How to calculate process parameters

Abstract

To provide a method for calculating process parameters to achieve a given process result.SOLUTION: A method for calculating process parameters includes the following steps of: providing a trained prediction model, obtained by machine learning of a dataset by a machine learning method, the dataset including a plurality of samples, each of the samples including a plurality of sample parameters, and the trained prediction model being used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters; setting a predicted result of the trained prediction model and providing at least one confirmed input parameter of the input parameters; and comparing the expected result, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset by a reverse derivation algorithm to determine at least one non-confirmed input parameter of the input parameters.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method for calculating process parameters, and more particularly to a method for calculating process parameters that can be applied to cell culture processes to obtain corresponding process steps based on desired process effects. [Background technology]

[0002] With advances in medical engineering technology, the application of regenerative medicine in clinical disease treatment is becoming increasingly diverse. Regenerative medicine is a medical technology that primarily utilizes the regenerative abilities of cells to repair damaged tissues and organs, and its range of applications is extremely broad. Furthermore, by combining medical technologies such as tissue engineering and molecular biology, it is expected that diseases that were previously considered difficult to treat, such as diabetes, neurological disorders, cardiovascular diseases, and cancer, can be improved and cured. Currently, regenerative medicine is primarily applied to organ repair, immune cell therapy, and stem cell therapy. Furthermore, research and applications of cell therapy in regenerative medicine are attracting increasing attention from various fields. Cell therapy involves culturing or processing human cells ex vivo, followed by transplantation into an individual's body.

[0003] In recent years, as governments in many countries have gradually lifted restrictions on the application of cell therapy, many scholars both at home and abroad have begun to engage in cell therapy research. As a result, significant progress has been made in cell therapy research to treat many diseases, such as the treatment of skin defects with autologous fibroblasts, the treatment of cartilage defects in knee joints with autologous chondrocytes, and the treatment of spinal cord injuries with autologous bone marrow mesenchymal stem cells. Previous research has shown that each individual's cells have unique characteristics, and therefore the type of cell preparation applied to each case varies depending on the disease and symptoms. Therefore, the need to customize cells appropriate for each individual's condition increases the complexity and difficulty of the cell preparation process. Furthermore, the quality of cell therapy products directly affects the safety and efficacy of the treatment. Therefore, cell culture processes require strict control of cell growth conditions and real-time monitoring of cell growth and environmental parameters to avoid cell contamination or quality deterioration during the culture process. However, due to the high variability of cells between different cases, the optimal culture and environmental parameters for cell preparations applied to different cases are not completely identical. Furthermore, because each process parameter must be adjusted within the process to achieve the desired results for each cell preparation, it is not possible to produce each cell preparation using fixed process parameters. In conventional techniques, a predictive model is obtained by training on a large set of sample data using machine learning. This predictive model can generate predicted cell culture results by inputting various process parameters, allowing users to simulate the effects of the cell culture process they are designing in advance. However, the above-mentioned predictive model trained using machine learning cannot reverse-estimate the process parameters of the cell culture process from the user's desired results. In other words, when designing a cell culture process, users must try a large number of different process parameters, which requires a great deal of effort and resources to design and improve the cell culture process.

[0004] Therefore, it is necessary to develop a method to further optimize cellular processes by enabling inverse estimation of process parameters that match the desired results. Summary of the Invention [Problem to be solved by the invention]

[0005] In view of the above, the present invention provides a method for calculating process parameters to solve the above-mentioned conventional problems. [Means for solving the problem]

[0006] The present invention provides a trained predictive model, which is obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, each sample including a plurality of sample parameters, and the trained predictive model is used to input the plurality of input parameters and generate a predicted result corresponding to the input parameters; setting a predicted result of the trained predictive model as a target result and providing at least one confirmed input parameter among the input parameters; and using an inverse inference algorithm to match the target result, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset to identify at least one unconfirmed input parameter among the input parameters.

[0007] The data set further includes a process data set, the process data set including a plurality of process samples, each process sample including a plurality of process sample parameters, each process sample parameter including a source parameter and an incubation parameter.

[0008] The source parameters for each process sample further include attribute data for the source of each process sample.

[0009] The incubation parameters for each process sample further include the parameters of the people, equipment, materials, methods, and environment in each process sample.

[0010] Each sample parameter includes at least one corresponding sample parameter and at least one unconfirmed reference parameter. The step of identifying at least one unconfirmed input parameter among the input parameters by using an inverse inference algorithm to match the target outcome, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset further includes: calculating a first vector of the at least one confirmed input parameter; obtaining at least one corresponding corresponding sample parameter from the sample parameters of the samples based on the at least one confirmed input parameter; calculating a second vector of the at least one corresponding sample parameter; matching the first vector with the second vector of each sample to select at least one candidate sample; inputting a combination of the at least one unconfirmed reference parameter and the confirmed input parameters of the at least one candidate sample into the trained prediction model to obtain a reference sample from the sample, wherein the second vector of the reference sample is similar to or consistent with the first vector, and the predicted outcome of the reference sample is similar to or consistent with the target outcome; and determining the at least one unconfirmed reference parameter in the sample parameters of the reference sample as the at least one unconfirmed input parameter.

[0011] In the process parameter calculation method of the present invention, the step of matching the first vector with the second vector of each sample to select at least one candidate sample further includes the step of calculating an included angle between the first vector and the second vector of each sample, and selecting the sample corresponding to the second vector that has the smallest included angle with the first vector as the candidate sample.

[0012] In the process parameter calculation method of the present invention, the step of matching the first vector with the second vector of each sample to select at least one candidate sample further includes the step of calculating the distance between the coordinates of the end point of the first vector and the coordinates of the end point of the second vector of each sample using a distance function, and selecting the sample corresponding to the second vector that has the smallest distance from the first vector as the candidate sample.

