Method for predicting pharmacokinetic parameters for coagulation factor viii and method for training a corresponding model

Abstract

The application discloses a method for predicting pharmacokinetic parameters of coagulation factor VIII and a corresponding model training method. The training method comprises the following steps: obtaining historical clinical data of children with hemophilia A; determining the true value label of the pharmacokinetic parameters corresponding to the historical clinical data, and storing the historical clinical data and the true value label in a table form as a first training data set; inputting the first training data set into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted value of the corresponding pharmacokinetic parameters; adjusting the model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII according to the true value label and the predicted value to obtain the trained pharmacokinetic parameter prediction model for coagulation factor VIII. The important PK parameters of coagulation factor VIII are accurately predicted while greatly reducing the demand for blood sampling frequency of patients.

Description

Technical Field

[0001] This disclosure generally relates to the fields of pharmacokinetics and machine learning, and more specifically, to a method for predicting pharmacokinetic parameters of coagulation factor VIII and a method for training a corresponding model. Background Technology

[0002] In the relevant field, hemophilia A is a rare inherited bleeding disorder characterized by a deficiency or dysfunction of clotting factor VIII (FVIII). The genetic background of hemophilia A involves mutations in the FVIII gene, located in the Xq28 region, approximately 186 kb in length, containing 26 exons and 25 introns. Point mutations are predominantly point mutations, involving multiple exons and some introns. Currently, exogenous replacement therapy is the primary treatment for hemophilia, involving the use of blood products such as plasma and clotting factors (e.g., recombinant human FVIII) to prevent and control bleeding. However, this treatment requires frequent injections, and some patients may develop anti-FVIII antibodies (inhibitors), leading to treatment ineffectiveness. Pharmacokinetics (PK) serves as a valuable tool for developing individualized FVIII-based treatment strategies for hemophilia A patients. Accurate prediction of PK parameters associated with hemophilia A patients is crucial for optimizing FVIII replacement therapy. Pharmacokinetics is a nonlinear process, and the prediction of its PK parameters is one of the key research directions in related fields. Summary of the Invention

[0003] The embodiments of this disclosure provide a method for predicting pharmacokinetic parameters of coagulation factor VIII and a corresponding model training method, the purpose of which is to achieve accurate prediction of important PK parameters of FVIII for hemophilia A while significantly reducing the need for patient blood sampling data.

[0004] In one general aspect, a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII is provided. The training method includes: acquiring historical clinical data for children with hemophilia A, the historical clinical data including historical demographic information associated with the child, historical drug administration records, and historical factor VIII activity data in the child's blood before and after drug administration (first and second historical factors VIII activity data); determining truth labels for pharmacokinetic parameters corresponding to the historical clinical data, and storing the historical clinical data and the truth labels in tabular form as a first training dataset; inputting the first training dataset into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain predicted values ​​for the corresponding pharmacokinetic parameters; adjusting the model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII based on the truth labels and the predicted values ​​to obtain a trained pharmacokinetic parameter prediction model for coagulation factor VIII, wherein the second historical factor VIII activity data only includes factor VIII activity data collected at three sampling time points.

[0005] Optionally, pharmacokinetic parameters may include recovery rate and half-life. The three sampling time points may include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

[0006] Optionally, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a pharmacokinetic parameter prediction model for coagulation factor VIII based on a linear regression algorithm, a pharmacokinetic parameter prediction model for coagulation factor VIII based on a random forest and XGBoost algorithm, a pharmacokinetic parameter prediction model for coagulation factor VIII based on a neural network algorithm, and / or a pharmacokinetic parameter prediction model for coagulation factor VIII based on a pre-trained language algorithm.

[0007] Optionally, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII can be a pharmacokinetic parameter prediction model for coagulation factor VIII based on a pre-trained language algorithm, and the step of inputting the first training dataset into the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters may include: performing a data serialization operation on the first training data to obtain a second training dataset in text form; and inputting the second training dataset into the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

[0008] Optionally, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a BERT-based BGE encoder and a linear projection layer. Backpropagation and / or gradient descent algorithms may be used to adjust the model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII based on the truth labels and the predicted values.

