A cell research-oriented experimental process data management method and system

Abstract

The application discloses a kind of experimental process data management methods and systems for cell research and development, it is related to cell research and development experimental data and process management technical field, including: the edge distance of each hole site in experimental plate is acquired and disturbance event chain is constructed, based on the duration and the amount of recording liquid, the delayed drift potential is extracted, the drift contribution value is determined in combination with hole site observation value, further, the drift zone depth and cumulative progression are extracted;The drift zone depth, cumulative progression and average drift intensity are constructed as environmental state and input into reinforcement learning model, and the process management action is output;Based on the drift contribution value, the reliable coefficient is determined, the hole site data routing processing is completed, and the management account book is generated.The application can identify the delayed drift influence of edge inward progression in multi-well plate cell experiment, improve the reliability of data hierarchical management, review archiving and result analysis.

Description

Technical Field

[0001] This invention relates to the field of cell research and development experimental data and process management technology, and in particular to an experimental process data management method and system for cell research and development. Background Technology

[0002] In cell research and development, multi-well experimental plates are widely used in cell expansion, drug screening, live-cell imaging, endpoint detection, and batch process evaluation. However, due to spatial differences between the edges and inner circles of different wells within the same plate, and the fact that experimental procedures typically involve multiple operations such as sample addition, door opening, transport, imaging, and medium replacement, the changes in culture medium volume, exposure time, and cumulative environmental fluctuations are often not entirely consistent across different wells. Especially under long-term culture and multi-node observation conditions, the impact of disturbances in the edge region can easily extend inwards along the experimental process, causing the observation results of some wells to deviate from the nominal experimental baseline, thus reducing the comparability between wells on the same template. The resulting technical challenge is how to uniformly manage the disturbance history, state changes, and data reliability of each well throughout the entire cell research and development experimental process, promptly identify delayed drift phenomena progressing from the edge inwards, and accordingly complete subsequent data classification, process management, and result archiving.

[0003] Current cell research and development experimental management methods mostly focus on recording experimental design parameters, collecting raw observation results, and saving endpoint analysis data. They typically assume that wells within the same experimental plate are under consistent conditions, lacking the ability to continuously record and analyze well-level temporal perturbations. For operations such as opening doors, transporting samples, imaging, and changing media that occur multiple times in the experimental process, existing methods often only allow for scattered recording, making it difficult to establish a complete perturbation chain for individual wells. Furthermore, it is difficult to quantify the data instability trends caused by differences in edge positions and the accumulation of process perturbations in different wells. When experimental results deviate, it is often difficult to distinguish whether the deviation originates from actual biological changes or from positional delayed drift effects in the experimental process. This can easily lead to delayed identification of abnormal wells, inaccurate retesting triggers, mixed data archiving, and distortion of subsequent statistical analysis, ultimately affecting the reliability and traceability of cell research and development experimental results. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as the inability to uniformly characterize the process disturbance history of each well in multi-well plate cell experiments, the difficulty in identifying the delayed drift effect that progresses from the edge inward, the tendency to misjudge the reliability of well data, and the insufficient reliability of subsequent review, archiving, and result analysis. Therefore, this invention proposes an experimental process data management method and system for cell research and development.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: A data management method for experimental workflows in cell research and development, comprising: S1. Obtain the edge distance of each hole in the experimental board, and map the disturbance event corresponding to each hole to the time axis to generate the disturbance event chain of the corresponding hole. S2. Based on the edge distance and the duration and volume of recording liquid contained in the disturbance event chain, extract the late drift potential accumulated over time for each of the well sites. S3. Determine the drift contribution value of the corresponding hole position based on the delayed drift potential and the observed value of each hole position, and extract the drift band depth and cumulative progression in the experimental plate according to the correspondence between the edge distance and the drift contribution value. S4. Construct the environment state by the drift zone depth, cumulative increment, and average drift intensity obtained based on the drift contribution value, and input it into the reinforcement learning model. Output the corresponding process governance action through the reinforcement learning model. S5. Determine the reliability coefficient of each of the holes based on the drift contribution value, complete the hole data routing process according to the reliability coefficient and the process governance action, and generate a management ledger.

[0006] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention continuously models the edge distance, disturbance event chain, delayed drift potential, drift contribution value, drift band depth, and cumulative progression of each well in the experimental plate. This transforms the disturbance information, originally scattered across process nodes such as sample addition, door opening, transfer, imaging, and liquid replacement, into calculable, traceable, and comparable well-level state quantities. This enables the identification of the range and extent of delayed drift effects that propagate inward from the edge during the experimental process. In multi-well plate cell experiments, it can more accurately distinguish between data offsets caused by accumulated process disturbances and the actual observed changes that should be presented under the nominal experimental conditions of the well itself, improving the ability to interpret well state changes and perceive the overall drift state of the experimental plate.

[0007] 2. This invention further completes the classification of well location data, generation of process governance actions, data routing processing, and construction of management ledgers based on drift contribution value and credibility coefficient. This enables data with different credibility levels to enter the corresponding review, archiving, and result summary paths. Furthermore, it combines a reinforcement learning model to dynamically adjust subsequent collection, retesting, and archiving actions. This not only improves the pertinence and temporal consistency of experimental data management but also reduces the continuous interference of data affected by delayed drift on experimental conclusions, enhances the reliability of results summary with the same template, the integrity of experimental process traceability, and the stability of subsequent analysis and evaluation. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart illustrating an experimental workflow data management method for cell research and development, provided as an embodiment of the present invention; Figure 2 This is a functional module diagram of an experimental workflow data management system for cell research and development, provided as an embodiment of the present invention. Detailed Implementation

[0009] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0010] Example: This example provides a data management method for experimental workflows in cell research and development. See [link to example]. Figure 1 Specifically, including: S1. Obtain the edge distance of each hole in the experimental board, and map the disturbance event corresponding to each hole to the time axis to generate the disturbance event chain of the corresponding hole. S2. Based on the edge distance and the duration and volume of recording liquid contained in the disturbance event chain, extract the late drift potential accumulated over time for each of the well sites. S3. Determine the drift contribution value of the corresponding hole position based on the delayed drift potential and the observed value of each hole position, and extract the drift band depth and cumulative progression in the experimental plate according to the correspondence between the edge distance and the drift contribution value. S4. Construct the environment state by the drift zone depth, cumulative increment, and average drift intensity obtained based on the drift contribution value, and input it into the reinforcement learning model. Output the corresponding process governance action through the reinforcement learning model. S5. Determine the reliability coefficient of each of the holes based on the drift contribution value, complete the hole data routing process according to the reliability coefficient and the process governance action, and generate a management ledger.

[0011] In an embodiment of the present invention, the edge distance of each hole position within the experimental board is obtained, and the disturbance events corresponding to each hole position are mapped to the time axis to generate a disturbance event chain for the corresponding hole position, including: For multi-well plate cell research experiments, the well coordinates were first defined, edge distances were calculated, and perturbation event chains were constructed. The total number of rows in the experimental plate was defined as follows. The total number of columns is , and All values ​​are positive integers, corresponding to the row and column dimensions of a standard multi-well cell culture plate. The first row at the top of the experimental plate is used as the starting position for row coordinate counting. From 1 to consecutive integers, corresponding to the first row to the second row of the experimental board. The first column on the left side of the experimental board is used as the starting position for column coordinate counting. From 1 to consecutive integers, corresponding sequentially to the first column to the second column of the experimental board. Each hole in the plate corresponds to a unique row coordinate. with column coordinates The edge distance of each hole is calculated based on row and column coordinates. The calculation formula is: In the formula, Let be the row coordinate of the i-th hole, and R be the total number of rows on the experimental board. Let be the column coordinate of the i-th hole, and C be the total number of columns on the experimental plate. To achieve the minimum value calculation, this formula precisely quantifies the distance from the hole to the nearest plate boundary by taking the minimum value of the distance from the hole to the four plate boundaries (top, bottom, left, and right). This directly corresponds to the basic risk of the hole being affected by the edge effect. The smaller the edge distance value, the higher the basic sensitivity of the hole to environmental disturbances, matching the physical characteristic that the edge effect of the perforated plate gradually decreases from the outside to the inside.

