adjacent filtered values ​​by Δt; the second-order difference can be obtained by subtracting the difference values ​​at adjacent time points based on the first-order difference and normalizing by the time interval. To suppress the amplification of residual noise by the difference, it is preferable to use a central difference or multi-point difference form instead of simple forward subtraction: the central difference forms a symmetrical estimate at the current time by using the preceding and following filtered values ​​and their timestamps; the multi-point difference can use a fixed-coefficient convolution kernel to convolve the filtered sequence to obtain the derivative estimate, where the convolution kernel coefficients are predetermined and fixed by the selected difference formula.

[0046] Step S303: Calculate the sliding window statistics based on the filtered sequence and the differential features to form the derived features.

[0047] In this example implementation, the sliding window statistics are used to transform the instantaneous fluctuations of a single point into a stable representation over a local time period. The sliding window can be defined by time length or by the number of sample points; when the channel sampling interval is not constant, it is preferable to define the window by time length, using the current timestamp as the alignment point and tracing back a preset time to form the window, with the valid samples within this window used as statistical input. The sliding window statistics preferably include at least the level and fluctuation statistics of the filtered sequence and the dynamic statistics of the difference features: for the filtered sequence, the window mean, window variance or standard deviation, and window range or quantile difference can be calculated; for first-order differences, the mean, absolute mean, and maximum absolute value within the window can be calculated; for second-order differences, the mean and extreme values ​​within the window can be calculated. The calculation of the above statistics preferably selects valid samples based on data quality indicators, and when the number of valid samples is lower than a threshold, the window statistics are marked as invalid or the previous window result is maintained according to a preset rule. The threshold and maintenance rule are fixed as part of the derived feature definition. In addition, to reduce real-time computational overhead, window statistics can be implemented using an incremental update method, that is, maintaining the cumulative amount within the window and updating the samples entering / leaving as the window slides, thereby obtaining the window mean and variance with a defined computational complexity; for quantile-class statistics, fixed binning histograms or approximate quantile summaries can be used, and the binning strategy or summary parameters can be fixed to limit the approximate error range.

[0048] Finally, the derived features are formed by combining the difference features and the sliding window statistics under the same time index. The combination method is preferably to concatenate them in a fixed order according to the channel and the feature type to obtain a derived feature sequence that corresponds one-to-one with the time index. When there is a multi-channel input, the channel concatenation order and the feature concatenation order are fixed as implementation configuration, so that different batches can obtain derived feature outputs with consistent dimensions and semantics under the same implementation configuration.

[0049] In step S220, based on the batch time series and the derived features, multiple biological milestone anchors that meet preset judgment conditions are identified, and corresponding anchor confidence scores are generated for the multiple biological milestone anchors.

[0050] In this example implementation, the above-mentioned event validity is defined by the discrimination rule that at least one candidate event simultaneously satisfies the preset judgment condition across multiple channels or multiple feature dimensions. The biological milestone anchor is preferably represented by a triplet of anchor type—anchor location—anchor attribute, where the anchor type distinguishes different biological event categories, the anchor location corresponds to a physical time index or timestamp of the batch time series, and the anchor attribute includes at least the anchor confidence generated in this step and anchor-related information used for subsequent constraints. In this step, the anchor-related information preferably covers at least the set of evidence features forming the anchor and the matching window parameters, so as to recalculate and verify the anchor validity. The anchor confidence is used to quantify the reliability of the anchor's validity; its calculation is based on the verification results in this step and is associated with and stored in relation to the specific anchor type and its evidence features, thereby ensuring that anchor determinations under different batches or different operating conditions have comparable measurement standards.

[0051] In some exemplary embodiments, anchor point identification can be completed through a chained process of candidate point generation, multi-channel feature verification, candidate point updating, anchor point confirmation, and confidence calculation. The output of the previous action is directly utilized by the subsequent action, thus ensuring a deterministic dependency of the processing result of this step on the input batch time series and the derived features. To ensure the locatability of anchor point positions on the batch time series, it is preferable to always use the time index of the batch time series as the primary index throughout the entire anchor point identification process, and to reference the feature vector of the derived feature at the same time index through this primary index, avoiding decoupling of features and positions due to inconsistent sampling times across different channels.

[0052] refer to Figure 4 As shown, in this example embodiment, multiple biological milestone anchor points that meet preset judgment conditions can be identified through the following steps S401 to S404. Wherein: Step S401: Extract local extreme points based on the batch time series to obtain the feature points to be tested.

[0053] In this example implementation, the local extrema are used to provide a set of initial candidate positions on the time axis, and the feature point to be measured is the position representation of the local extrema in the batch time series. Local extrema extraction is preferably performed on a sequence where noise is controlled; therefore, in implementation, extrema search can be performed directly based on a filtered sequence consistent with the time index of the batch time series or based on a selected channel sequence of the batch time series. When the derived feature contains second-order difference or curvature-type features, candidate extrema can also be generated at the point of sign change of the second-order difference to improve the coverage of inflection points by the candidate points.