[0013] The trained predictive models include Artificial Neural Networks (ANN), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Decision Trees, Support Vector Machines (SVM), Random Forests, K-Nearest Neighbors (KNN), K-Means Clustering, Principal Component Analysis (PCA), Linear Regression, Logistic Regression, Gradient Boosting Machines, Deep Belief Networks (DBN), Recursive Neural Networks (RecNN), Reinforcement Learning, Autoencoders, and Gaussian Processes. It is obtained by machine learning a dataset using either Neural Networks (NNs) or Complex Neural Networks (CNNs).

[0014] In the process parameter calculation method of the present invention, the step of using an inverse inference algorithm to match the target result, at least one confirmed input parameter, and sample parameters of the samples in the dataset to identify at least one unconfirmed input parameter among the input parameters further includes the steps of: using an inverse inference algorithm to match the target result, the at least one confirmed input parameter, and sample parameters of the samples in the dataset to identify sample parameters of a first reference sample among the samples as reference input parameters, wherein the reference input parameters include at least one unconfirmed reference parameter; inputting the at least one confirmed input parameter and the at least one unconfirmed reference parameter into a trained predictive model to generate a reference prediction result; matching the reference prediction result with the target result; and, if the reference prediction result matches the target result, identifying the at least one unconfirmed reference parameter as the at least one unconfirmed input parameter.

[0015] The process parameter calculation method of the present invention further includes the step of excluding the first reference sample if the reference prediction result does not match the target result, and again using the inverse inference algorithm to compare the target result, at least one confirmed input parameter and the sample parameters of the samples in the dataset, to obtain the sample parameters of the second reference sample among the samples as reference input parameters.

[0016] The process parameter calculation method of the present invention further includes the steps of: matching at least one corresponding sample parameter of the samples in the dataset, and inputting the at least one confirmed input parameter and at least one unconfirmed reference parameter corresponding to each sample into the trained prediction model to generate corresponding reference prediction results; and matching the reference prediction results corresponding to each sample with the target results to find reference samples whose reference prediction results match the target results, and further forming a reference sample set.

[0017] The process parameter calculation method of the present invention further includes the step of linearly combining the first K data in the reference sample set and finding the optimal combination of unconfirmed input parameters by an optimization search method.

[0018] The present invention provides a trained predictive model, which is obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, each sample including a plurality of sample parameters. The trained predictive model is used to input the plurality of input parameters and generate a predicted result corresponding to the input parameters; setting a predicted result of the trained predictive model as a target result, and providing at least one confirmed input parameter among the input parameters, wherein each sample parameter includes at least one corresponding sample parameter corresponding to the at least one confirmed input parameter and at least one unconfirmed reference parameter; inputting the sample parameters of the at least one sample into the trained predictive model to obtain at least one candidate sample, wherein the predicted result of the at least one candidate sample approximates or matches the target result; calculating a first vector of the at least one confirmed input parameter and a second vector of at least one corresponding sample parameter for the at least one candidate sample; matching the first vector with the second vector to select a reference sample, wherein the second vector of the reference sample approximates or matches the first vector; and determining the at least one unconfirmed reference parameter of the reference sample as the at least one unconfirmed input parameter.

[0019] The present invention provides a trained prediction model, the trained prediction model being obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, each sample including a plurality of sample parameters, the trained prediction model being used to input the plurality of input parameters and generate a prediction result corresponding to the input parameters; setting the prediction result of the trained prediction model as a target result, and providing at least one confirmed input parameter among the input parameters, each sample parameter including at least one corresponding sample parameter corresponding to the at least one confirmed input parameter and at least one unconfirmed reference parameter; calculating a first vector of the at least one confirmed input parameter; and calculating at least one unconfirmed reference parameter of the reference sample as at least one unconfirmed input parameter.

[0020] The present invention provides a trained prediction model, which is obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, each sample including a plurality of sample parameters, each sample parameter including at least one corresponding sample parameter and at least one unconfirmed reference parameter, and the trained prediction model is used to input the plurality of input parameters and generate a prediction result corresponding to the input parameters; setting the prediction result of the trained prediction model as a target result and providing at least one confirmed input parameter among the input parameters; inputting a combination of the at least one confirmed input parameter and at least one unconfirmed reference parameter for each sample parameter into the trained prediction model to obtain a reference sample from the sample, so that the prediction result of the reference sample approximates or matches the target result; and using the at least one unconfirmed reference parameter in the sample parameters of the reference sample as at least one unconfirmed input parameter. [Effects of the Invention]

[0021] In summary, the present invention provides a process parameter calculation method that uses a desired target result and confirmed input parameters as inputs and a trained prediction model dataset to inversely estimate unconfirmed input parameters, thereby effectively optimizing the process method and achieving the target result of the process. In addition to applications in the field of cellular processes, the process parameter calculation method of the present invention can also be applied to any other trained machine learning model to inversely estimate process parameters and unconfirmed input parameters. [Brief explanation of the drawings]

[0022] [Figure 1] FIG. 1 shows a flow chart of the steps of a method for calculating process parameters according to an embodiment of the present invention. [Figure 2]FIG. 2 shows a block diagram of a data set and a sample in a method for calculating process parameters according to a specific embodiment of the present invention. [Figure 3] FIG. 3 shows a flow chart of the steps of a method for calculating process parameters according to an illustrative embodiment of the present invention. [Figure 4] FIG. 4 shows a flow chart of steps of a method for calculating process parameters according to another embodiment of the present invention. [Figure 5] FIG. 5 shows a flow chart of steps of a method for calculating process parameters according to another embodiment of the present invention. [Figure 6] FIG. 6 shows a flow chart of steps of a method for calculating process parameters according to another exemplary embodiment of the present invention. [Figure 7] FIG. 7 shows a block diagram of a data set and a sample in a method for calculating process parameters according to another specific embodiment of the present invention. [Figure 8] FIG. 8 shows a flow chart of steps of a method for calculating process parameters according to another exemplary embodiment of the present invention. [Figure 9] FIG. 9 shows a flow chart of steps of a method for calculating process parameters according to another exemplary embodiment of the present invention. [Figure 10] FIG. 10 shows a flow chart of steps of a method for calculating process parameters according to another exemplary embodiment of the present invention. [Figure 11] FIG. 11 shows a flow chart of steps of a method for calculating process parameters according to another exemplary embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0023] To make the advantages, spirit, and features of the present invention more easily and clearly understandable, the present invention will be described and discussed in detail below using specific examples and with reference to the drawings. It should be noted that these specific examples are merely representative examples of the present invention, and the specific methods, devices, conditions, materials, etc. exemplified do not limit the present invention or the corresponding specific examples. Furthermore, the components in the drawings are used only to represent their relative positions and are not drawn to actual scale. Furthermore, the step numbers of the present invention are used only to distinguish different steps and do not represent the order of the steps. The above will be explained in advance.