[0012] Complete the collection and standardized recording of disturbance events. All events occurring throughout the experimental process, including sample addition, door opening, transport, imaging, and medium replacement, are considered disturbance events. For each disturbance event, define the start and end points of its duration. The duration of a sample addition event begins when culture medium is injected into the well and ends when all wells are filled. The duration of a door opening event begins when the incubator door opens and ends when it is completely closed. The duration of a transport event begins when the experimental plate leaves the culture environment and ends when it is returned to the target culture environment. The duration of an imaging event begins when the experimental plate enters the imaging device and ends when it leaves the imaging device after imaging all set fields of view. The duration of a medium replacement event begins when the original culture medium is aspirated from the well and ends when all target wells are replenished with new culture medium. For each disturbance event, record the occurrence time synchronously according to the chronological order of its occurrence. The duration of the disturbance event, determined according to the aforementioned rules, and the recorded culture medium volume at the corresponding well location at the time of the event occurrence are recorded. The recorded culture medium volume is uniformly taken as the steady-state volume of the culture medium in the well before the starting node of the corresponding disturbance event. For disturbance events executed uniformly across the entire plate, the recorded culture medium volume of all well locations across the entire plate is recorded simultaneously at the time of the event occurrence. For disturbance events executed independently in a single well or some well locations, only the recorded culture medium volume of the corresponding operating well location at the time of the event occurrence is recorded. Finally, the disturbance event chain is generated. For each well location in the plate, all disturbance events corresponding to that well location in the entire experimental process are arranged in ascending order according to the order of their occurrence time. The occurrence time, duration, and recorded culture medium volume at the corresponding occurrence time of each disturbance event are treated as individual event entries and assembled sequentially to form the disturbance event chain corresponding to that well location. This ensures that the disturbance event chain of each well location completely covers all environmental disturbance operations it experiences in the entire experimental process, providing complete temporal data support for the subsequent calculation of the late drift potential.

[0013] It should be noted that an experimental plate refers to a plate-shaped experimental carrier containing multiple wells, commonly a multi-well cell culture plate, such as a 6-well plate, 24-well plate, 96-well plate, or 384-well plate. Experimental plates are used to perform cell culture, drug treatment, imaging detection, and endpoint analysis in parallel. A well is an individual, independent unit on the experimental plate used to hold cells, culture medium, or treatment reagents. Each well is a relatively independent experimental space, corresponding to a specific set of cell culture conditions, treatment conditions, and observation results. Edge distance refers to the distance from a well to the nearest outer boundary of the experimental plate. The smaller the edge distance, the closer the well is to the edge of the experimental plate, and the more susceptible it is to boundary effects such as edge evaporation and temperature and humidity fluctuations.

[0014] It's important to note that a sample addition event refers to the process of adding cell suspension, culture medium, drugs, dyes, or other reagents to the wells. This event alters the liquid composition, total volume, and starting point of subsequent reactions within the wells. A door opening event refers to the opening and closing of the door of an incubator, environmental control chamber, or related enclosed equipment. This process may cause changes in temperature, humidity, gas concentration, and external disturbances, affecting the well environment. A transport event refers to the movement of the experimental plate between different devices, spatial locations, or environments, such as from an incubator to a microscope, imaging equipment, or worktable. This event may introduce vibration, temperature differences, and changes in exposure time. An imaging event refers to the process of acquiring images or visually inspecting the wells within the experimental plate using a microscope, live-cell imaging system, or other detection equipment. Imaging events typically involve equipment dwell time, environmental changes, or additional exposure. A medium change event refers to the process of removing the existing culture medium from the wells and adding new culture medium, processing solution, or buffer solution. This event alters the liquid composition, volume, and the cell microenvironment within the wells. A disturbance event is a collective term for operational events that alter the actual environmental state of a well during an experiment. In this passage, disturbance events specifically include sample addition events, door opening events, transport events, imaging events, and medium change events. Essentially, they are procedural behaviors that may affect the actual cell exposure conditions within the well. A disturbance event chain refers to a sequence of events for a specific well, formed by connecting all disturbance events it experiences during the entire experimental process in chronological order. This event chain can comprehensively describe the environmental disturbance history experienced by that well from the start to the end of the experiment.

[0015] In an embodiment of the present invention, based on the edge distance and the duration and recording volume included in the disturbance event chain, the late drift potential accumulated over time for each of the well sites is extracted, including: The volume sensitivity corresponding to a single disturbance event is calculated. Volume sensitivity characterizes the responsiveness of the well culture medium system to disturbance events. For a single disturbance event, the volume sensitivity is calculated based on the duration of the event and the corresponding recorded volume. The calculation formula is as follows: In the formula, Let be the duration of the k-th perturbation event. Let represent the volume of culture medium recorded at the i-th well at the moment the k-th perturbation event occurs. This calculation formula quantifies the correlation between perturbation duration and system volume by using the ratio of duration to recorded volume. A longer duration and a smaller volume of culture medium in the well result in a higher volume sensitivity value, indicating a stronger influence of the perturbation event on the corresponding well. This aligns with the cell culture experiment characteristic where small-volume culture systems are more sensitive to environmental disturbances. The edge response level corresponding to this well is calculated. The edge response characterizes the difference in basic perturbation response caused by the difference in distance between the well and the plate boundary. The edge response level is calculated based on the edge distance of the well, using the following formula: In the formula, Let be the edge distance of the i-th hole. This calculation formula uses the reciprocal of the edge distance to quantify the higher the edge response of holes closer to the plate edge. The outermost holes with an edge distance of 0 correspond to the highest basic response intensity, while the basic response intensity of the innermost holes with larger edge distances gradually decreases, accurately matching the physical law that the edge effect of a multi-hole plate gradually weakens from the outside to the inside. The formula also calculates the delayed propagation degree corresponding to a single disturbance event. The delayed propagation degree characterizes that the impact of a disturbance event is not synchronously manifested at the time of the event, but rather exhibits a delayed propagation characteristic from the outside to the inside over time. Based on the time difference between the current moment and the time of the corresponding disturbance event, and combined with the edge distance of the hole, the delayed propagation degree is calculated using the following formula: In the formula, For the current calculation time, Let k be the time when the perturbation event occurs. The time difference between the current moment and the moment the disturbance event occurred. Let be the edge distance of the i-th hole. This calculation formula uses the ratio of the time difference to the edge distance superimposed on the time difference to quantify the gradual manifestation of the disturbance effect over time. For the same disturbance event, the larger the time difference, the higher the degree of delayed expansion and the more significant the lag effect of the disturbance. For different holes with the same time difference, the hole with the smaller edge distance has a higher degree of delayed expansion and the corresponding disturbance effect appears earlier. The inner hole with a larger edge distance has a more gradual increase in the degree of delayed expansion and the corresponding disturbance effect appears later.

[0016] Based on the aforementioned volume sensitivity, edge response degree, and delayed propagation degree, the single delayed contribution of a single disturbance event to the corresponding aperture is calculated using the following formula: Substituting the aforementioned three sub-calculation formulas, the complete formula for calculating the contribution of a single delayed payment is as follows: The definitions of each parameter in the formula are consistent with the definitions of the parameters in the aforementioned sub-item calculation formula. This calculation formula fully quantifies the hysteretic effect of a single disturbance event on the corresponding well site at the current moment through the product of three core influencing factors. It also covers three core dimensions: system volume sensitivity, edge-based response difference, and disturbance delayed propagation characteristics, ensuring that the calculation results of the single delayed contribution match the actual disturbance influence law in multi-well plate cell experiments.