[0054] In practical implementation, a neighborhood-based extreme value determination method can be adopted: for a certain time index i, if its corresponding value is greater than all valid samples in the preceding and following neighborhoods, it is determined to be a local maximum; if its corresponding value is less than all valid samples in the preceding and following neighborhoods, it is determined to be a local minimum. To avoid noise introducing dense pseudo-extremes, it is preferable to introduce minimum interval constraints and significance constraints: the minimum interval constraint is used to limit the minimum distance between adjacent local extreme value points on the time axis. If the interval between two extreme value points is less than a threshold, the one with the more significant amplitude is retained; the significance constraint is used to limit the amplitude difference of the candidate extreme value relative to its neighborhood baseline to exceed a preset threshold. The threshold can be determined by the channel noise level, the in-batch standard deviation, or robust scaling estimation. For channels with non-constant sampling intervals, the minimum interval constraint is preferably defined by the time length rather than the number of sample points, and the neighborhood range is determined by time window backtracking, so that the extreme value determination is insensitive to changes in sampling density.

[0055] In the formation of the feature points to be tested, in addition to recording their time index, their source channel and extreme value type are also recorded, so as to aggregate evidence from different channels within the same time window during subsequent multi-channel verification. Specifically, a candidate label field can be attached to each feature point to be tested. The candidate label field at least indicates which channel triggered the feature point, whether it is a maximum or minimum value, and its significance measure value on the triggering channel. The significance measure value can be the ratio of the difference between the point value and the mean of the neighborhood, the difference between the point value and the median of the neighborhood, or the standard deviation of the neighborhood, and is used as one of the inputs for subsequent confidence calculation.

[0056] Step S402: Extract the slope reversal feature of the pH channel and the metabolic rate change feature of the lactate channel at the feature point to be tested, and determine the intersection position within the same time window based on the slope reversal feature and the metabolic rate change feature to obtain the metabolic shift feature.

[0057] In this example implementation, the slope reversal feature of the pH channel is used to characterize the turning behavior of the pH curve from decreasing to increasing or from increasing to decreasing, and the metabolic rate change feature of the lactate channel is used to characterize the behavior of changes in the net production rate or net consumption rate of lactate over time; when both are established simultaneously within the same time window, they constitute the consistency of "metabolic shift" in multi-channel evidence. In implementation, the slope reversal feature of the pH channel is preferably obtained from the first-order difference sequence in the derived features: for the time index i of the feature point to be tested, the first-order difference statistics in the preceding and following small windows are taken respectively, and it is determined whether the sign of the first-order difference mean of the preceding window and the first-order difference mean of the following window changes; if a sign change occurs and the change amplitude exceeds the preset slope threshold, then it is determined that the feature point to be tested has the slope reversal feature. The metabolic rate change characteristics of lactate channels are preferably obtained from the first-order difference sequence of lactate channels or its window statistics: the mean or median of the first-order difference of lactate is calculated in the small windows before and after the feature point to be tested, and it is determined whether it crosses from a positive change (net generation) to a negative change (net consumption) or from a negative to a positive change, and the crossing amplitude exceeds the preset rate threshold; when the lactate channel sampling is offline intermittent sampling, resulting in sparse difference, the equivalent change rate of two adjacent effective samples can be used as the metabolic rate estimate on the lactate channel, and normalized by the time interval before participating in the above crossing determination.

[0058] The intersection location within the same time window is used to bind the evidence of pH slope reversal and the evidence of lactate rate change as the same metabolic event. In specific implementation, an aligned time window W(i) can be constructed at the time index i of the feature point to be tested, with its center at the timestamp corresponding to the time index and the window width being a preset duration. Within this time window, candidate index sets satisfying the pH slope reversal feature and candidate index sets satisfying the lactate metabolic rate change feature are searched respectively, and the intersection of the two is taken as the intersection location set. When the intersection location set is empty, the nearest neighbor matching substitution rule can be used, that is, if both sets have elements within the time window, the pair with the smallest time distance is taken as the approximate intersection location, and this approximation degree is used as the penalty term for subsequent confidence. When the intersection location set contains multiple indices, the index with the largest significance metric or the index with the highest degree of synchronization between the two channels can be used as the representative location of the metabolic shift feature, and the remaining indices are recorded as auxiliary evidence.

[0059] Step S403: Match the metabolic shift characteristics with the dissolved oxygen rebound characteristics of the dissolved oxygen channel within a time window to update the anchor candidate points, and determine the multiple biological milestone anchor points based on the updated anchor candidate points' positions in the batch time series.