[0009] In another general aspect, a method for predicting pharmacokinetic parameters of coagulation factor VIII is provided. The method includes: acquiring clinical data to be analyzed for children with hemophilia A, the clinical data including demographic information associated with the child, medication records, and data on the activity of the first coagulation factor VIII in the child's blood before and after medication; inputting the clinical data to be analyzed into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain predicted values ​​of the corresponding pharmacokinetic parameters, wherein the second coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points, and wherein the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is trained using the training method for pharmacokinetic parameter prediction models for coagulation factor VIII described above.

[0010] In another general aspect, a training apparatus is provided for a pharmacokinetic parameter prediction model for coagulation factor VIII, the training apparatus comprising: a data acquisition module configured to: acquire historical clinical data for children with hemophilia A, the historical clinical data including historical demographic information associated with the children, historical drug administration records, and historical data on the first historical coagulation factor VIII activity in the children's blood before drug administration and on the second historical coagulation factor VIII activity in the children's blood after drug administration; and a data preprocessing module configured to: determine truth labels for pharmacokinetic parameters corresponding to the historical clinical data, and preprocess the historical clinical data and the data... The truth labels are stored in tabular form as the first training dataset; the parameter prediction module is configured to input the first training dataset into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters; the parameter tuning module is configured to adjust the model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII according to the truth labels and the predicted values ​​to obtain the trained pharmacokinetic parameter prediction model for coagulation factor VIII, wherein the second historical coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points.

[0011] Optionally, pharmacokinetic parameters may include recovery rate and half-life. The three sampling time points may include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

[0012] Optionally, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a pharmacokinetic parameter prediction model for coagulation factor VIII based on a linear regression algorithm, a pharmacokinetic parameter prediction model for coagulation factor VIII based on a random forest and XGBoost algorithm, a pharmacokinetic parameter prediction model for coagulation factor VIII based on a neural network algorithm, and / or a pharmacokinetic parameter prediction model for coagulation factor VIII based on a pre-trained language algorithm.

[0013] Optionally, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII can be a pharmacokinetic parameter prediction model for coagulation factor VIII based on a pre-trained language algorithm. The operation of the parameter prediction module inputting the first training dataset into the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters may include: performing a data serialization operation on the first training data to obtain a second training dataset in text form; and inputting the second training dataset into the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

[0014] Optionally, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a BERT-based BGE encoder and a linear projection layer. Backpropagation and / or gradient descent algorithms may be used to adjust the model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII based on the truth labels and the predicted values.

[0015] In another general aspect, a device for predicting pharmacokinetic parameters of coagulation factor VIII is provided. The device includes: an acquisition module configured to acquire clinical data to be analyzed for a child with hemophilia A, the clinical data including demographic information associated with the child, medication record data, and data on the activity of a first coagulation factor VIII in the child's blood before administration and data on the activity of a second coagulation factor VIII in the child's blood after administration; and a prediction module configured to input the clinical data to be analyzed into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain predicted values ​​of the corresponding pharmacokinetic parameters, wherein the second coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points, and wherein the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is trained using the training method for the pharmacokinetic parameter prediction model for coagulation factor VIII described above.

[0016] In another general aspect, a computer program product is provided, the computer program product comprising a computer program / instruction that, when executed by a processor, implements the training method for a pharmacokinetic parameter prediction model for coagulation factor VIII as described above, and the method for predicting pharmacokinetic parameters for coagulation factor VIII as described above.

[0017] In another general aspect, a computer-readable storage medium is provided, wherein instructions in the computer-readable storage medium, when executed by a processor of an electronic device / server, enable the electronic device / server to perform the training method for a pharmacokinetic parameter prediction model for coagulation factor VIII as described above, and the prediction method for pharmacokinetic parameters of coagulation factor VIII as described above.