[0017] To calculate the delayed drift potential of the borehole at the current time, for all perturbation events in the perturbation event chain of the target borehole, the single delayed contribution corresponding to each perturbation event at the current time is accumulated to obtain the delayed drift potential of the borehole at the current time. The calculation formula is as follows: In the formula, K represents the total number of disturbance events included in the target aperture disturbance event chain. Let be the single delayed contribution of the k-th perturbation event to the i-th aperture at the current time t. For summation, this calculation formula quantifies the potential drift degree accumulated at the current moment by the sum of the single delayed contributions of all disturbance events experienced by the aperture in the entire experimental process. The higher the value of the delayed drift potential, the stronger the trend of the aperture being affected by the cumulative process disturbance, and the higher the probability of observation drift, providing a core quantitative indicator for the subsequent identification of drift bands.

[0018] It should be noted that delayed drift potential refers to the potential drift intensity accumulated at a specific well location due to multiple disturbance events experienced during the experimental process at the current moment. This quantity does not directly represent the actual deviation that has occurred in a particular observation, but rather characterizes the strength of the tendency for the well location to continue to exhibit state shifts or result distortions under the combined influence of factors such as edge location, recording fluid volume, duration of disturbance, and delayed propagation over time after the event. A larger delayed drift potential indicates a more significant cumulative impact of process disturbances on the well location during the experiment, making subsequent observations more likely to deviate from the nominal baseline of the same template well location; a smaller delayed drift potential indicates a weaker cumulative disturbance effect on the well location, and a higher probability that the well location data will remain stable. Delayed drift potential can serve as a fundamental state quantity for identifying potential drift risks at well locations, analyzing delayed drift bands that progress from the edge inwards, and supporting the generation of subsequent process control actions.

[0019] In an embodiment of the present invention, the drift contribution value of the corresponding hole position is determined based on the delayed drift potential and the observed values ​​of each hole position, and the drift band depth and cumulative progression within the experimental plate are extracted according to the correspondence between the edge distance and the drift contribution value, including: Obtain the nominal experimental template identifier for each well. This identifier is a unique group identifier pre-assigned to each well during the experimental design phase. Wells with the same nominal experimental template identifier correspond to the same nominal experimental conditions: cell seeding density, culture medium composition, drug treatment conditions, culture duration, and observation node settings. This ensures that wells with the same template have a consistent nominal experimental baseline. Subsequently, among wells with the same template identifier, extract the median value of the observation at the current observation time as the nominal reference center. The calculation formula is as follows: In the formula, For the current observation time, This is the nominal experimental template identifier for the i-th well position. This serves as the nominal experimental template identifier for the j-th well position. For the j-th aperture at the current observation time The observed values ​​are quantitative detection results of cell viability, proliferation rate, fluorescence intensity, etc., collected at corresponding observation times during cell research and development experiments. To achieve median-based calculations, this formula uses the median rather than the arithmetic mean as the reference center. This effectively avoids interference from individual extreme outliers in the same template wells, ensuring that the nominal reference center accurately reflects the baseline experimental results of the same template wells under drift-free conditions. This matches the real-world scenario where random outliers are prone to occur in multi-well cell development experiments. Before calculating the deviation of the observed values ​​from the nominal reference center, all observed values ​​in the same template wells undergo 0-1 normalization preprocessing. The normalization formula is as follows: In the formula, These are the normalized borehole position observations. This represents the minimum value of the observed values ​​at the same template borehole location at the current observation time. Normalization, where the maximum observed value of each well site within the same template is the maximum value at the current observation time, reduces the impact of dimensional and range differences in different types of observed values ​​on subsequent calculation results. This ensures a unified benchmark for calculating drift contribution values ​​under different experimental scenarios. The normalized observed values ​​are used for calculating the degree of deviation and drift contribution value. In actual calculations, to avoid a situation where all observed values ​​of the same well site within the same template are equal at a certain observation time, resulting in a zero denominator for the normalization, this embodiment further stipulates that when the maximum and minimum observed values ​​of the same well site within the same template are equal at the current observation time, the normalized observed values ​​of all well sites within that group are uniformly recorded as zero at the current observation time, and the corresponding nominal reference center is simultaneously recorded as zero. At this time, the degree of deviation of each well site relative to the nominal reference center is recorded as zero, and the corresponding drift contribution value only retains the state record function of the late drift potential without triggering amplification calculations based on the degree of deviation.

[0020] The deviation between the observed values ​​at each well site and the corresponding nominal reference center is calculated using the following formula: In the formula, For the i-th aperture at the current observation time Normalized observations For the i-th borehole, the boreholes on the same template at the current observation time... The nominal reference center of the normalized observations is calculated using an absolute value to quantify the degree of deviation. This formula can indiscriminately cover both positive and negative deviations of the observations relative to the nominal reference center, ensuring that the magnitude of the deviation from the baseline result is accurately quantified regardless of whether the observations are too high or too low, avoiding underestimation of the drift degree caused by unidirectional deviation calculation. Then, the delayed drift potential and deviation degree corresponding to each borehole are combined to obtain the drift contribution value of the corresponding borehole at the corresponding time. The calculation formula is as follows: In the formula, For the i-th aperture at the current observation time The delayed drift potential, For the i-th aperture at the current observation time The calculation formula, relative to the deviation from the nominal reference center, achieves coupled quantification of the cumulative risk of process disturbance and the actual observation deviation through the product of the late drift potential and the deviation degree. The late drift potential characterizes the potential trend of the aperture position being affected by the late drift that progresses from the edge inward, while the deviation degree characterizes the deviation magnitude of the actual observation results of the aperture position. The combination of the two can accurately distinguish between systematic deviation and random experimental error caused by process-related late drift. The higher the drift contribution value, the greater the contribution of the observation deviation of the aperture position to the influence of the late drift zone targeted by this invention, providing a core judgment index for the subsequent calculation of drift zone depth and cumulative progression.

[0021] It should be noted that the drift contribution value refers to the comprehensive quantitative result formed by the late-drift potential of a certain well site and the degree of deviation of the observed value of that well site relative to the nominal reference center at a corresponding time. It is used to characterize the strength of the contribution of the current observation deviation of the well site attributable to the influence of process-related late-drift. It does not reflect the potential disturbance trend of the well site alone, nor does it reflect the difference between the observed value and the same template reference alone. Instead, it combines the two to identify whether the current abnormal offset of a well site is more likely to originate from a late-drift band that progresses from the edge inward. The larger the drift contribution value, the stronger the cumulative disturbance trend of the well site is, and the more significant the deviation of its current observation result has been relative to the nominal reference of the same template well site. Therefore, the degree to which the current deviation of the well site can be explained by the influence of late-drift is higher. The smaller the drift contribution value, the more likely the well site has some disturbance accumulation, but has not yet formed a significant reference deviation, or although there is some deviation, its correlation with the late-drift trend is weak.