[0060] In this example implementation, the dissolved oxygen rebound feature is used to characterize the behavior of the dissolved oxygen curve rebounding after a period of decline. This behavior can be characterized by the first-order difference of the dissolved oxygen channel turning from negative to positive and maintaining a certain persistence. In specific implementation, it is preferable to calculate the first-order difference of dissolved oxygen on the filtered sequence of the dissolved oxygen channel, and construct a matching time window W_DO near the feature point to be measured or the location of the metabolic shift feature. Within this matching time window, candidate indices that satisfy the condition that the difference between the preceding window is negative and the difference between the following window is positive are searched, and the positive difference of the following window is required to be maintained within a continuous number of sampling points or a continuous time length to exclude short-term noise rebound. The time window matching is used to determine whether the metabolic shift feature and the dissolved oxygen rebound feature can be attributed to the same milestone event: when the location of the metabolic shift feature and the candidate index of the dissolved oxygen rebound meet the maximum allowable time deviation within the same matching time window, the two are bound as the same anchor point candidate point; when there are multiple candidate indices of dissolved oxygen rebound, it is preferable to select the index with the smallest time distance from the location of the metabolic shift feature and the largest significance of the dissolved oxygen rebound as the binding point. The anchor candidate point update involves filtering and refining the original set of test feature points generated from local extrema, combining the matching results of metabolic shift features and dissolved oxygen rebound features: candidate points that satisfy the matching relationship are retained, and their positions can be corrected from the original test feature point positions to metabolic shift feature positions or a weighted position of both; candidate points that do not satisfy the matching relationship are eliminated or downgraded to low-confidence candidate points. The updated anchor candidate points' positions in the batch time series are output as the anchor positions of the multiple biological milestone anchors, and their anchor type is marked as a milestone category related to metabolic shift, a milestone category related to dissolved oxygen rebound, or a composite milestone category combining both.

[0061] Step S404: Calculate the anchor confidence level corresponding to each of the multiple biological milestone anchor points based on the verification results of the time window matching.

[0062] In this example implementation, the input for calculating the anchor confidence score preferably includes at least: the significance of the pH slope reversal feature, the significance of the lactate metabolism rate change feature, the significance of the dissolved oxygen rebound feature, and the synchronicity of the three types of features in the time window matching. Furthermore, the above quantities can be normalized to a uniform scale to construct a weighted scoring function to obtain the anchor confidence score. The weights reflect the relative contribution of different pieces of evidence to the determination of this type of anchor point; the weights can be preset or obtained from historical batch statistics. The synchronicity score can be taken as a monotonically decreasing function of the time difference between the metabolic shift feature position and the dissolved oxygen rebound feature position, so that the smaller the time difference, the higher the confidence score.

[0063] In step S230, a piecewise mapping function from physical time to pseudo time is constructed based on the multiple biological milestone anchors, and the piecewise mapping function is used to convert the batch time series into a pseudo time series.

[0064] In this example implementation, the pseudo-time is not a new sample, but a mapping representation of the existing physical time index. Therefore, the pseudo-time series can maintain the same set of values ​​as the batch time series, only replacing the index from physical time to pseudo-time, or simultaneously storing the association table of physical time index and pseudo-time index to support traceable dual-index access. The segmentation of the segmented mapping function originates from the position division of the multiple biological milestone anchor points on the physical time axis, that is, the physical time interval between adjacent anchor points corresponds to the adjacent pseudo-time node interval on the pseudo-time axis, so that the mapping function is calculated with the same set of parameters in each anchor point interval and remains continuous at the anchor point position. The basic constraints for constructing the segmented mapping function preferably include at least monotonicity and continuity. Monotonicity is used to ensure that the pseudo-time does not backtrack as physical time advances, so that the mapping can be used as a valid time axis for sequence reparameterization; continuity is used to ensure that there are no jumps at the boundaries of adjacent segments, so that the pseudo-time coordinates at the anchor point position are consistent with the segment boundary. Furthermore, each segment of the segmented mapping function can determine the calculation relationship from physical time to pseudo-time with a finite number of segmented mapping parameters. Furthermore, to avoid locking the implementation to a specific function form, the piecewise mapping function can be linearly piecewise or alternative forms such as monotonic splines or monotonic polynomials, and is not limited to the methods listed in this example implementation.

[0065] refer to Figure 5 As shown, in this example implementation, a piecewise mapping function from physical time to pseudo-time can be constructed through the following steps S501 to S503. Wherein: Step S501: Assign preset pseudo-time nodes to the multiple biological milestone anchor points respectively.

[0066] In this example implementation, pseudo-time nodes can be set in two ways: one is discrete sequence setting, where the k-th anchor point is assigned a pseudo-time node τ_k=k or τ_k=k·Δτ, where Δτ is a fixed step size; the other is physical meaning setting, where pseudo-time nodes with business meaning are pre-assigned to different anchor point types, for example, using a normalized interval of [0,1] to represent the process from the start to the end of a batch, and assigning fixed normalized coordinates to key anchor points. Preferably, the assignment of pseudo-time nodes is accompanied by the output of a list of corresponding anchor point positions and pseudo-time nodes, where the anchor point position is taken from the timestamp or time index of the multiple biological milestone anchor points on the physical time axis, and the pseudo-time node is τ_k determined by the above rules.