[0018] In another general aspect, a computing device is provided, the computing device comprising: at least one processor; at least one memory storing computer-executable instructions, wherein, when executed by the at least one processor, the computer-executable instructions cause the at least one processor to perform the training method for a pharmacokinetic parameter prediction model for coagulation factor VIII as described above, and the prediction method for pharmacokinetic parameters of coagulation factor VIII as described above.

[0019] The method for predicting pharmacokinetic parameters of coagulation factor VIII and the corresponding model training method according to embodiments of this disclosure, through a machine learning-based method for predicting FVIII pharmacokinetic parameters in hemophilia A patients and the corresponding model training method, achieves accurate prediction of two important PK parameters, including the recovery rate and half-life of FVIII, by using background information such as FVIII activity at predetermined blood collection time points and patient dosing records, while significantly reducing the number of blood collections. Furthermore, the method for predicting pharmacokinetic parameters of coagulation factor VIII and the corresponding model training method according to embodiments of this disclosure can provide beneficial assistance for subsequent PK assessment and personalized dosing regimens by achieving accurate prediction of PK parameters. Attached Figure Description

[0020] The above and other objects and features of the embodiments of this disclosure will become clearer from the following description taken in conjunction with the accompanying drawings illustrating the embodiments, wherein:

[0021] Figure 1 This is a flowchart illustrating a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure.

[0022] Figure 2 This is a flowchart illustrating an example of a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure.

[0023] Figure 3 This is a schematic diagram illustrating the serialization of tabular data according to an embodiment of the present disclosure;

[0024] Figure 4 This is a schematic diagram illustrating a language model-based network architecture according to an embodiment of the present disclosure;

[0025] Figure 5 This is a schematic diagram of a training process illustrating an example of a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure.

[0026] Figure 6 This is a schematic diagram of a test process illustrating an example of a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure.

[0027] Figure 7 This is a flowchart illustrating a method for predicting pharmacokinetic parameters of coagulation factor VIII according to an embodiment of the present disclosure;

[0028] Figure 8This is a structural block diagram illustrating a training apparatus for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure.

[0029] Figure 9 This is a structural block diagram illustrating a device for predicting pharmacokinetic parameters of coagulation factor VIII according to an embodiment of the present disclosure.

[0030] Figure 10 This is a block diagram illustrating a computing device according to an embodiment of the present disclosure. Detailed Implementation

[0031] The following detailed embodiments are provided to aid the reader in gaining a comprehensive understanding of the methods, apparatus, and / or systems described herein. However, various changes, modifications, and equivalents of the methods, apparatus, and / or systems described herein will become apparent upon understanding this disclosure. For example, the order of operations described herein is merely illustrative and is not limited to those orders set forth herein, but may be changed as will become clear upon understanding this disclosure, except for operations that must occur in a specific order. Furthermore, for clarity and conciseness, descriptions of features known in the art may be omitted.

[0032] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings, examples of which are illustrated in the drawings, wherein the same reference numerals always refer to the same parts. The embodiments will now be described with reference to the accompanying drawings in order to explain this disclosure.

[0033] As mentioned above, because pharmacokinetics is a non-linear process, sampling data (e.g., concentration) at a single time point is insufficient for accurate assessment. According to guidelines regarding recombinant human coagulation FVIII, data obtained from blood samples taken before injection of a 25-50 IU / kg dose (i.e., pre-dose) and after injection of a 25-50 IU / kg dose (i.e., post-dose) at 10-15 minutes, 30 minutes, 1 hour, 3 hours, 6 hours, 9 hours, 24 hours, 28 hours, and 32 hours can be used to assess pharmacokinetic parameters for FVIII. For injection doses greater than or equal to 50 IU / kg, blood samples may be collected within 48 hours to obtain more sets of data.