[0022] Hole positions were grouped and their drift contributions were calculated based on concentric circles with the same edge distance. The edge distance was used as the unique identifier for each circle, with holes of the same edge distance grouped into the same circle. Holes with an edge distance of 0 formed the outermost edge circle, and circles with increasing edge distances were arranged inwards. For each observation time, the drift contribution of all holes within the same circle was averaged to obtain the corresponding circle's drift contribution at that observation time. The calculation formula is as follows: In the formula, For the current observation time, This represents the edge distance value corresponding to the circle. The edge distance is The contribution of the sphere to the sphere drift at the current observation time. The edge distance is The total number of holes contained within the concentric rings. Let be the edge distance of the i-th hole position. Let i be the drift contribution value of the i-th aperture at the current observation time. To perform the summation calculation, this formula uses the average value of the drift contribution within the same sphere as the sphere drift contribution. This eliminates the influence of differences in the number of apertures in different spheres on the calculation results, ensures that the drift intensity of different spheres is comparable laterally, and accurately quantifies the average degree of influence of delayed drift on each sphere.

[0023] The drift contribution of each layer is accumulated from the outermost layer inwards to determine the drift zone depth at the current moment. Starting from the outermost layer, the cumulative average drift contribution from the outermost layer to the current layer is calculated sequentially in ascending order of edge distance. The edge distance of the layer that maximizes the cumulative average drift contribution is determined as the drift zone depth at the current observation moment. The calculation formula is as follows: In the formula, The depth of the drift zone at the current observation time. This represents the maximum edge distance value corresponding to the layers participating in the cumulative calculation. The edge distance values ​​corresponding to the layers participating in the cumulative calculation. The edge distance is The contribution of the sphere to the sphere drift at the current observation time. To obtain the value of the independent variable that maximizes the calculation result within the parentheses, this calculation formula determines the depth of the drift zone by averaging the results layer by layer from the edge inward. This can accurately pinpoint the inner boundary of the banded region affected by delayed drift, avoiding the underestimation of the drift zone range caused by determining the boundary solely based on the drift contribution of a single layer.

[0024] All observation times throughout the experiment are arranged in ascending order of their occurrence. The drift band depths of adjacent observation times are compared sequentially. When the drift band depth of a later observation time is greater than that of an earlier observation time, the difference between the two is recorded as the stage progression for that adjacent observation time. When the drift band depth of a later observation time is not greater than that of an earlier observation time, the stage progression is recorded as zero. This calculation rule only counts the positive change in the inward advancement of the drift band, excluding non-systematic influences such as drift band depth fluctuations or retractions, accurately reflecting the progressive characteristics of late-occurring drift bands. Finally, the stage progressions corresponding to all adjacent observation times are summed to obtain the cumulative progression up to the current observation time. The calculation formula is as follows: In the formula, This represents the cumulative increment up to the m-th observation time, where m is the total sequence number of the observation times arranged in ascending order of time. These are the sequence numbers of adjacent observation times. For the first At each observation time, For the first The previous adjacent observation time of each observation time For the first The drift zone depth corresponding to each observation time For the first The drift zone depth corresponding to each observation time To perform the maximum value operation, For summation calculation, this formula quantifies the total magnitude of the late-onset drift band advancing inward during the experimental period by accumulating the progressive amounts of each stage of the entire process. The larger the cumulative progressive amount, the more significant the phenomenon of the drift band expanding inward due to the accumulation of disturbance events during the experiment, providing core time-series status indicators for the generation of subsequent process governance actions.

[0025] It should be noted that the drift band depth refers to the innermost layer position at the corresponding observation time, where the cumulative contribution of drift from the outermost layer of the experimental plate reaches its maximum after statistical analysis layer by layer from the edge inwards. It characterizes the extent to which the delayed drift effect, progressing from the edge inwards, has extended into the interior of the experimental plate. This quantity is not a local anomaly marker for a single well location, but rather a spatial boundary characterization of the banded drift region formed on the entire experimental plate at the current moment. A larger drift band depth indicates that the delayed drift effect caused by edge disturbance has been further transmitted from the outermost layer to the innermost well locations, resulting in a wider affected area and a relatively smaller stable region within the experimental plate that can directly participate in the routine statistical analysis. A smaller drift band depth indicates that the current delayed drift effect is mainly concentrated in a limited number of layers near the edge and has not yet significantly extended into the deeper layers of the plate.

[0026] It should be noted that the cumulative progression refers to the total progression obtained by successively extracting and summing the changes in drift band depth at each observation time relative to the previous observation time across multiple chronologically arranged observation moments. This measure characterizes the overall magnitude of the late-onset drift band's continuous inward advancement throughout the experimental process. This quantity only counts the positive advancement corresponding to the increase in drift band depth, excluding changes occurring when the drift band depth remains constant or retracts. Therefore, it effectively reflects the true expansion trend of the late-onset drift effect within the experimental period. A larger cumulative progression indicates a more significant inward advancement of late-onset drift at the edges during the experiment, and a more pronounced continuous intrusion of process disturbances into the stability of the borehole positions within the plate. A smaller cumulative progression indicates limited expansion of the drift band between observation moments, and a relatively stable overall state of the experimental plate.

[0027] In an embodiment of the present invention, the drift band depth, cumulative increment, and average drift intensity obtained based on the drift contribution value are constructed as an environmental state and input into a reinforcement learning model. The reinforcement learning model outputs corresponding process governance actions, including: Based on the drift contribution values ​​of all holes within the experimental plate at the current observation time, the corresponding plate-level average drift intensity is determined using the following formula: In the formula, For the current observation time, The average drift intensity corresponds to the current observation time, and N is the total number of holes in the experimental plate. Let i be the drift contribution value of the i-th aperture at the current observation time. For summation, the calculation formula uses the arithmetic mean of the drift contribution values ​​of all well positions on the entire plate as the average drift intensity. This can quantify the global severity of the experimental plate's impact from delayed drift, eliminate the interference of differences in the number of well positions on the plate-level state determination, and provide a core quantitative indicator of the global drift level for the environmental state input of reinforcement learning. The current remaining process node identifier of the experimental plate is obtained. The current remaining process node identifier is the unique sequence identifier of the process node that has not been executed after the current observation time in the full experimental process node sequence pre-set in the experimental design stage. The process nodes include subsequent imaging acquisition nodes, liquid replacement operation nodes, and endpoint detection nodes. Each process node is assigned a unique continuous numerical identifier according to the execution order. The current remaining process node identifier is the numerical identifier corresponding to the first executed node among the unexecuted process nodes, which is used to accurately characterize the remaining execution progress of the experimental process and the type of subsequent operations to be executed. Next, the drift band depth, cumulative progression, average drift intensity, and current remaining process node identifier obtained above are used to construct a one-dimensional state vector as the environmental state input of the reinforcement learning model. The calculation formula for the state vector is as follows: In the formula, This is the environmental state vector corresponding to the current observation time. The depth of the drift zone at the current observation time. This represents the cumulative increment up to the current observation time. The average drift intensity at the current observation time. This state vector, which identifies the remaining process nodes, fully covers four core dimensions: the spatial boundary of the late drift band, the degree of temporal progression, the overall drift level at the board level, and the progress of the remaining experimental process. It represents the actual operating state of the current experimental process, provides complete environmental input for the sequential decision-making of reinforcement learning, and ensures that the output decision action matches the current experimental state.