[0067] Furthermore, to ensure that subsequent piecewise fitting is solvable, it is preferable to perform consistency checks on the above-mentioned correspondence list. For example, after the anchor points are sorted by physical time, it is checked that their corresponding pseudo-time nodes are also strictly increasing or non-decreasing; if there are cases where pseudo-time nodes are not increasing, it is preferable to correct them according to preset rules, such as reordering conflicting nodes or merging them into the same node and marking the merging relationship in the anchor point information, so as to ensure that the boundary conditions of monotonic mapping are met.

[0068] Step S502: Divide the mapping segments based on the anchor positions of the multiple biological milestone anchors on the physical time axis and the pseudo time nodes.

[0069] In this example implementation, mapping segmentation refers to dividing the entire physical time domain of the batch into several adjacent intervals according to the anchor point positions, and mapping each interval to an adjacent node interval on the pseudo-time axis. Specifically, the anchor point position sequence t_0, t_1, ..., t_m is obtained by arranging the anchor point positions in ascending order, and the corresponding pseudo-time node sequence τ_0, τ_1, ..., τ_m is taken. Then, the j-th mapping segment can be defined as the pairing of the physical time interval [t_j, t_{j+1}] and the pseudo-time interval [τ_j, τ_{j+1}]. In addition, to cover the start and end boundaries of the batch, boundary anchor points are preferably introduced. For example, the start point of the batch can be the earliest timestamp of the batch time sequence as t_0 and assigned τ_0, and the end point of the batch can be the latest timestamp as t_m and assigned τ_m. When the multiple biological milestone anchor points already include start and end anchor points, their positions can be directly used as boundary anchor points.

[0070] In some exemplary embodiments, when the interval between adjacent anchor points is less than a preset threshold, the segment length may be too short, leading to unstable fitting. In this case, anchor points can be merged according to rules or treated as the same segment boundary, and pseudo-time nodes can be merged or expanded at minimal intervals. This merging strategy is recorded in the segment mapping parameters to maintain recalculation. When some intermediate anchor points are not identified, resulting in a large interval between adjacent anchor points, the mapped segments will cover a longer physical time interval. This step can still divide the segments according to existing anchor points, but it is preferable to use stronger monotonic constraints or a more robust fitting form when fitting the segment parameters to avoid local overfitting.

[0071] Step S503: Fit monotonically continuous piecewise mapping parameters for each of the mapping segments, and summarize the piecewise mapping parameters to generate the piecewise mapping function.

[0072] In this example implementation, the piecewise mapping parameters are used to determine the computational relationship from any physical time t to pseudo-time τ within a segment. Taking linear piecewise mapping as an example, for the j-th segment, the slope a_j = (τ_{j+1} - τ_j) / (t_{j+1} - t_j) and the intercept b_j = τ_j - a_j·t_j are taken, thus obtaining τ = a_j·t + b_j; this form satisfies endpoint continuity and monotonicity when t_{j+1} > t_j and τ_{j+1} ≥ τ_j. To avoid numerical instability caused by excessively large sampling timestamp scales, it is preferable to normalize t within the segment by translation, for example, by using t' =(t-t_j) / (t {j+1}-t_j) represents the relative position within the segment, then τ=τ_j+(τ_{j+1}-τ_j)·t', and this parameterization is stored as a ternary parameter group of segment start point, segment end point, and segment pseudo-time span for quick recalculation.

[0073] In addition, during the fitting process of the piecewise mapping parameters, anchor point-related information can be introduced as a source of fitting weights, besides endpoint constraints. However, the use of weights is limited to the handling of parameter stability within this step. For example, when the confidence of the anchor point is low, the influence of the anchor point as a hard boundary on the piecewise shape can be reduced. Specifically, this can be done by setting the anchor point as a soft constraint point and approximating its pseudo-time node with a lower weight in the fitting objective function. When using linear piecewise segments, this soft constraint is usually not needed, but when using monotonic splines or higher-order piecewise forms, this weight can be used to adjust the estimation of the tangent slope or the fitting strength of intermediate control points. After fitting, it is preferable to perform monotonicity and continuity checks on the piecewise mapping parameters. Monotonicity checks can be performed by checking that the slope of each segment is non-negative or that the derivative of the spline is non-negative. Continuity checks can be performed by checking that the mapping values ​​of the segment endpoints are consistent, that is, the output of the previous segment at t_{j+1} is equal to the output of the next segment at t_{j+1}. If the verification fails, you can revert to linear segmentation or trim the local tangent slope according to the preset rules until the monotonic continuity constraint is met.