[0024] Please refer to Figures 1 and 2. Figure 1 shows a flowchart of the steps of the process parameter calculation method according to a specific embodiment of the present invention. Figure 2 shows a block diagram of data sets and samples in the process parameter calculation method according to a specific embodiment of the present invention. As shown in Figure 1, in this specific embodiment, the process parameter calculation method includes:

[0025] Step S1: Provide a trained prediction model. The trained prediction model is obtained by machine learning a dataset using a machine learning method. The dataset includes a plurality of samples (including sample 1 to sample n). Each sample includes a plurality of sample parameters. The trained prediction model is used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters.

[0026] Step S2: The predicted result of the trained predictive model is set as the target result, and the confirmed input parameters 10 of the input parameters are provided.

[0027] Step S3: Using an inverse inference algorithm, the target result, the confirmed input parameters and the sample parameters of the samples in the data set are matched to identify unconfirmed input parameters 12 among the input parameters.

[0028] In this specific embodiment, the trained predictive model in step S1 may be any model that has completed machine learning training and is available on an open platform, or may be a model that has been trained by a user. The dataset in this specific embodiment may be a set of data that can be used for machine learning training, testing, and validation. In practical applications, the process parameter calculation method in this specific embodiment may be applied to any other trained machine learning model, where the confirmed input parameters are used to match data in a database, and the confirmed input parameters are combined with unconfirmed reference parameters to inversely estimate the unconfirmed input parameters.In addition, the trained prediction model in this specific embodiment may be Artificial Neural Networks (ANN), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Decision Trees (Decision Trees), Support Vector Machines (SVM), Random Forests, K-Nearest Neighbors (KNN), K-Means Clustering, Principal Component Analysis (PCA), Linear Regression, Logistic Regression, Gradient Boosting Machines (GVMMs), Deep Belief Networks (DBN), Recursive Neural Networks (RecNN), Reinforcement Learning, Autoencoders, Gaussian Processes (GVMMs), and others. The data set is obtained by machine learning using neural networks (NNs), neural processes (NNs), complex neural networks (CNs), or any other machine learning or neural network algorithm. The choice of machine learning or neural network algorithm depends on the needs of the user.

[0029] In this specific embodiment, the dataset may further include a process dataset. The process dataset includes a plurality of process samples. Each process sample includes a plurality of process sample parameters, and each process sample parameter includes a source parameter and a culture parameter. When the process parameter calculation method in this specific embodiment is applied to the field of cell culture processing, the process dataset may be a cell dataset. In practice, the cell dataset may be any data available from an open platform or data collected by a user. In addition, the types of cell samples may be immune cells (e.g., dendritic cells (DC cells), cytokine-induced killer cells (CIK), tumor-infiltrating lymphocytes (TIL), natural killer cells (NK cells), and CAR-T cells), stem cells (e.g., peripheral blood stem cells, adipose stem cells, and bone marrow mesenchymal stem cells), chondrocytes, fibroblasts, etc. However, in actual applications, the types of cell samples are not limited to these and can be determined according to the type of cell culture that the user wants to perform. In addition, in this specific embodiment, the source parameters of each cell sample may further include attribute data of the source of each cell sample. Each cell sample in the cell dataset may include a cell source corresponding to the cells and attribute data of the source, such as physiological data of the source or other data related to the source, such as the source's gender, age, medical history, living environment, and residential area, etc. However, in practical applications, the source parameters of the cell sample may also include other parameters related to the cell source that may affect cellular processes.

[0030] In this specific embodiment, the culture parameters for each cell sample further include human, equipment, material, method, and environmental parameters for each cell sample. The cell culture process includes many steps, each of which is associated with many culture parameters. These culture parameters include human-related parameters such as the gender and age of the cell source, the experience and stability of the cell culture operator, differences in the personnel performing the surgery, location, environment, and transportation; equipment-related parameters such as the type and grade of the cell operation platform and the stability and accuracy of the temperature and humidity control of the cell culture device; material-related parameters such as the material of the cell culture dish and the components and ratios and formulations of the cell culture medium; method-related parameters such as the technique of the cell culture operator and the method of the cell culture process; and environmental-related parameters such as the environmental temperature, humidity, carbon dioxide, and organic molecule concentration of the cell culture environment.

[0031] In this specific embodiment, the process parameter calculation method uses an inverse inference algorithm to find unconfirmed process parameters in a process waiting for confirmation by comparing them with a dataset for generating a trained predictive model. In detail, if the trained model used by the user has excellent predictive effect, it is not possible to inversely estimate the input process parameters based on the trained model alone, but since the trained model is obtained by machine learning the dataset, it is possible to use the process parameter data in the dataset as the basis for inversely estimating the process parameters.

[0032] Please refer to Figures 1, 2 and 3 together. Figure 3 shows a flowchart of the steps of a method for calculating process parameters according to a specific embodiment of the present invention. This specific embodiment will be further described below with reference to Figure 3. In this specific embodiment, each sample parameter includes at least one corresponding sample parameter and at least one unidentified reference parameter. As shown in Figure 3, step S3 of the method for calculating process parameters in this specific embodiment further includes steps S31, S32, S33, S34, S35 and S36, which are sequentially performed after step S2.

[0033] Step S31: Calculate the first vector of confirmed input parameters 10.