[0028] A set of candidate actions covering the entire process of governance is pre-constructed. This set includes actions such as adding risk markers, adjusting inner-circle priority acquisition, initiating retesting, and freezing data archiving. The risk marker action involves continuing the original experimental procedure while adding drift risk markers to well data within the drift zone depth range. The inner-circle priority acquisition action adjusts the acquisition order of the next acquisition node to prioritize inner-circle wells before acquiring outer-circle wells, reducing the exposure time of inner-circle wells to disturbances caused by door opening and transfer. The retesting action initiates supplementary acquisition or retesting under the same conditions for wells with drift contribution values ​​exceeding a set threshold within the same template. The set threshold is not a fixed constant but is adaptively determined based on the distribution of drift contribution values ​​of wells within the same template at the current observation time. Specifically, for wells with the same nominal experimental template identifier, all drift contribution values ​​are extracted at the current observation time, and the median value of this group of drift contribution values ​​is taken as the retest trigger threshold corresponding to that template group at the current observation time. The value; when the drift contribution value of the target well is greater than the retest trigger threshold corresponding to its template group, the target well is determined to meet the conditions for initiating the retest action. The median value of the drift contribution value of the same template is used as the retest trigger threshold, which can keep the retest trigger rule consistent with the actual drift distribution within the same template, avoiding the problem of insufficient adaptability of fixed thresholds under different experimental templates, different observation indicators and different experimental stages; the data freezing and archiving action freezes the well data within the depth range of the drift band and transfers it to the risk archiving process, not including it in the regular experimental results statistics; the aforementioned constructed environmental state is input into the pre-trained reinforcement learning model, and the optimal process governance action is determined from the candidate action set according to the action value function and output. The reinforcement learning model uses the Q-learning algorithm to complete the iterative update of the action value function, which is suitable for the multi-step sequential decision-making scenario in the experimental process of this invention, and can adapt to the uncertainty of the disturbance effect in cell research and development experiments, ensuring that the decision action can be optimized for long-term benefits, and taking into account the experimental data quality and process execution efficiency.

[0029] The model was pre-trained using a historical cell research multi-well plate experimental dataset. The pre-training dataset used was a publicly available long-term experimental dataset for cell research multi-well plates. This dataset included well coordinates, records of all perturbation events, quantitative observations of well positions at each observation point, and the reliability annotations of well positions from manual quality control at the experimental endpoint. The dataset contained complete experimental data from at least 100 standard multi-well plates. The rules for state vector construction, action value function updates, and immediate reward calculation during pre-training were consistent with those used in the actual online experimental phase. The convergence termination condition for pre-training was a low average change in the action value function over 100 consecutive iterations. At 0.001, the action decision accuracy of the converged model is no less than 90%, and the model parameters after pre-training can be directly used for online inference in actual experiments; furthermore, the construction method, sample organization method, and manual quality control annotation rules of the historical cell research and development multi-well plate experimental dataset are limited as follows: the historical cell research and development multi-well plate experimental dataset is constructed with experimental plates as basic sample units, and each experimental plate contains at least experimental plate specification information, well position coordinate information, full-process disturbance event records, well position observation values ​​corresponding to each observation time, drift zone depth and cumulative progression calculation results corresponding to each observation time, and... The experiment endpoint yielded reliable labeling results for each well location after manual quality control. For each observation time on each experimental board, the drift band depth, cumulative progression, average drift intensity, and current remaining process node identifier were extracted to form an environmental state sample. Combined with the actual governance actions taken in the historical manual process governance records under this environmental state sample, a corresponding state-action sample was formed. Then, based on the change in the state quality function at the next observation time after the governance action and the additional process burden introduced by the action, a corresponding immediate reward was generated, thus forming a state-action-reward system for pre-training. The next state quadruple sample; the well location credibility labeling results are at least divided into three categories: credibility labeling, retest labeling, and risk archiving labeling. Among them, credibility labeling indicates that well location data can be directly included in the statistics of routine experimental results, retest labeling indicates that well location data needs to be re-sampled or retested under the same conditions before judgment, and risk archiving labeling indicates that well location data is only retained for process traceability and drift review and is not included in the statistics of experimental conclusions. During pre-training, the action value function is updated in batches according to the time sequence of the quadruple sample, and the average change of the action value function after several consecutive iterations is lower than the convergence threshold as the pre-training convergence condition.The iteration update step size during the online operation phase is consistent with that during the pre-training phase. During the actual experimental execution, the model can continuously perform online iterative optimization based on the real-time state data of this experiment to ensure that the action decision adapts to the actual drift situation of this experiment. Finally, the model calculates the action value function value corresponding to each action in the candidate action set under the current environmental state, selects the action with the largest action value function value as the optimal process governance action and outputs it, providing a clear execution basis for subsequent data routing and archiving management.

[0030] It should be noted that the risk labeling action refers to a process governance action that, without changing the established execution order of subsequent experimental procedures on the current experimental board, adds a risk identification label to the well data currently determined to be within the drift band or with a high drift contribution value. The purpose of this action is to preserve the original experimental data and original process arrangement, while simultaneously tracking the relevant well data separately from regular stable well data during subsequent data analysis, review, and archiving. The risk labeling action does not directly trigger additional experimental operations; rather, it identifies the potential impact of delayed drift at the data level, preventing the data of affected wells from being treated as regular valid data in subsequent statistical analysis.

[0031] It should be noted that adjusting the inner-circle priority acquisition action refers to a process management action that, when imaging acquisition, detection acquisition, or other observation acquisition nodes still need to be performed subsequently, changes the acquisition order of the boreholes, which was originally performed in the conventional order, to prioritize the acquisition of the inner-circle boreholes of the experimental plate, followed by the acquisition of the outer-circle boreholes. The purpose of this action is to reduce the extra waiting time of the inner-circle boreholes under conditions of door opening, transportation, or imaging exposure, allowing the relatively stable inner-circle boreholes to complete data acquisition first, thereby reducing the new delayed drift impact on the inner-circle boreholes caused by continued diffusion from subsequent process exposure. This action is essentially a rearrangement of the subsequent acquisition order, rather than a change in the experimental processing conditions themselves.

[0032] It should be noted that initiating a retest action refers to a process remediation action taken for well sites that are currently deemed to have a high risk of delayed drift and whose observation results have deviated significantly from the nominal reference center of the same template. This involves initiating supplementary sampling, repeated observation, re-imaging, or repeated detection under the same conditions. The purpose of this action is to determine, through one or more additional data acquisitions under the same conditions, whether the current abnormal results of the target well site are caused by process-related delayed drift, or by instantaneous random errors, one-time acquisition anomalies, or occasional operational deviations. Initiating a retest action typically introduces additional experimental steps and additional data recording; therefore, it is suitable for well sites where there are substantial doubts about the reliability of the current data and further verification is required.

[0033] It should be noted that freezing data archiving refers to the process governance action of separating well site data that is currently determined to be significantly affected by the late-drift band and is no longer suitable for inclusion in the routine experimental results statistics from the regular analysis process and transferring it to the risk archiving path. The purpose of this action is to retain the complete original records, drift status information, and process context information of the relevant well sites for subsequent experimental retrospectives, drift debriefing, and scheme optimization, rather than using this data as a regular valid sample in the generation of current experimental conclusions. Freezing data archiving does not mean deleting data, but rather changing its archiving attributes and scope of use to prevent data significantly affected by late-drift from continuously interfering with experimental conclusions.

[0034] The action value function of the reinforcement learning model is iteratively updated to provide a stable value judgment benchmark for the optimization decision of process governance actions. First, based on the average drift intensity at the current observation time and combined with the ratio of the drift band depth to the maximum hole edge distance, the state quality function value corresponding to the current observation time is determined. The state quality function is used to quantify the reliability of the experimental data within the experimental board and the global impact level of late drift. The calculation formula is as follows: In the formula, Let m be the state-mass function value at the m-th observation time. Let m be the average drift intensity corresponding to the m-th observation time. Let m be the drift zone depth corresponding to the m-th observation time. The formula represents the maximum edge distance of all wells within the experimental board. It achieves a comprehensive quantification of state quality through the product of two core components. The first component uses the reciprocal of the average drift intensity plus one, quantifying that a higher average drift intensity corresponds to a lower state quality function value, reflecting the negative impact of overall board-level drift on data quality. The second component uses a normalization ratio minus the drift band depth. Normalization is first performed using the ratio of the drift band depth to the maximum edge distance of the well plus one, reducing the interference of size differences in multi-well boards of different specifications on state determination. This quantifies that a greater drift band depth corresponds to a lower state quality function value, reflecting the negative impact of the inward expansion of the late-developing drift band on data quality. The combination of these two components covers the two core influence dimensions of global drift intensity and drift band spatial range, ensuring that the state quality function value truly reflects the overall reliability of the experimental data.