[0074] Finally, when summarizing the segmented mapping parameters to generate the segmented mapping function, the effective physical time interval, the corresponding pseudo-time interval, and the segmented mapping parameters of each segment are stored in a structured manner, so that the segmented mapping function can be queried and called.

[0075] In step S240, a preset reference trajectory is retrieved, and in the pseudo-time domain, the multiple biological milestone anchors and their related information are used as alignment constraints to perform dynamic time warping alignment between the pseudo-time series and the reference trajectory, so as to output the aligned reparameterized data.

[0076] In this example implementation, the aforementioned retrieval of a preset reference trajectory involves reading the target process trajectory data corresponding to the same process object in the current batch from a preset storage area and organizing it into a reference sequence structure that can be compared point-by-point with the pseudo-time series. The reference trajectory is preferably established using pseudo-time as an index, meaning each sampling point contains a reference pseudo-time coordinate and the corresponding multi-channel reference value. When the original record of the reference trajectory is still a physical time index, the pseudo-time coordinateization rule consistent with the current batch can be completed and stored at the preset time, or the index conversion can be completed according to the fixed pseudo-time coordinate table during retrieval, so that the reference trajectory has a definite pseudo-time axis semantic before being input into the dynamic time warping alignment. The source of the reference trajectory can be a pseudo-time series of a single gold batch, or a template trajectory generated by statistical rules in the pseudo-time domain from multiple qualified batches. When using a template trajectory, each pseudo-time coordinate of the template trajectory preferably contains at least a reference mean vector and a dispersion parameter, so that subsequent distance measurements can use this dispersion as scale information in the calculation.

[0077] The anchor point information includes at least the anchor point type, anchor point confidence level, pseudo-time coordinates of the anchor point in the pseudo-time series, and target pseudo-time coordinates or target anchor point location identifiers of the anchor point in the reference trajectory. When the reference trajectory uses a template trajectory, the pseudo-time nodes of anchor points of the same type in the template can be directly used as the target pseudo-time coordinates. Therefore, when performing dynamic time warping alignment, the alignment path is no longer solely determined by the general minimum distance principle, but rather searched within the allowable area defined by the anchor point and its related information, giving the correspondence near the stage boundary a clear source of constraint.

[0078] refer to Figure 6 As shown, in this example embodiment, dynamic time warping alignment can be performed through the following steps S601 to S604. Wherein: Step S601: Generate an alignment constraint set based on the multiple biological milestone anchors and their related information.

[0079] In this example implementation, the alignment constraint set can include at least two types: anchor constraints and bandwidth constraints. Anchor constraints specify that pseudo-time sampling points near a given anchor point should preferentially or necessarily match sampling points near the corresponding anchor point in the reference trajectory. Bandwidth constraints limit the range of path deviation from the diagonal to restrict unreasonable stretching or compression. Anchor constraints are preferably constructed in the form of anchor pairs, that is, matching the position u_k of an anchor point in the current batch (indication or pseudo-time coordinate of the pseudo-time series) with the corresponding position v_k of the reference trajectory, and adding a constraint radius and constraint strength to the anchor pair. The constraint radius can be obtained by mapping the anchor point confidence level; higher confidence levels result in a smaller radius and a harder constraint, while lower confidence levels result in a larger radius or a softer constraint. Bandwidth constraints can be constructed using segmented bandwidth, that is, specifying the allowable path offset range for each interval between adjacent anchor pairs. Specifically, this can be represented by upper and lower bound functions of the allowable window, ensuring that any pseudo-time series index i is only allowed to fall within the interval [j_min(i), j_max(i)] with the reference trajectory index j. This interval can be determined by the position difference between the two anchor point pairs and a preset multiplier threshold. Different multiplier thresholds can be set for different anchor point intervals, giving the constraint set a definite parameter input for limiting the distortion amplitude at different stages. As an alternative implementation, the alignment constraint set can also be represented by a reachable domain mask matrix, that is, using a matrix to identify whether (i,j) is allowed to match, setting the hard constraint region of the anchor point to strongly allowed and the forbidden region of the anchor point to disallowed, so that the constraint directly acts on the cost matrix filling and path recursion.

[0080] Step S602: Calculate the alignment cost matrix based on the alignment constraint set.

[0081] In this example implementation, the alignment cost matrix is ​​used to quantify the matching cost between each sample point of the pseudo-time series and each sample point of the reference trajectory, and provides input for subsequent minimum cumulative cost path search. Specifically, the alignment cost matrix can be calculated as follows: First, the signal channel weights are determined based on the anchor point confidence level, and a weighted parameter set is generated.