[0034] Step S32: Based on the confirmed input parameters 10, corresponding sample parameters C1 to Cn are obtained from the sample parameters of each sample.

[0035] Step S33: A second vector of each of the corresponding sample parameters C1 to Cn is calculated.

[0036] Step S34: Match the first vector with the second vector of each sample to select a candidate sample i (ie, sample i).

[0037] Step S35: The unconfirmed reference parameters Pi of the candidate sample i are combined with the confirmed input parameters 10 and input into the trained prediction model to obtain a reference sample i from the sample, where the second vector of the reference sample i is close to or consistent with the first vector, and the prediction result of the reference sample i is close to or consistent with the target result.

[0038] Step S36: The unconfirmed reference parameter Pi in the sample parameters of the reference sample i is set as the unconfirmed input parameter 12.

[0039] As shown in Figures 1, 2, and 3, the process awaiting confirmation may include an incubation stage and an incubation waiting stage (not shown). Corresponding to the incubation stage and the incubation waiting stage, the process awaiting confirmation may further include incubation process parameters (i.e., confirmed input parameters 10) and incubation waiting process parameters (i.e., unconfirmed input parameters 12). In this specific example, the dataset may include process parameter data having an incubation time and flow rate related to or similar to the process awaiting confirmation, as a basis for inversely estimating the unconfirmed input parameters 12. The dataset includes Sample 1 to Sample n. Each sample includes sample parameters (not shown), which further include corresponding sample parameters C1 to Cn and unconfirmed reference parameters P1 to Pn. For example, a user may be attempting to implement a 21-day process and have already implemented the process for 7 days (i.e., the user already has confirmed input parameters 10), but may not have confirmed how to implement the process for the subsequent 8th to 21st days or adjust the process parameters (i.e., unconfirmed input parameters 12) to achieve the target results the user desires on the pre-set 21st day.

[0040] In this specific embodiment, the inverse inference algorithm in step S3 first selects a candidate sample from the dataset by comparing the confirmed input parameters 10 obtained from the user's seven-day test with the corresponding sample parameters C1 to Cn for the first seven days of all samples in the dataset (i.e., sample 1 to sample n). The unconfirmed reference parameters of the candidate sample are then combined with the confirmed input parameters 10 of the awaiting confirmation process to form a reference sample. The reference sample is then input into the trained prediction model to confirm whether the reference sample matches the desired target result. If so, the unconfirmed reference parameters of the reference sample can be used as the unconfirmed input parameters 12 of the awaiting confirmation process. For example, if the confirmed input parameters 10 are compared with all corresponding sample parameters C1 to Cn of sample 1 to sample n and it is found that the corresponding sample parameters Ci of sample i are most similar or identical to the confirmed input parameters 10, the unconfirmed reference parameters Pi of sample i and the confirmed input parameters 10 are used to form a reference sample i, and all parameters of the reference sample i are input into the trained prediction model. If the predicted result is equal to or better than the target result, the unconfirmed reference parameter Pi can be set as the unconfirmed input parameter 12 of the process awaiting confirmation. That is, it can be set as the process parameter for the 8th to 21st days. In this specific embodiment, the selection of candidate samples may further include calculating the included angle between the first vector and the second vector of each sample, and the sample corresponding to the second vector with the smallest included angle with the first vector may be set as the candidate sample. In practice, the process awaiting confirmation may not have an incubation stage but only an incubation stage, and may include only corresponding incubation process parameters (i.e., unconfirmed input parameters 12), and the first vector in this embodiment may be an empty vector.

[0041] In the specific example described above, in step S3, the confirmed input parameters are matched with the sample parameters of the samples in the data set to obtain matching samples, and the process parameters for days 8 to 21 (unconfirmed input parameters 12) are identified from the sample parameters. Therefore, the following describes how to obtain matching samples using the confirmed input parameters by matching. Here, reference is made to FIGS. 2 and 3. As shown in FIG. 3, in this specific example, the first vector of the confirmed input parameters 10 in the process awaiting confirmation is matched with the second vector of the corresponding sample parameters C1 to Cn in the sample parameters of all samples (samples 1 to n) in the data set. If the second vector of sample i is close to or identical to the first vector of the confirmed input parameters 10 in the process awaiting confirmation (i.e., the included angle between the two vectors is close to or equal to 0 degrees), and if inputting sample i into the trained prediction model produces a prediction result equal to or better than the target result, sample i can be used as reference sample i. Subsequently, the user can set sample parameters other than the corresponding sample parameters in this reference sample i (ie, unconfirmed reference parameters Pi) as unconfirmed input parameters 12.

[0042] Furthermore, in this specific embodiment, when step S34 is performed to compare the first vector with the second vector of each sample, if the second vector of sample 1 is not similar to or consistent with the first vector of the confirmed input parameters 10 in the confirmation process (i.e., the included angle between the two vectors is not approximately 0 degrees), or if sample 1 is input into the trained prediction model and a prediction result that does not meet the target result is obtained, sample 1 is rejected. Then, other samples in the dataset are taken, and steps S32 to S35 are performed again until a similar or consistent sample is found. Since these parameters have all been verified to have excellent predictive effects in the trained model, after completing the confirmed process steps, the user can continue to perform subsequent process steps using the unconfirmed input parameters obtained by the above method, and ultimately achieve the user's target result. In practice, the user may adjust and set the indicators of similarity and consistency according to different circumstances and needs. It should be noted that the other steps of the process parameter calculation method in this specific embodiment are essentially the same as the corresponding steps in the above specific embodiments, and therefore will not be described in detail here.

[0043] In addition to the above specific embodiment, the present invention also provides another method for selecting candidate samples. Now, refer to Fig. 4. Fig. 4 shows a flowchart of steps of a method for calculating process parameters according to another specific embodiment of the present invention. This specific embodiment may further include step S341, which is sequentially performed after step S33.