[0035] Based on the change between the state quality function value at the next observation time and the state quality function value at the current observation time, and combined with the additional process burden introduced by executing the corresponding process governance action, the immediate reward for executing the action is determined. The immediate reward is used to quantify the benefits and costs generated after executing the process governance action, providing a core optimization objective for updating the action value function. The calculation formula is as follows: In the formula, To execute process governance actions at the m-th observation time The corresponding instant rewards, To implement process governance actions The state-mass function value corresponding to the (m+1)th observation time. Let m be the state-mass function value at the m-th observation time. To implement process governance actions The introduced additional process burden is pre-defined by assigning a fixed additional process burden value to each process governance action within the candidate action set. Actions without additional experimental operations or process modifications have an additional process burden of zero. The more additional operations introduced and the longer the process time, the higher the corresponding additional process burden value. The risk labeling action has no additional process operations and has an additional process burden of zero. The inner circle priority collection action only changes the collection order and has no additional experimental operations, so its additional process burden is 0.1. The retesting action requires additional experimental operations and data collection, so its additional process burden is 0.5. The data freezing and archiving action requires additional data diversion and review processes, so its additional process burden is... The additional process burden is set to 0.2, based on the proportion of additional experimental time and human resource costs introduced by the action. The value ranges from 0 to 1. Actions without additional process operations are set to 0, and the value increases as the proportion of additional operations increases. Technical personnel in the relevant field can adjust the values ​​of the above parameters within a reasonable range according to the process complexity and cost requirements of the actual experiment. The calculation formula takes the change in the state quality function value before and after the execution of the action as the core benefit item and the additional process burden as the cost deduction item, so that the immediate reward can simultaneously reflect the improvement effect of the action on data quality and the additional process cost brought about, guiding the model to prioritize governance actions that can significantly improve data quality without introducing excessive process burden.

[0036] Based on the aforementioned immediate rewards and the preset discount factor and learning rate, the action value function corresponding to the current environmental state and process governance action combination is iteratively updated. The action value function is used to quantify the long-term cumulative benefits that can be obtained by performing a specific process governance action under a specific environmental state. The iteration is completed using a temporal difference update method, and the update formula is: In the formula, The environmental state at the m-th observation time Next execution process governance actions The corresponding action value function value, The preset learning rate is 0.2. A reasonable range for the learning rate is 0.1 to 0.3. A larger value results in a faster response to new real-time data, while a smaller value leads to stronger stability after convergence. This embodiment selects 0.2 as the default value to balance the model's response speed and stability. The preset discount factor is 0.8. A reasonable range for the discount factor is 0.7 to 0.9. A larger value indicates a greater focus on long-term returns, while a smaller value indicates a greater focus on immediate returns. This embodiment selects 0.8 as the default value to suit the sequential decision-making requirements of long-term cell experiments. To implement process governance actions The corresponding instant rewards, For a pre-defined set of candidate actions, To implement process governance actions The environmental state obtained at the (m+1)th observation time after the transition. Given the environmental state at the (m+1)th observation time, let be the maximum value of the action value function corresponding to all actions in the candidate action set. The iterative update symbol represents assigning the calculation result on the right side of the arrow to the action value function on the left side of the arrow, completing one iterative update. This update formula combines the immediate reward of the current action with the maximum expected future reward of the next state, and uses a learning rate weighted approach to complete the incremental update. During the model pre-training phase, the historical cell research and development multi-well plate experimental dataset is used to complete batch iterative updates until the action value function converges to a stable range. During the actual online experimental phase, after completing the environmental state acquisition, process governance action execution, and environmental state acquisition at each observation time, a single-step iterative update of the action value function is performed to achieve continuous optimization and adaptive adjustment of the model.

[0037] In an embodiment of the present invention, a reliability coefficient for each of the aperture locations is determined based on the drift contribution value; aperture location data routing is completed based on the reliability coefficient and the process governance action; and a management ledger is generated, including: Based on the drift contribution value of each well site at all historical observation times, the reliability coefficient of each well site is determined. The reliability coefficient is used to quantify the degree of reliability of the well site observation data affected by late drift. The calculation formula is as follows: In the formula, Let be the confidence coefficient for the i-th well position, and M be the total number of observation times up to the current stage of the experiment. The drift contribution value of the i-th well site at the m-th observation time is calculated using the reciprocal of the cumulative sum of drift contributions at all historical observation times plus one. This constrains the range of the confidence coefficient to between 0 and 1. The closer the confidence coefficient is to 1, the smaller the cumulative impact of late drift on the well site, and the higher the confidence level of the observation data. The closer the confidence coefficient is to 0, the greater the cumulative impact of late drift on the well site, and the lower the confidence level of the observation data. This also avoids meaningless infinite results when the cumulative drift contribution value is 0, matching the quantitative requirements of the confidence level of well site data and providing a standardized confidence weight benchmark for subsequent data routing and result statistics.

[0038] It should be noted that the credibility coefficient is a quantitative indicator used to characterize the overall credibility of the observation data of a well site, determined based on the cumulative drift contribution value at each historical observation time. This quantity does not reflect the instantaneous anomalous state at a single moment, but rather the comprehensive credibility level of the well site after being continuously affected by the progressively inward late drift throughout the entire experimental process. A higher credibility coefficient indicates a smaller cumulative drift contribution from the well site during historical observations, and a lower degree of interference from process-related late drift, making it more suitable as an effective data source for routine experimental conclusion analysis and template-based result summarization. Conversely, a lower credibility coefficient indicates a higher cumulative drift contribution from the well site across multiple observation times, and its data is more likely to be continuously affected by process-related disturbances, thus requiring priority for review, risk labeling, or risk archiving.

[0039] All wells in the experimental board are sorted in ascending order according to their confidence coefficients. Based on the sorting results and the output process management actions, a set of low-confidence wells is determined. During the sorting process, wells with smaller confidence coefficients are ranked higher. The rules for determining the set of low-confidence wells correspond to the types of process management actions. Furthermore, to ensure that the determination results of the set of low-confidence wells have consistent executability under different experimental board specifications and different observation nodes, this embodiment stipulates that the selection of the set of low-confidence wells is based solely on the drift zone depth, confidence coefficient, and statistical results of the same template that have been calculated at the current observation time. According to the process governance action, when the retest action is initiated, in addition to including holes with a confidence coefficient lower than the median confidence coefficient of the corresponding hole in the same template into the low confidence hole set, the drift contribution value of the hole at the observation time is also checked for consistency to see if it is higher than the median drift contribution value of the same template. Only when both conditions are met is the hole finally determined as the hole to be retested. By using the dual consistency judgment of confidence coefficient and drift contribution value, misjudgment caused by a single moment anomaly or a single cumulative indicator can be avoided, ensuring a more stable correspondence between the low confidence hole set and the process governance action.