[0082] The signal channel weights are used to allocate the contribution of each channel to the matching cost in a multi-channel distance metric, ensuring that the cost matrix reflects the consistent principle of reliable evidence dominating and low-reliability evidence being weighted less. In a specific implementation, a mapping rule from anchor confidence to weight adjustment factors is first established, and then this adjustment factor is applied to the channel base weights to obtain the final channel weights. The channel base weights can be determined by channel sampling density, channel data quality identification statistics, channel noise level, or preset process experience weights. The weight adjustment factor can be derived from the set of evidence channels related to the anchor point. For example, when the evidence for a certain type of anchor point mainly comes from the pH, lactic acid, and dissolved oxygen channels, the confidence of that anchor point preferably applies a stronger adjustment to the weights of the aforementioned channels, while maintaining the base weights or applying a weaker adjustment to channels unrelated to that anchor point.

[0083] Secondly, the weighted distance between the pseudo-time series and the reference trajectory at each pseudo-time sampling point is calculated based on the weighted parameter set. In this example embodiment, the weighted distance refers to the scalar distance obtained by weighting and aggregating the differences in the corresponding multi-channel values ​​for each pair of (i,j) matching candidates according to the channel weights. In a specific implementation, a dimensionless processing is performed on each channel before calculating the difference. The dimensionless processing can reuse the scale normalization rules fixed in step S210, or use the same channel scale parameters for the reference trajectory and the pseudo-time series; subsequently, the channel differences can be calculated using weighted L1 distance, weighted L2 distance, or weighted Mahalanobis distance.

[0084] Finally, the weighted distance is filled into the alignment cost matrix for use in the search for the minimum cumulative cost path. Preferably, the filling process is executed in conjunction with the alignment constraint set; that is, the weighted distance is first calculated and written into the matrix for (i,j) pairs that satisfy the reachability conditions of the constraint set, and an unreachable flag or a large penalty value is written into the matrix for unreachable (i,j) pairs.

[0085] Step S603: Under the constraints of the alignment constraint set, search for the minimum cumulative cost path based on the alignment cost matrix to obtain the alignment mapping.

[0086] In this example implementation, the search for the minimum cumulative cost path employs a dynamic programming recursive approach, satisfying the monotonicity and continuous step constraints of dynamic time warping. Specifically, the path index (i,j) increments with i while j does not regress, and the allowed moves between adjacent steps are limited to a preset set of step types. In the specific implementation, for each reachable (i,j), the cumulative cost D(i,j) = C(i,j) + min{D(i-1,j), D(i,j-1), D(i-1,j-1)} is calculated, where C(i,j) is an element of the cost matrix, and the three minimum values ​​correspond to three step types. The bandwidth constraint in the alignment constraint set ensures that the recursion only occurs within the allowed region by limiting the reachable domain; the anchor constraint influences path selection by considering both the reachable domain of the anchor's neighborhood and the cost bias. During path backtracking, the (i,j) sequence can be obtained by backtracking from the endpoint or termination boundary according to the predecessor pointer of the record, and this sequence is then organized into an alignment map.

[0087] Step S604: Reparameterize the pseudo-time series based on the alignment mapping to output aligned reparameterized data.

[0088] In this example implementation, reparameterization refers to adding or replacing the reference coordinates of each sampling point in the pseudo-time series without changing the original channel values, so that it can form a directly comparable data sequence in the reference trajectory coordinate system. For example, the alignment mapping is that the current batch pseudo-time coordinate u_i corresponds to the reference pseudo-time coordinate v_{f(i)}, and an aligned pseudo-time coordinate is generated for each sampling point accordingly. When the reference trajectory is a discrete sampling point sequence, v_{f(i)} can directly take the reference pseudo-time coordinate corresponding to the reference trajectory index. When the reference trajectory is a continuous template, the template can be queried at v_{f(i)} to obtain the reference value and the coordinate can be written into the reparameterized data. The reparameterized data can be output in the form of records. Each record contains at least one of the following: the original pseudo-time coordinate of the current batch, the aligned reference pseudo-time coordinate, the channel value vector, and the local cost or cumulative cost of the mapping path, for subsequent auditing. The local cost can be given by C(i,f(i)), and the stage cumulative cost can be obtained by the projection of D(i,f(i)) onto the backtracking path. If a reference trajectory index corresponds to multiple current batch indices (due to many-to-one mapping caused by horizontal / vertical steps in the path), these records can be retained as multiple records according to the rules, or multiple records under the same reference index can be sorted and aggregated according to physical time or original pseudo time.

[0089] In one exemplary embodiment of this disclosure, the output aligned reparameterized data further includes compensation for missing sampling points and uncertainty annotation. Specifically: First, missing sampling points are identified in the reparameterized data, and the corresponding pseudo-time coordinates are determined. Missing sampling points can be identified by records marked as missing / invalid by the channel data quality identifier, and their aligned reference pseudo-time coordinates are used as the pseudo-time coordinates of the missing point. When the missing point occurs in the original pseudo-time series but the coordinates are repeated after alignment, the unique key of the missing point is determined by the reference pseudo-time coordinates and the channel identifier.