[0044] Step S341: Using a distance function, the distance between the coordinates of the end point of the first vector and the coordinates of the end point of the second vector of each sample is calculated, and the sample corresponding to the second vector that has the smallest distance from the first vector is determined as the candidate sample.

[0045] In practical applications, distance functions commonly used in the field of machine learning include Euclidean distance, Manhattan distance, cosine similarity, Jaccard similarity, and Mahalanobis distance. In practice, the calculation method and application scope of the minimum distance can be set according to user needs. It should be noted that the other steps of the process parameter calculation method in this specific embodiment are substantially the same as the corresponding steps in the above-mentioned specific embodiment, and therefore will not be described in detail again here.

[0046] As shown in the above specific examples, vectorizing the parameters in the dataset can provide a trained prediction model for more efficient calculation and processing. However, in actual applications, when vectorizing the parameters in the dataset to identify a reference sample, in addition to the above-mentioned methods of calculating the included angle between the first vector and the second vector and calculating the distance between the first vector and the second vector, other vector calculation methods or other matching methods may be selected to identify the reference sample according to the user's needs, the application scenario, and the type of dataset.

[0047] In the above specific embodiment, the parameters in the data set are vectorized, and reference samples that match or are compatible with the confirmed input parameters are obtained by matching using the included angle and distance between the parameters, thereby identifying the unconfirmed input parameters. However, if a matching reference sample cannot be obtained by matching using the included angle and distance between the parameters, or if the obtained included angle and distance exceed the applicable range of the preset minimum included angle and minimum distance, other matching conditions may be used to obtain a more compatible reference sample. Now, please refer to FIG. 5. FIG. 5 shows a flowchart of steps of a process parameter calculation method according to another specific embodiment of the present invention. As shown in FIG. 5, the differences between this specific embodiment and the above-mentioned specific embodiments are as follows: That is, the process parameter calculation method in this specific embodiment further includes steps S31′, S32′, S33′, S341′, and S342′, which are sequentially performed after step S2.

[0048] Step S31': By using the inverse inference algorithm, the target result, the confirmed input parameters 10 and the sample parameters of the samples in the data set are matched to confirm the sample parameters of the first reference sample among the samples as reference input parameters, where the reference input parameters include unconfirmed reference parameters.

[0049] Step S32': The confirmed input parameters 10 and the unconfirmed reference parameters of the reference input parameters are input into the trained prediction model to generate a reference prediction result.

[0050] Step S33': The reference prediction result is compared with the target result.

[0051] Step S341′: If the reference prediction result matches the target result, the unconfirmed reference parameter P1 is identified as the unconfirmed input parameter 12.

[0052] Step S342′: If the reference prediction result does not match the target result, the first reference sample is excluded, and the inverse inference algorithm is again used to match the target result, the confirmed input parameters and the sample parameters of the samples in the dataset, and the sample parameters of the second reference sample among the samples are obtained as the reference input parameters 12.

[0053] In this specific example, a more appropriate reference sample is obtained by sequentially evaluating the reference prediction results of the complete parameter vector formed by the first vector of confirmed input parameters 10 and the unconfirmed reference parameters P1 to Pn of each sample (sample 1 to sample n). For example, when a user is conducting a 21-day culture process, after completing the first 7 days of the process (i.e., confirmed input parameters 10), he or she may want to confirm the process parameters for the subsequent 8 to 21 days (i.e., unconfirmed input parameters 12) to achieve the target result (95% cell viability) on the 21st day. However, if the user is unable to obtain a matching or compatible reference sample by matching the included angle and distance between parameters to obtain the unconfirmed input parameters 12 that can achieve the desired target result, the following method may also be used. That is, in practice, an inverse inference algorithm is first used to match the target result, the confirmed input parameters, and the sample parameters of the samples in the dataset, and the samples are sorted in order of the similarity between the first vector and the two vectors. Next, the complete parameters formed by combining the confirmed input parameters 10 and the unconfirmed reference parameters of each sample are input into the trained prediction model in the above order to generate a reference prediction result, and the reference prediction result is further compared with the target result. At this time, if the reference prediction result of the first reference sample in the order matches the target result (95% cell viability), the unconfirmed reference parameters of the first reference sample can be identified as unconfirmed input parameters 12. Furthermore, if the reference prediction result of the first reference sample does not match the target result (95% cell viability), the first reference sample is excluded, and the unconfirmed reference parameters of other samples are combined with the confirmed input parameters 10 in order and input into the trained prediction model until the reference prediction result of the reference sample (i.e., the second reference sample) matches the target result (95% cell viability).

[0054] Furthermore, in practical applications, if the reference prediction result of the cell reference sample ultimately fails to match the target result (95% cell viability) through matching, the current matching process may be stopped. In practice, the user may further adjust and set the matching index according to different circumstances and needs. That is, if the reference prediction result of the cell reference sample fails to match the target result (95% cell viability) through matching using the inverse inference algorithm, the current matching process may be stopped. Alternatively, the target result may be adjusted (e.g., the original 95% cell viability may be adjusted to 93% cell viability), or the goal may be changed to obtaining a process reference sample whose reference prediction result is closest to the target result through matching. In this specific embodiment, the inverse inference algorithm automatically matches sample parameters in the dataset to inversely estimate the process parameters required to achieve the desired target result. This effectively optimizes the process method and achieves the target process result. It should be noted that the other steps of the process parameter calculation method in this specific embodiment are essentially the same as the corresponding steps in the above-mentioned specific embodiments, and therefore will not be described in detail again.

[0055] In addition, the present invention may simultaneously compare the sample parameters of all process samples in a process dataset to find a process reference sample whose reference prediction result matches the target result (95% cell viability), thereby forming a process reference sample set, and further select an optimal combination of unconfirmed input parameters. Please refer to FIGS. 6 and 7. FIG. 6 shows a flowchart of the steps of a process parameter calculation method according to another specific embodiment of the present invention. FIG. 7 shows a block diagram of a dataset and samples in the process parameter calculation method according to another specific embodiment of the present invention. As shown in FIG. 6, the process parameter calculation method in this specific embodiment differs from the above-mentioned specific embodiments in the following respects. That is, this specific embodiment may further include steps S32″ to S33″, which are sequentially performed after step S31′.