[0040] When the process governance action is to add risk markers, all holes within the edge distance range corresponding to the drift zone depth are included in the low-confidence hole set. When the process governance action is to adjust the inner circle priority collection action, all holes with an edge distance less than or equal to the drift zone depth are included in the low-confidence hole set. When the process governance action is to initiate a retest, holes with a confidence coefficient lower than the median confidence coefficient of the corresponding hole in the same template are included in the low-confidence hole set. When the process governance action is to freeze data archiving, all holes within the edge distance range corresponding to the drift zone depth are included in the low-confidence hole set. This rule matches the control objectives of different process governance actions. To ensure that the screening of low-confidence wells corresponds to the current experimental drift status and governance needs, all well data corresponding to the low-confidence well set are written into the review queue. The data in the review queue are set with review priority in ascending order of confidence coefficient, with the lower the confidence coefficient, the higher the review priority. At the same time, the well data corresponding to the remaining wells outside the low-confidence well set are written into the regular archiving queue. The data in the regular archiving queue directly enters the standardized experimental data archiving process and is included in the regular experimental results statistics. According to the output process governance actions, the corresponding well data that meet the risk archiving conditions are adjusted and written into the risk archiving path to complete the well data routing processing.

[0041] The risk archiving condition is set to "freeze data archiving" as the process governance action. When the process governance action is "freeze data archiving," all well data within the low-confidence well set meets the risk archiving condition. The corresponding well data is then moved from the review queue or regular archiving queue and written into the risk archiving path. Well data in the review queue needs to undergo manual review and experimental verification. Data that passes the review is removed from the risk flag and transferred to the regular archiving queue; data that fails the review is transferred to the risk archiving path. The manual review and experimental verification include at least the following criteria: First, checking whether the disturbance event chain corresponding to the target well is complete and whether there are missing events, incorrect time sequence, or missing recorded liquid volume; Second, checking the original observation value of the target well at the corresponding observation time. The criteria for determining whether a target well location passes the review are: 1) the difference between the target well location and the repeated observations; 2) the difference between the target well location and the repeated observations within the same template group; 3) whether the target well location is within the current drift zone depth coverage area and whether its confidence coefficient is lower than the median confidence coefficient of its same template group; 4) when all three criteria meet the normal conditions, the target well location is deemed to have passed the review, its risk label is removed, and it is transferred to the regular archiving queue; 5) when any one of the criteria fails the review, the target well location is deemed to have failed the review, and it is transferred to the risk archiving path; 6) adopting the above review criteria provides a clear basis for determining whether a review passes or fails, avoiding reliance on subjective experience for manual review, thereby improving the repeatability and feasibility of data routing processing.Data within the risk archiving path is stored separately in a closed loop, synchronously annotated with drift risk labels, and the original experimental records are permanently retained. This data is used solely for experimental process tracing, drift impact review, and subsequent experimental scheme optimization, and is not included in any statistical analysis of experimental conclusions. Data in the regular archiving queue is synchronously associated with the corresponding well position's reliability coefficient, used for calculating the weighted summary results of the same template and analyzing the core experimental conclusions. This ensures that data routing matches the decision requirements of process governance actions, achieving the diversion and control of data with different reliability levels. Finally, the edge distance, late drift potential, drift contribution value, drift band depth, cumulative progression, reliability coefficient, and output process governance actions of each well position are assembled into a standardized target data structure, generating a management ledger that completely records the entire process data status of the experimental board. The target data structure adopts a structured storage format that corresponds one-to-one with the unique identifier of each well position. The data entries corresponding to the well locations fully include all the aforementioned core indicators. The edge distance is derived from the calculation results of the well location coordinates and experimental board specifications. The late drift potential is derived from the calculation results of the well location corresponding to each observation time. The drift contribution value is derived from the calculation results of the well location corresponding to each observation time. The drift band depth is derived from the board-level calculation results at each observation time. The cumulative progressive amount is derived from the board-level time series calculation results up to the current stage. The reliability coefficient is derived from the calculation results of the well locations corresponding to all historical observation times. The process governance actions are derived from the decision results output by the reinforcement learning model for the current experimental state. The management ledger fully covers the entire chain of information on drift identification, state determination, decision actions, and data reliability classification throughout the entire experimental process. This approach does not compromise the integrity of the original experimental data and provides a complete structured management basis for experimental data quality control, process traceability, result review, and subsequent model training.

[0042] After completing the calculation of the reliability coefficients of each well in the experimental plate and constructing the basic data for the management ledger, the weighted summarization and archiving of the observations of wells with the same template were completed. All wells in the experimental plate were grouped according to the nominal experimental template identifier. Wells with the same nominal experimental template identifier were grouped into the same template well group. The nominal experimental template identifier is a unique group identifier pre-assigned to each well during the experimental design stage. The nominal experimental conditions of the same template wells are consistent, including cell seeding density, culture medium composition, drug treatment conditions, culture time, and observation node settings, to ensure that the observations of the same template wells have a unified benchmark for statistical summarization. Using the reliability coefficients of each well in the same template group as weights, the observations of each well in the same template group at the corresponding observation time were weighted and summed to generate the same template weighted summarization result for that group. The calculation formula is as follows: In the formula, For the target observation time, The nominal experimental template identifier corresponding to the target template group. This is the nominal experimental template identifier for the i-th well position. Let be the confidence coefficient for the i-th hole position. For the i-th aperture at the target observation time The calculation formula uses a confidence coefficient-weighted arithmetic average method, with the confidence coefficient of the well as the basis for weight allocation. This ensures that well observations with higher confidence levels account for a higher proportion in the summary results, while well observations with lower confidence levels and greater impact from late drift accumulation account for a lower proportion. This avoids the misleading influence of low-confidence well observations affected by process drift on the experimental conclusions of the same template in the traditional equal-weighted average method. At the same time, it does not directly remove low-confidence well data, thus fully retaining the information of all experimental samples and avoiding the invalid loss of effective experimental data. The weighted summarization calculation method can ensure that the final summary results can truly reflect the real experimental effects of the same template wells under drift-free interference, providing a quantitative benchmark for drug efficacy determination, cell state assessment, and batch difference analysis in cell research and development experiments.

[0043] After calculating the weighted summaries of all groups within the same template at each observation time throughout the entire experimental process, the weighted summaries of all observation times for each nominal experimental template identifier are associated and bound with the identifier list, confidence coefficient distribution, and original observation data of the corresponding wells within the same template. This data is then written into the management ledger of the corresponding experimental board for archiving. An independent entry for summaries is created in the management ledger for each nominal experimental template identifier. This entry fully records the weighted summaries of all observation times for the corresponding template, the basis for weight calculation, the total number of wells within the same template, and the confidence coefficient statistics. This ensures the end-to-end traceability and reproducibility of the summaries. The archived weighted summaries of the same template can be directly used for core conclusion analysis scenarios in cell research and development. Simultaneously, the management ledger retains the original observations and equally weighted average results of the wells within the same template, providing complete control data support for cross-validation of experimental results and review of the effects of late drift.

[0044] like Figure 2 The diagram shown is a functional block diagram of an experimental workflow data management system for cell research and development provided in an embodiment of the present invention.

[0045] In this embodiment, the functions of each module / unit are as follows: The perturbation mapping module is used to obtain the edge distance of each hole in the experimental board and map the perturbation events corresponding to each hole to the time axis to generate the perturbation event chain of the corresponding hole. The drift calculation module is used to extract the late drift potential accumulated over time for each of the well sites based on the edge distance and the duration and volume of the recording liquid contained in the disturbance event chain. The state assessment module is used to determine the drift contribution value of the corresponding hole position based on the delayed drift potential and the observed values ​​of each hole position, and to extract the drift band depth and cumulative progression in the experimental plate according to the correspondence between the edge distance and the drift contribution value. The intelligent decision-making module is used to construct the environmental state by taking the drift zone depth, cumulative increment, and average drift intensity obtained based on the drift contribution value and inputting them into the reinforcement learning model, and outputting the corresponding process governance actions through the reinforcement learning model. The data ledger module is used to determine the reliability coefficient of each of the holes based on the drift contribution value, complete the hole data routing process according to the reliability coefficient and the process governance action, and generate a management ledger.