[0090] Subsequently, the pseudo-temporal conditional distribution of historical batches is retrieved, and distribution parameters are obtained at the pseudo-temporal coordinates. The pseudo-temporal conditional distribution refers to the distribution of historical observations in the same channel under the same pseudo-temporal coordinates (or their neighborhood intervals); this distribution can be estimated from the sample set of historical qualified batches aligned in the pseudo-temporal domain and stored in parameterized form. The distribution parameters can be implemented as the mean and variance of a Gaussian distribution, a set of quantiles, the bandwidth parameter of kernel density estimation, or the weights and component parameters of a mixture model. When using a set of quantiles, the quantile interval can be directly used as the source of uncertainty output during compensation, thus avoiding strong assumptions about the distribution shape.

[0091] Next, posterior correction is performed based on the distribution parameters and the pseudo-temporal neighborhood observations of the missing sampling points to obtain a compensation value. Preferably, the neighborhood observations are the nearest valid points before and after the missing point in the same channel, along with their pseudo-temporal coordinates. The neighborhood observations and the conditional distribution are used together to generate the compensation value. For example, in one implementation, the conditional distribution can be considered as a prior, and the locally linear extrapolated or locally smoothed predictions given by the neighborhood observations can be considered as the likelihood center. A weighted fusion is used to obtain the posterior mean as the compensation value. The fusion weights can be determined by the neighborhood span, the quality of the neighborhood data, and the dispersion of the historical distribution, so that a larger neighborhood span or lower neighborhood quality depends more on the historical distribution, and vice versa.

[0092] Finally, when writing the uncertainty value corresponding to the compensation value as a metadata tag into the reparameterized data, the uncertainty value can be the posterior variance, confidence interval width, or confidence score, and written together with the compensation value, channel identifier, and pseudo-time coordinate; when the uncertainty exceeds the threshold, a participation flag can be written into the metadata at the same time to indicate that the compensation value will be processed according to the deweighting or elimination rules in subsequent analysis.

[0093] This example embodiment also provides an automated data acquisition and processing system for a biopharmaceutical laboratory. (Reference) Figure 7 As shown, the automated data acquisition and processing system 700 for a biopharmaceutical laboratory may include a signal acquisition module 710, an anchor point recognition module 720, a sequence conversion module 730, and a regularization and alignment module 740. Wherein: The signal acquisition module 710 is configured to acquire experimental process signals during the biopharmaceutical batch experiment to form a batch time series, and calculate derived features for event detection based on the batch time series. The anchor point identification module 720 is configured to identify multiple biological milestone anchor points that meet preset judgment conditions based on the batch time series and the derived features, and generate corresponding anchor point confidence scores for the multiple biological milestone anchor points. The sequence conversion module 730 is configured to construct a piecewise mapping function from physical time to pseudo-time based on the multiple biological milestone anchor points, and use the piecewise mapping function to convert the batch time series into a pseudo-time series. The normalization and alignment module 740 is configured to retrieve a preset reference trajectory, and perform dynamic time normalization and alignment on the pseudo-time series and the reference trajectory in the pseudo-time domain using the multiple biological milestone anchor points and their related information as alignment constraints, so as to output aligned reparameterized data.

[0094] The specific details of each module of the aforementioned automated data acquisition and processing system for biopharmaceutical laboratories have been described in detail in the corresponding methods for automated data acquisition and processing in biopharmaceutical laboratories, so they will not be repeated here.

[0095] refer to Figure 8 As shown, an electronic device capable of implementing the above-described method is also provided. The electronic device 800 includes a processor 801 and a memory 802. The memory 802 stores computer-readable instructions, which, when executed by the processor 801, implement the method of this disclosure.

[0096] In an exemplary embodiment of this disclosure, a computer-readable storage medium is also provided, having stored thereon computer program code instructions that, when invoked by a processor, execute the method described in the embodiments.

[0097] refer to Figure 9 As shown, a program product 900 for implementing the above-described method according to an embodiment of the present disclosure is described. This product may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0098] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0099] Finally, the above preferred embodiments are only used to illustrate the technical solutions of this disclosure and are not restrictive. Although this disclosure has been described in detail, those skilled in the art should understand that changes in form and detail can be made without departing from the scope defined by the claims of this disclosure. The dimensions in the drawings are not related to the specific physical object, and the physical object dimensions can be arbitrarily changed.