[0034] To achieve accurate prediction of pharmacokinetic parameters while significantly reducing the number of blood samplings, this disclosure provides a machine learning-based method for predicting FVIII pharmacokinetic parameters in children with hemophilia A, along with a corresponding model training method. By combining deep learning and algorithmic applications, using FVIII activity detected from a subset of sampling data at the aforementioned time points, and combining this with, for example, medication records and background information stored in the patient's electronic medical record, a predetermined machine learning model is used to predict PK parameters. This effectively solves the problems of existing methods requiring a large number of blood samples and having low accuracy in predicting PK parameters. The machine learning-based method for predicting FVIII pharmacokinetic parameters in children with hemophilia A proposed in this disclosure can achieve accurate prediction of two important PK parameters—FVIII recovery rate and half-life—using FVIII activity data at the blood sampling time points, along with the patient's medication records and relevant background information, while significantly reducing the number of blood samplings. This provides assistance for further PK assessment and the development of personalized treatment plans.

[0035] The following reference Figures 1 to 10 This disclosure provides a detailed description of the training method and apparatus for a pharmacokinetic parameter prediction model for coagulation factor VIII, and the prediction method and apparatus for pharmacokinetic parameters of coagulation factor VIII, according to embodiments of the present disclosure.

[0036] First, refer to Figures 1 to 6 A detailed description is provided of a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII according to embodiments of the present disclosure.

[0037] Figure 1 This is a flowchart illustrating a training method 100 for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure. Figure 2 This is a flowchart illustrating an example of a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure. Figure 3 This is a schematic diagram illustrating the serialization of tabular data according to an embodiment of the present disclosure. Figure 4 This is a schematic diagram illustrating a language model-based network architecture according to an embodiment of the present disclosure. Figure 5 This is a schematic diagram illustrating an example of a training process for a training method of a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure. Figure 6 This is a schematic diagram of a test process illustrating an example of a training method for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure.

[0038] Reference Figures 1 to 6According to an embodiment of this disclosure, in step S101, historical clinical data for children with hemophilia A (e.g., severe hemophilia A) are obtained.

[0039] Here, historical clinical data may include, for example, historical demographic information associated with the child from the child's electronic medical record, historical medication records, and historical data on the activity of the first historical coagulation factor VIII in the child's blood before administration and the activity of the second historical coagulation factor VIII in the child's blood after administration (e.g., such as...). Figure 2 (as shown in (1)).

[0040] For example, the second historical coagulation factor VIII activity data only includes coagulation factor VIII activity data collected at three sampling time points. Preferably, the three sampling time points may include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

[0041] By using coagulation factor VIII activity data collected at only three sampling time points to predict PK parameters, a novel approach is achieved that can significantly reduce the number of blood samplings and accurately predict PK parameters. This addresses the problem of significantly increased patient suffering and assessment costs, especially for pediatric patients, in existing methods that require intensive multiple blood samplings (e.g., at least six times (e.g., 1 hour, 3 hours, 9 hours, 24 hours, 48 ​​hours, and 72 hours after administration)) to detect FVIII activity.

[0042] According to an embodiment of this disclosure, in step S102, the true value labels of pharmacokinetic parameters corresponding to historical clinical data are determined, and the historical clinical data and true value labels are stored in tabular form as a first training dataset.

[0043] As an example, pharmacokinetic parameters may include recovery rate (IVR) and half-life (HL).

[0044] Specifically, the two PK parameters mentioned above for the children (which also serve as labels for the regression task, representing the PK parameters to be fitted) are calculated by Phoenix WinNonlin software and used as ground truth labels. In other words, the features and labels defined above together constitute the clinical dataset of the children (e.g., ...). Figure 2 As shown in (2) in the figure), and the dataset is stored in tabular form.