[0046] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A data management method for experimental workflows in cell research and development, characterized in that, include: S1. Obtain the edge distance of each hole in the experimental board, and map the disturbance event corresponding to each hole to the time axis to generate the disturbance event chain of the corresponding hole. S2. Based on the edge distance and the duration and volume of recording liquid contained in the disturbance event chain, extract the late drift potential accumulated over time for each of the well sites. The specific steps for extracting the late drift potential accumulated over time at each of the aforementioned pore sites include: The volume sensitivity corresponding to a single disturbance event is determined based on the duration of the disturbance event and the corresponding recorded volume. The edge response level corresponding to the hole position is determined based on the edge distance of the hole position; Based on the time difference between the current time and the time when the corresponding disturbance event occurred, and in conjunction with the edge distance, the degree of delayed spread corresponding to a single disturbance event is determined; The single delayed contribution of a single disturbance event to the aperture is determined based on the volume sensitivity, the edge response degree, and the delayed propagation degree. The delayed contribution of each individual event corresponding to all the disturbance events in the disturbance event chain is accumulated to obtain the delayed drift potential of the aperture at the current moment. S3. Determine the drift contribution value of the corresponding hole position based on the delayed drift potential and the observed value of each hole position, and extract the drift band depth and cumulative progression in the experimental plate according to the correspondence between the edge distance and the drift contribution value. The specific steps for extracting the drift band depth and cumulative increment within the experimental plate include: According to the concentric circles with the same edge distance, all the drift contribution values ​​within the same concentric circle are summarized to obtain the concentric circle drift contribution of the corresponding concentric circle. The drift contribution of each layer is accumulated layer by layer from the outermost layer to the innermost layer, and the layer that maximizes the accumulation result is determined as the drift zone depth at the current moment. Compare the drift zone depths corresponding to two adjacent observation times; When the drift zone depth at the later observation time is greater than the drift zone depth at the earlier observation time, the difference between the two is recorded as the stage progression amount; when the drift zone depth at the later observation time is not greater than the drift zone depth at the earlier observation time, the stage progression amount is recorded as zero. The cumulative increment is obtained by summing the stage increments corresponding to all adjacent observation times. S4. Construct the environment state by the drift zone depth, cumulative increment, and average drift intensity obtained based on the drift contribution value, and input it into the reinforcement learning model. Output the corresponding process governance action through the reinforcement learning model. S5. Determine the reliability coefficient of each of the holes based on the drift contribution value, complete the hole data routing process according to the reliability coefficient and the process governance action, and generate a management ledger.

2. The experimental workflow data management method for cell research and development according to claim 1, characterized in that, Obtain the edge distance of each hole position within the experimental board, and map the perturbation events corresponding to each hole position to the time axis to generate a perturbation event chain for the corresponding hole position, including: Obtain the row and column coordinates of each hole location in the experimental plate; The edge distance of each hole is determined based on the distances from the row coordinates and column coordinates to the corresponding boundaries of the experimental board. The sample addition event, door opening event, transfer event, imaging event, and liquid change event are treated as disturbance events. The occurrence time, duration, and corresponding recording volume of each disturbance event are recorded in chronological order. The disturbance events corresponding to each of the holes are assembled in chronological order to generate a disturbance event chain for the corresponding hole.

3. The experimental workflow data management method for cell research and development according to claim 1, characterized in that, The drift contribution value of each aperture is determined based on the delayed drift potential and the observed values ​​of each aperture, including: Obtain the nominal experimental template identifier to which each of the aforementioned apertures belongs; Among the same template wells with the same nominal experimental template identifier, the median value of the current observation is extracted as the nominal reference center; Calculate the degree of deviation between the observed value at each aperture position and the corresponding nominal reference center; By combining the delayed drift potential corresponding to each of the aforementioned aperture positions with the degree of deviation, the drift contribution value of the corresponding aperture position at the corresponding time is obtained.

4. The experimental workflow data management method for cell research and development according to claim 1, characterized in that, The drift band depth, cumulative increment, and average drift intensity obtained based on the drift contribution value are used to construct the environment state and input into the reinforcement learning model, including: Based on the drift contribution values ​​of all holes in the experimental plate at the current observation time, determine the corresponding average drift intensity; Obtain the current remaining process node identifier of the experimental board; The drift zone depth, the cumulative increment, the average drift intensity, and the current remaining process node identifier are constructed into a state vector as the environment state. The environmental state is input into the reinforcement learning model, and the process governance action is determined from the candidate action set according to the action value function and output. The candidate action set includes actions such as adding risk markers, adjusting the priority collection of the inner circle, initiating retesting, and freezing data archiving.

5. The experimental workflow data management method for cell research and development according to claim 4, characterized in that, The action value function of the reinforcement learning model is updated in the following way: Based on the average drift intensity and the ratio of the drift zone depth to the maximum aperture edge distance, the state quality function value corresponding to the current observation time is determined. The immediate reward is determined based on the change between the state quality function value at the next observation time and the state quality function value at the current observation time, and in combination with the additional process burden introduced by the process governance action. Based on the instant reward, discount factor, and learning rate, the action value function corresponding to the current environmental state and process governance action combination is iteratively updated.

6. The experimental workflow data management method for cell research and development according to claim 1, characterized in that, Based on the drift contribution value, a reliability coefficient is determined for each of the aperture locations. Based on the reliability coefficient and the process governance actions, aperture location data routing is completed, and a management ledger is generated, including: The confidence coefficient of each of the aforementioned apertures is determined based on the drift contribution value corresponding to each aperture at the historical observation time. The holes are sorted according to the confidence coefficient, and the set of low-confidence holes is determined based on the sorting results and the process governance actions. Write the hole data corresponding to the low-confidence hole set into the review queue, and write the hole data corresponding to the remaining holes into the regular archive queue. According to the process governance actions, the corresponding hole location data that meets the risk archiving conditions is adjusted and written into the risk archiving path to complete the hole location data routing process. The edge distance, late drift potential, drift contribution value, drift band depth, cumulative increment, confidence coefficient, and process governance action of each hole are assembled into a target data structure to generate a management ledger that records the data status of the experimental board.

7. The experimental workflow data management method for cell research and development according to claim 6, characterized in that, After determining the reliability coefficient of each of the aforementioned aperture locations, the process further includes: For well sites with the same nominal experimental template identifier, the observations of each well site are weighted and summed using the confidence coefficient of the corresponding well site as the weight, to generate a weighted summation result of the same template. Write the weighted summary results of the same template into the corresponding management ledger for archiving.

8. A data management system for experimental workflows in cell research and development, applied in the data management method for experimental workflows in cell research and development as described in any one of claims 1-7, characterized in that, The system includes: The perturbation mapping module is used to obtain the edge distance of each hole in the experimental board and map the perturbation events corresponding to each hole to the time axis to generate the perturbation event chain of the corresponding hole. The drift calculation module is used to extract the late drift potential accumulated over time for each of the well sites based on the edge distance and the duration and volume of the recording liquid contained in the disturbance event chain. The state assessment module is used to determine the drift contribution value of the corresponding hole position based on the delayed drift potential and the observed values ​​of each hole position, and to extract the drift band depth and cumulative progression in the experimental plate according to the correspondence between the edge distance and the drift contribution value. The intelligent decision-making module is used to construct the environmental state by taking the drift zone depth, cumulative increment, and average drift intensity obtained based on the drift contribution value and inputting it into the reinforcement learning model, and output the corresponding process governance action through the reinforcement learning model. The data ledger module is used to determine the reliability coefficient of each of the holes based on the drift contribution value, complete the hole data routing process according to the reliability coefficient and the process governance action, and generate a management ledger.

Images