[0056] Step S32: Match the corresponding sample parameters C1 to Cn of all samples (including sample 1 to sample n) in the dataset, and input the confirmed input parameters 10 and the unconfirmed reference parameters (including P1 to Pn) corresponding to each sample into the trained prediction model to generate the corresponding reference prediction results.

[0057] Step S33: Match the reference prediction result corresponding to each sample with the target result to find the reference samples whose reference prediction result matches the target result, and further form a reference sample set.

[0058] In this specific embodiment, the sample parameters of all process samples in the process dataset are collated at once to obtain one or more process reference samples whose reference predicted results match the target result (95% cell viability), and a cell reference sample set can be formed. Furthermore, based on the conditions and constraints set by the user, unconfirmed input parameters 12 are selected for the process reference samples in the process reference sample set. In practice, the user may set conditions and constraints taking into account their own consumable inventory and costs, and further select process parameters from the reference sample set that meet the target result and the user's needs. It should be noted that the other steps of the process parameter calculation method in this specific embodiment are substantially the same as the corresponding steps in the above-mentioned specific embodiments, and therefore will not be described in detail again here.

[0059] In addition, the process parameter calculation method of the present invention may be implemented in other modes than those described above. Now, reference is made to FIGS. 7 and 8. FIG. 8 shows a flowchart of the steps of a process parameter calculation method according to another specific embodiment of the present invention. As shown in FIG. 8, the process parameter calculation method of this specific embodiment differs from the above-described specific embodiments in the following respects. That is, this specific embodiment may further include step S34" executed after step S33".

[0060] Step S34): Linearly combine the first K data in the reference sample set and find the optimal combination of unconfirmed input parameters using an optimization search method.

[0061] In this specific embodiment, the user may use a process reference sample set obtained by simultaneously matching the process data set, and further arrange the process reference samples in the process reference sample set using conditions and constraints set by the user. The first K data in the process reference sample set are then linearly combined to find the optimal combination of unidentified input parameters using an optimization search method. In practice, the user may set conditions and constraints based on their own consumable inventory and costs, and further arrange the process reference sample set to select the reference sample that best meets the user's needs and target results using an optimization search method. In practical applications, the optimization search method may be the finite difference method, gradient descent method, Newton's method, penalty function method, barrier function method, or Lagrange multiplier method, or other calculation methods may be selected based on the user's needs. It should be noted that the other steps of the method for calculating process parameters in this specific embodiment are almost the same as the corresponding steps in the above-mentioned specific embodiment, and therefore will not be described in detail again here.

[0062] The process parameter calculation method of the present invention may be implemented in other embodiments besides the above. Further explanations are provided below. Reference is now made to FIG. 9, which shows a flowchart of steps of a process parameter calculation method according to another specific embodiment of the present invention. As shown in FIG. 9, in this specific embodiment, the process parameter calculation method includes:

[0063] Step S1': Provide a trained prediction model. The trained prediction model is obtained by machine learning a dataset using a machine learning method. The dataset includes a plurality of samples, and each sample includes a plurality of sample parameters. The trained prediction model is used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters.

[0064] Step S2': Set the predicted result of the trained prediction model as the target result, and provide confirmed input parameters among the input parameters. Each sample parameter includes a corresponding sample parameter corresponding to the confirmed input parameter and an unconfirmed reference parameter.

[0065] Step S3': Input the sample parameters of at least one sample into the trained prediction model to obtain at least one candidate sample, and the predicted result of the at least one candidate sample is close to or coincides with the target result.

[0066] Step S4': Calculate a first vector of at least one confirmed input parameter and a second vector of at least one corresponding sample parameter for at least one candidate sample.

[0067] Step S5': Match the first vector with the second vector to select a reference sample, where the second vector of the reference sample is close to or coincides with the first vector.

[0068] Step S6': At least one unconfirmed reference parameter of the reference sample is set as at least one unconfirmed input parameter.

[0069] The process parameter calculation method in this specific embodiment differs from the above-described specific embodiments in the following respects. Specifically, in this specific embodiment, a combination of corresponding sample parameters and unconfirmed reference parameters for each sample in a dataset is directly input into a trained predictive model to obtain a predicted result. If the predicted result of a sample in the dataset matches the target result, the sample can be further considered a candidate sample. Next, a first vector of confirmed input parameters is calculated and compared with a second vector of corresponding sample parameters for the candidate sample. A sample whose second vector is similar to or identical to the first vector is found from the candidate samples and designated as a reference sample. The corresponding unconfirmed reference parameters can then be designated as unconfirmed input parameters for the current process. In practice, the process parameter calculation method in this specific embodiment is suitable for processes involving a large amount of data in the sample dataset. First, a reference sample whose predicted result is similar to or identical to the target result and whose second vector is similar to or identical to the first vector is found from the sample dataset. This eliminates the need to combine confirmed input parameters with unconfirmed reference parameters and input them into a trained predictive model, thereby efficiently and quickly obtaining unconfirmed input parameters for the process. It should be noted that the other steps of the method for calculating process parameters in this specific embodiment are almost the same as the corresponding steps in the above-mentioned specific embodiment, and therefore will not be described in detail again here.

[0070] Furthermore, the method for calculating process parameters in the present invention may also be implemented in other ways. Now, refer to FIG. 10. FIG. 10 shows a flowchart of steps of a method for calculating process parameters according to another specific embodiment of the present invention. As shown in FIG. 10, in this specific embodiment, the method for calculating process parameters includes:

[0071] Step S1'': Provide a trained prediction model. The trained prediction model is obtained by machine learning a dataset using a machine learning method. The dataset includes a plurality of samples, and each sample includes a plurality of sample parameters. The trained prediction model is used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters.