[0045] According to an embodiment of this disclosure, in step S103, the first training dataset is input into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

[0046] As an example, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a pharmacokinetic parameter prediction model for coagulation factor VIII based on linear regression algorithms (e.g., Lasso, Ridge regression models), a pharmacokinetic parameter prediction model for coagulation factor VIII based on random forest and XGBoost algorithms, and a pharmacokinetic parameter prediction model for coagulation factor VIII based on neural network algorithms (e.g., MLP and FT-Transformer models). Figure 2 (as shown in (4)) and / or a pharmacokinetic parameter prediction model for coagulation factor VIII based on a pre-trained language algorithm (e.g., as shown in (4)) and / or a model for predicting pharmacokinetic parameters ... Figure 2 (as shown in (6)).

[0047] By employing the aforementioned methods, including traditional machine learning (ML) algorithms and language models (LM), to predict pharmacokinetic parameters, we have achieved the ability to analyze FVIII pharmacokinetic parameters for hemophilia A patients using machine learning methods. Furthermore, through effective evaluation and validation on real-world clinical datasets, we have demonstrated a significant improvement in the predictive performance of FVIII pharmacokinetic parameters for hemophilia A patients.

[0048] Furthermore, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII can be a pre-trained language algorithm-based pharmacokinetic parameter prediction model for coagulation factor VIII. In this case, step S103 may further include the following steps S1031 and S1032:

[0049] In step S1031, a data serialization operation is performed on the first training data to obtain a second training dataset in text form.

[0050] Here, since training method 100 can use a pre-trained language model, the feature name can be appropriately expanded before performing data serialization operations, for example, expanding the feature name "FFM" to "Fat Free Mass", thereby enriching the semantic information that the language model can learn.

[0051] In addition, the categorical features in the training dataset can be processed separately. For the case of using a language model, the original category name is retained, while for other machine learning models, the category is encoded as a number. For example, for the blood type category, the original "A\B\O\AB" is retained when using a language model, while it can be encoded as "1\2\3\4" when using other machine learning models.

[0052] Furthermore, since hemophilia A is a rare disease, the sample size may be relatively small. Therefore, this disclosure divides the dataset into a training set and a test set in an 8:2 ratio (e.g., as shown in the figure). Figure 2 As shown in (3) in the figure), the features and labels are then standardized and scaled to the same dimension.

[0053] As an example, refer to Figure 3 For cases using a language model, the operation of serializing the features of the input samples into text form is as follows: Define a serialization function. , C For feature name, x These are the corresponding feature values. Apply this to each sample. S， Obtain the serialized text t The specific serialization method involves concatenating the feature names (column names) and their corresponding feature values ​​into a long text. t， The middle part is a conjunction, which can be "and" or a separator [SEP], for example, " C 1is x 1 , and C 2is x 2 ,......,C n is x n That is, "feature name is feature value" is concatenated using the conjunction "and". For example, for an input sample, the serialized text (also known as: text sequence) t i It can be expressed as "The Blood Type isA and The Fat Free Mass is 30, ..., The FVIII activity is 90".

[0054] For example, in Figure 3 In the example, Blood Group represents blood type, Height represents height, and Body weight represents weight. The serialized example expression is: "The Blood Group is A and The Height is 114 and The Body weight is 20".

[0055] As another example, see Figure 2 In (5), the serialized example expression of features Column 1, Column 2, ..., Columnn and their corresponding values ​​x1, x2, ..., xn is: "the Column 1 is Value1, and the Column 2 is Value2, and, ..., the Column n is Value n".

[0056] In step S1032, the second training dataset is input into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters (e.g., such as...). Figure 2 (7) shows the predicted IVR and predicted Half-Life.

[0057] In addition, Figure 2 In this context, PK_ID represents the PK sequence number.

[0058] The method disclosed herein is the first to use multiple machine learning models, including pre-trained language models, to predict pharmacokinetic parameters for FVIII, which can improve the feasibility and accuracy of pharmacokinetic analysis for FVIII.

[0059] According to an embodiment of this disclosure, in step S104, the model parameters of a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII are adjusted based on the truth label and the predicted value to obtain a trained pharmacokinetic parameter prediction model for coagulation factor VIII.