Claims

1. An automated data acquisition and processing method for a biopharmaceutical laboratory, characterized in that, include: The experimental process signals during the batch experiment of biopharmaceuticals are collected to form a batch time series, and derived features for event detection are calculated based on the batch time series. Based on the batch time series and the derived features, multiple biological milestone anchors that meet preset judgment conditions are identified, and corresponding anchor confidence scores are generated for the multiple biological milestone anchors. Based on the multiple biological milestone anchors, a piecewise mapping function from physical time to pseudo time is constructed, and the piecewise mapping function is used to convert the batch time series into a pseudo time series. A preset reference trajectory is retrieved, and in the pseudo-time domain, the multiple biological milestone anchors and their related information are used as alignment constraints to perform dynamic time warping alignment between the pseudo-time series and the reference trajectory, so as to output the aligned reparameterized data.

2. The automated data acquisition and processing method for biopharmaceutical laboratories according to claim 1, characterized in that, The calculation of derived features for event detection includes: Denoising filtering is performed on the batch time series to obtain the filtered sequence; Based on the filtered sequence, calculate the first-order difference sequence and / or the second-order difference sequence to obtain the difference features; The derived features are formed by calculating the sliding window statistics based on the filtered sequence and the differential features.

3. The automated data acquisition and processing method for biopharmaceutical laboratories according to claim 1, characterized in that, The identification of multiple biological milestone anchors that meet preset judgment conditions includes: Local extreme points are extracted from the batch time series to obtain the feature points to be tested; At the feature points to be measured, the slope reversal features of the pH channel and the metabolic rate change features of the lactate channel are extracted, and the intersection position within the same time window is determined based on the slope reversal features and the metabolic rate change features to obtain the metabolic shift features. The metabolic shift characteristics are matched with the dissolved oxygen rebound characteristics of the dissolved oxygen channel within a time window to update the anchor candidate points, and the multiple biological milestone anchor points are determined based on the positions of the updated anchor candidate points in the batch time series. The anchor confidence level corresponding to each of the multiple biological milestone anchor points is calculated based on the verification results of the time window matching.

4. The automated data acquisition and processing method for biopharmaceutical laboratories according to claim 1, characterized in that, The construction of the piecewise mapping function from physical time to pseudo-time includes: Each of the multiple biological milestone anchor points is assigned a preset pseudo-time node; Based on the anchor point positions of the multiple biological milestone anchor points on the physical time axis and the pseudo-time nodes, the mapping segments are divided. For each of the aforementioned mapping segments, a monotonically continuous piecewise mapping parameter is fitted, and the piecewise mapping parameters are summarized to generate the piecewise mapping function.

5. The automated data acquisition and processing method for a biopharmaceutical laboratory according to claim 1, characterized in that, The execution of dynamic time warping alignment includes: An alignment constraint set is generated based on the multiple biological milestone anchors and their related information. Calculate the alignment cost matrix based on the aforementioned alignment constraint set; Under the constraints of the alignment constraint set, the minimum cumulative cost path is searched based on the alignment cost matrix to obtain the alignment mapping; The pseudo-time series is reparameterized based on the alignment mapping to output aligned and reparameterized data.

6. The automated data acquisition and processing method for a biopharmaceutical laboratory according to claim 5, characterized in that, The calculation of the alignment cost matrix includes: The signal channel weights are determined based on the anchor point confidence level, and a weighted parameter set is generated. Calculate the weighted distance between the pseudo-time series and the reference trajectory at each pseudo-time sampling point based on the weighted parameter set; The weighted distance is filled into the alignment cost matrix for use in the search of the minimum cumulative cost path.

7. The automated data acquisition and processing method for a biopharmaceutical laboratory according to claim 5, characterized in that, The output aligned reparameterized data includes: Identify missing sampling points in the reparameterized data and determine the pseudo-time coordinates corresponding to the missing sampling points; Retrieve the pseudo-time conditional distribution of historical batches and obtain the distribution parameters at the pseudo-time coordinates; Based on the distribution parameters and combined with the pseudo-temporal neighborhood observations of the missing sampling points, a posterior correction is performed to obtain the compensation value; The uncertainty value corresponding to the compensation value is written as a metadata tag into the reparameterized data.

8. An automated data acquisition and processing system for a biopharmaceutical laboratory, characterized in that, include: The signal acquisition module is configured to acquire experimental process signals during the batch experiment of biopharmaceuticals to form a batch time series, and to calculate derived features for event detection based on the batch time series. An anchor point identification module is configured to identify multiple biological milestone anchor points that meet preset judgment conditions based on the batch time series and the derived features, and generate corresponding anchor point confidence scores for the multiple biological milestone anchor points. The sequence conversion module is configured to construct a segmented mapping function from physical time to pseudo time based on the multiple biological milestone anchors, and to use the segmented mapping function to convert the batch time series into a pseudo time series. The normalization and alignment module is configured to retrieve a preset reference trajectory and perform dynamic time normalization and alignment on the pseudo-time series and the reference trajectory in the pseudo-time domain using the multiple biological milestone anchors and their related information as alignment constraints, so as to output the aligned reparameterized data.

9. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the method of any one of claims 1-7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

Images