[0072] Step S2'': Set the predicted result of the trained prediction model as the target result, and provide confirmed input parameters among the input parameters. Each sample parameter includes a corresponding sample parameter corresponding to the confirmed input parameter and an unconfirmed reference parameter.

[0073] Step S3''): Calculate the first vector of confirmed input parameters.

[0074] Step S4''): Based on the confirmed input parameters, obtain the corresponding sample parameters from the sample parameters of each sample.

[0075] Step S5''): Calculate the second vector of parameters for each corresponding sample.

[0076] Step S6'': Match the first vector with the second vector of each sample to select a candidate sample whose second vector is close to or coincides with the first vector.

[0077] Step S7″): Input the sample parameters of the candidate sample into the trained prediction model to obtain a reference sample, and the prediction result of the reference sample is close to or coincides with the target result.

[0078] Step S8″): The unconfirmed reference parameters in the sample parameters of the reference sample are set as unconfirmed input parameters.

[0079] The difference between this specific embodiment and the above-mentioned specific embodiment is as follows: first, the first vector is compared with the second vector of each sample, and then the predicted result is compared with the target result. In this case, the unknown input parameters of the process can be obtained more accurately than in the above-mentioned specific embodiment. It should be noted that the other steps of the process parameter calculation method in this specific embodiment are almost the same as the corresponding steps in the above-mentioned specific embodiment, and therefore will not be described in detail again here.

[0080] Furthermore, the method for calculating process parameters in the present invention may also be implemented in other ways. Now, refer to FIG. 11. FIG. 11 shows a flowchart of steps of a method for calculating process parameters according to another specific embodiment of the present invention. As shown in FIG. 11, in this specific embodiment, the method for calculating process parameters includes:

[0081] Step S1'''): Provide a trained predictive model. The trained predictive model is obtained by machine learning a dataset using a machine learning method. The dataset includes a plurality of samples, and each sample includes a plurality of sample parameters. The trained predictive model is used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters.

[0082] Step S2'''): Set the predicted result of the trained prediction model as the target result, and provide the confirmed input parameters among the input parameters.

[0083] Step S3'''): The confirmed input parameters and unconfirmed reference parameters for each sample parameter are combined and input into the trained prediction model to obtain a reference sample from these samples. The predicted result of the reference sample is close to or coincides with the target result.

[0084] Step S4'''): Unconfirmed reference parameters in the sample parameters of the reference sample are set as unconfirmed input parameters.

[0085] This specific embodiment differs from the above-described specific embodiment in the following respects: In this specific embodiment, the confirmed input parameters and the unconfirmed reference parameters for each sample parameter are directly combined and input into the trained prediction model to obtain a prediction result without needing to match the first vector with the second vector for each sample. Then, a reference sample is obtained based on whether the prediction result corresponding to each sample is close to or consistent with the target result. Compared to the above-described specific embodiment, which requires two steps to obtain the reference sample, this specific embodiment can obtain the reference sample in one step, aiming to perform comprehensive matching and find the optimal solution. It should be noted that the other steps of the process parameter calculation method in this specific embodiment are substantially the same as the corresponding steps in the above-described specific embodiment, and will not be described in detail again.

[0086] In summary, the present invention provides a process parameter calculation method that uses a desired target result and confirmed input parameters as inputs and a trained prediction model dataset to inversely estimate unconfirmed input parameters, thereby effectively optimizing the process method and achieving the target result of the process. In addition to applications in the field of cellular processes, the process parameter calculation method of the present invention can also be applied to any other trained machine learning model to inversely estimate process parameters and unconfirmed input parameters.

[0087] The detailed description of the preferred specific embodiments above is intended to more clearly describe the features and spirit of the present invention, and is not intended to limit the scope of the present invention to the preferred specific embodiments disclosed above. Rather, it is intended that various modifications and equivalent structures be covered within the scope of the claims intended to be claimed by the present invention. Therefore, the scope of the claims intended to be claimed by the present invention should be interpreted in the broadest possible manner based on the above description so as to cover all possible modifications and equivalent structures. [Explanation of symbols]

[0088] S1~S3, S31~S35, S341 steps S31'~S33', S341', S342' steps S32''~S34'', S1''''~S6'' steps S1''''~S8'''', S1''''''~S4''''' steps 10 Verified Input Parameters 12 Unvalidated Input Parameters C1, C2, C3, Ci···Cn Corresponding sample parameters P1,P2,P3,Pi···Pn Unconfirmed reference parameters

Claims

1. providing a trained prediction model, the trained prediction model being obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, and each of the samples including a plurality of sample parameters, the trained prediction model being used to input the plurality of input parameters and generate a prediction result corresponding to the input parameters; setting the predicted result of the trained predictive model as a target result and providing at least one verified input parameter of the input parameters; and matching the target result, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset using an inverse inference algorithm to identify at least one unconfirmed input parameter among the input parameters.

2. 2. The method of claim 1, wherein the data set further comprises a process data set, the process data set comprising a plurality of process samples, each of the process samples comprising a plurality of process sample parameters, each of the process sample parameters comprising a source parameter and a culture parameter.

3. 3. The method of claim 2, wherein the source parameters of each of the process samples further include attribute data of the source of each of the process samples.

4. The method for calculating process parameters according to claim 2 , wherein the culture parameters for each of the process samples further include parameters of people, equipment, materials, methods, and environments in each of the process samples.

5. Each of the sample parameters includes at least one corresponding sample parameter and at least one unconfirmed reference parameter, and the step of identifying the at least one unconfirmed input parameter among the input parameters by the inverse inference algorithm by matching the target result, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset further includes: calculating a first vector of said at least one verified input parameter; respectively obtaining the at least one corresponding sample parameter from the sample parameters of the samples based on the at least one confirmed input parameter; calculating a second vector of said at least one corresponding sample parameter; selecting at least one candidate sample by matching the first vector with the second vector of each sample; obtaining a reference sample from the at least one candidate sample by combining the at least one unconfirmed reference parameter and the confirmed input parameter of the at least one candidate sample and inputting them into the trained prediction model, wherein the second vector of the reference sample is close to or coincides with the first vector, and the prediction result of the reference sample is close to or coincides with the target result; 2. The method for calculating a process parameter according to claim 1, further comprising the step of: determining the at least one unconfirmed reference parameter in the sample parameters of the reference sample as the at least one unconfirmed input parameter.