[0060] In addition, the Scikit-Learn and PyTorch machine learning frameworks can be used for model building. For example, see [link to example]. Figure 4 The predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a BERT-based BGE (BAAI General Embedding) encoder and a linear projection layer. Figure 4 As can be seen, the BGE encoder first processes the input text sequence. t i Extract representation E, and then map representation E to predicted values ​​through a linear projection layer. p .

[0061] For example, methods used to adjust model parameters may include backpropagation and / or gradient descent.

[0062] As an example, refer to Figure 5 and Figure 6 This paper illustrates an example of the training and testing process for a training method of a pharmacokinetic parameter prediction model for coagulation factor VIII. Specifically, tabular data and serialized text data are input into the machine learning model and language model, respectively, for training, and the model's prediction results for pharmacokinetic parameters are verified on the test set.

[0063] exist Figure 5 and Figure 6In this context, "Blood Group" refers to blood type, "Height" refers to height, "Body weight" refers to body weight, and "Half Life" refers to half-life. Figure 5 The serialized example expression is: "The BloodGroup is O and The Height is 135 and The Body weight is 30". Figure 6 The serialized example expression is: "The Blood Group is A and The Height is 114 and The Bodyweight is 20".

[0064] Furthermore, for example, ensemble learning models can be trained using unnormalized raw data. For neural network models and language models, backpropagation and / or gradient descent can be used to update model parameters.

[0065] By employing the aforementioned predictive model, two key pharmacokinetic parameters of FVIII (i.e., recovery rate (IVR) and half-life (HL)) in children with hemophilia A (e.g., severe hemophilia A) can be accurately predicted. Furthermore, unlike traditional population pharmacokinetic methods that require intensive sampling and stringent parameter assumptions, the method proposed in this disclosure utilizes only three FVIII activity measurements, combined with demographic and prior dosing information, to predict PK parameters, thereby significantly reducing the amount of blood sample data and thus lowering the blood collection burden on patients.

[0066] The significant effects of the predictive model of this disclosure are illustrated below with reference to Tables 1 and 2. The comparative method for the predictive models of this disclosure (i.e., Lasso, Ridge, Random Forest, XGBoost, MLP, FT-Transformer, and LanguageModel in Tables 1 and 2) (i.e., WAPPS-Hemo in Tables 1 and 2) is the clinically widely used FVIII pharmacokinetic analysis software WAPPS-Hemo. The calculation results of WAPPS-Hemo are derived from FVIII activity in blood samples taken at least six times at 1 hour, 3 hours, 9 hours, 24 hours, 48 ​​hours, and 72 hours. The performance of this invention and the comparative method on a real-world clinical dataset of children with hemophilia A is shown in Tables 1 and 2 below.

[0067] Table 1: Results of the disclosed and comparative methods on the recovery test set.

[0068]

[0069] Table 2: Results of the present disclosure and comparative methods on the half-life test set

[0070]

[0071] In Tables 1 and 2 and Figures 4 to 6 In this context, MAE represents the mean absolute error, RMES represents the root mean square error, and R0 represents the root mean square error. 2 The coefficient of determination is represented by the coefficient of determination. Figures 4 to 6 The P in the table and the P value in Tables 1 and 2 both represent P values.

[0072] As can be seen from the test results in Tables 1 and 2 above, compared with the pharmacokinetic parameters calculated from the corresponding FVIII activity obtained by WAPPS-Hemo using data from six blood collections, this disclosure can achieve accurate prediction of the two important pharmacokinetic parameters, recovery rate and half-life, while significantly reducing the number of blood collections. Furthermore, it has a significant improvement in the predictive effect of FVIII recovery rate and half-life for children with hemophilia A.