6. The step of selecting the at least one candidate sample by matching the first vector with the second vector of each of the samples further includes:

6. The method for calculating a process parameter according to claim 5, further comprising the step of calculating an included angle between the first vector and the second vector of each of the samples, and determining the sample corresponding to the second vector that has the smallest included angle with the first vector as the candidate sample.

7. The step of selecting the at least one candidate sample by matching the first vector with the second vector of each of the samples further includes:

6. The method for calculating a process parameter according to claim 5, further comprising the step of calculating a distance between a coordinate of an end point of the first vector and a coordinate of an end point of the second vector of each of the samples using a distance function, and determining the sample corresponding to the second vector having the smallest distance from the first vector as the candidate sample.

8. The trained prediction models include artificial neural networks (ANN), convolutional neural networks (CNNs), recurrent neural networks (RNNs), decision trees (Decision Trees), support vector machines (SVMs), random forests (Random Forests), k-nearest neighbors (KNN), k-means clustering (K-Means Clustering), principal component analysis (PCA), linear regression (Linear Regression), and so on.

2. The method of claim 1, wherein the dataset is obtained by machine learning using any of Deep Belief Networks (DBN), Logistic Regression, Gradient Boosting Machines, Deep Belief Networks (DBN), Recursive Neural Networks (RecNN), Reinforcement Learning, Autoencoders, Gaussian Processes, and Complex Neural Networks.

9. The step of identifying the at least one unconfirmed input parameter among the input parameters by the inverse inference algorithm by matching the target result, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset further includes: using the inverse inference algorithm to match the target result, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset to confirm the sample parameters of a first reference sample among the samples as reference input parameters, wherein the reference input parameters include at least one unconfirmed reference parameter; inputting the at least one confirmed input parameter and the at least one unconfirmed reference parameter into the trained predictive model to generate a reference prediction result; matching the reference prediction result with the target result; 2. The method of claim 1, further comprising identifying the at least one unconfirmed reference parameter as the at least one unconfirmed input parameter if the reference predicted result matches the target result.

10. Furthermore, 10. The method for calculating process parameters according to claim 9, further comprising: if the reference prediction result does not match the target result, excluding the first reference sample, and again using the inverse inference algorithm to compare the target result, the at least one confirmed input parameter, and the sample parameters of the samples in the dataset, to obtain the sample parameters of a second reference sample among the samples as the reference input parameters.

11. Furthermore, matching at least one corresponding sample parameter of the samples in the dataset, and inputting at least one confirmed input parameter and the at least one unconfirmed reference parameter corresponding to each of the samples into the trained prediction model to generate the corresponding reference prediction result; 11. The method for calculating a process parameter according to claim 10, further comprising: matching the reference prediction result corresponding to each of the samples with the target result to find a reference sample whose reference prediction result matches the target result, and further forming a reference sample set.

12. Furthermore, 12. The method for calculating process parameters according to claim 11, further comprising the step of linearly combining the first K data in the reference sample set and finding an optimal combination of unconfirmed input parameters by an optimization search method.

13. providing a trained prediction model, the trained prediction model being obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, and each of the samples including a plurality of sample parameters, the trained prediction model being used to input the plurality of input parameters and generate a prediction result corresponding to the input parameters; setting the predicted result of the trained predictive model as a target result and providing at least one validated input parameter among the input parameters, each of the sample parameters including at least one corresponding sample parameter corresponding to the at least one validated input parameter and at least one unvalidated reference parameter; inputting the sample parameters of at least one of the samples into the trained prediction model to obtain at least one candidate sample, and the predicted result of the at least one candidate sample approximates or matches the target result; calculating a first vector of the at least one confirmed input parameter and a second vector of the at least one corresponding sample parameter for the at least one candidate sample; selecting a reference sample by comparing the first vector with the second vector, the second vector of the reference sample being close to or identical to the first vector; a step of determining the at least one unconfirmed reference parameter in the sample parameters of the reference sample as at least one unconfirmed input parameter.

14. providing a trained prediction model, the trained prediction model being obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, and each of the samples including a plurality of sample parameters, the trained prediction model being used to input the plurality of input parameters and generate a prediction result corresponding to the input parameters; setting the predicted result of the trained predictive model as a target result and providing at least one validated input parameter among the input parameters, each of the sample parameters including at least one corresponding sample parameter corresponding to the at least one validated input parameter and at least one unvalidated reference parameter; calculating a first vector of said at least one verified input parameter; respectively obtaining the at least one corresponding sample parameter from the sample parameters of the samples based on the at least one confirmed input parameter; calculating a second vector of said at least one corresponding sample parameter; selecting at least one candidate sample by matching the first vector with the second vector of each of the samples, the second vector of the at least one candidate sample being close to or identical to the first vector; inputting the sample parameters of the at least one candidate sample into the trained prediction model to obtain a reference sample, and the predicted result of the reference sample approximates or matches the target result; a step of determining the at least one unconfirmed reference parameter in the sample parameters of the reference sample as at least one unconfirmed input parameter.

15. providing a trained prediction model, the trained prediction model being obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, each of the samples including a plurality of sample parameters, each of the sample parameters including at least one corresponding sample parameter and at least one unconfirmed reference parameter, the trained prediction model being used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters; setting the predicted result of the trained predictive model as a target result and providing at least one verified input parameter of the input parameters; obtaining a reference sample from the sample by inputting a combination of the at least one confirmed input parameter and at least one unconfirmed reference parameter for each of the sample parameters into the trained predictive model, so that the predicted result of the reference sample approximates or matches the target result; a step of determining the at least one unconfirmed reference parameter in the sample parameters of the reference sample as at least one unconfirmed input parameter.

Images