[0073] By using the training method described above for the pharmacokinetic parameter prediction model of coagulation factor VIII, accurate prediction of two important PK parameters, including the recovery rate and half-life of FVIII, can be achieved by using background information such as FVIII activity at predetermined blood collection time points and patient dosing records, under the condition of significantly reducing the number of blood collections from patients.

[0074] Furthermore, this disclosure is the first to use multiple machine learning models, including pre-trained language models, to predict the pharmacokinetic parameters of FVIII, demonstrating the feasibility and accuracy of applying machine learning to pharmacokinetic analysis. This is beneficial for assisting clinicians, for example, in developing personalized dosing regimens for patients based on pharmacokinetic principles.

[0075] Furthermore, by using machine learning with strong nonlinear fitting capabilities (e.g., language models), it is possible to capture complex trends, patterns, and contextual relationships in clinical datasets (e.g., large datasets) related to patients, thereby further improving the accuracy of parameter prediction.

[0076] Next, refer to Figure 7 A detailed description of a method 700 for predicting pharmacokinetic parameters of coagulation factor VIII according to embodiments of the present disclosure.

[0077] Figure 7 This is a flowchart illustrating a method 700 for predicting pharmacokinetic parameters of coagulation factor VIII according to an embodiment of the present disclosure.

[0078] Reference Figure 7In step S701, clinical data to be analyzed are obtained for children with hemophilia A (e.g., severe hemophilia A).

[0079] Here, the clinical data to be analyzed includes demographic information associated with the child, medication record data, and data on the activity of factor VIII I in the child's blood before administration and the activity of factor VIII II in the child's blood after administration.

[0080] As an example, the second coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points.

[0081] In step S702, the clinical data to be analyzed is input into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

[0082] As an example, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is trained using the training method 100 for the pharmacokinetic parameter prediction model for coagulation factor VIII as described above.

[0083] It should be noted that the above operations can be referenced. Figure 1 The related content is similar, so I will not repeat it here.

[0084] Next, refer to Figure 8 A detailed description of a training apparatus 800 for a pharmacokinetic parameter prediction model for coagulation factor VIII according to embodiments of the present disclosure.

[0085] Figure 8 This is a structural block diagram illustrating a training device 800 for a pharmacokinetic parameter prediction model for coagulation factor VIII according to an embodiment of the present disclosure.

[0086] Reference Figure 8 The training apparatus 800 for a pharmacokinetic parameter prediction model for coagulation factor VIII according to embodiments of the present disclosure may include: a data acquisition module 810, a data preprocessing module 820, a parameter prediction module 830, and a parameter tuning module 840.

[0087] According to embodiments of this disclosure, the data acquisition module 810 can perform the following: acquire historical clinical data for children with hemophilia A (e.g., severe hemophilia A).

[0088] Here, historical clinical data may include historical demographic information associated with the child, historical medication records, and historical data on the activity of first historical coagulation factor VIII in the child's blood before administration and the activity of second historical coagulation factor VIII in the child's blood after administration.

[0089] As an example, second historical coagulation factor VIII activity data may include coagulation factor VIII activity data collected at three sampling time points. For example, the three sampling time points may include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

[0090] According to an embodiment of this disclosure, the data preprocessing module 820 may perform the following: determine the true labels of pharmacokinetic parameters corresponding to historical clinical data, and store the historical clinical data and the true labels in tabular form as a first training dataset.

[0091] As an example, pharmacokinetic parameters may include recovery rate and half-life.

[0092] According to an embodiment of this disclosure, the parameter prediction module 830 can perform the following: inputting a first training dataset into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

[0093] For example, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a pharmacokinetic parameter prediction model for coagulation factor VIII based on a linear regression algorithm, a pharmacokinetic parameter prediction model for coagulation factor VIII based on a random forest and XGBoost algorithm, a pharmacokinetic parameter prediction model for coagulation factor VIII based on a neural network algorithm, and / or a pharmacokinetic parameter prediction model for coagulation factor VIII based on a pre-trained language algorithm.

[0094] As an example, in the case where the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is a pharmacokinetic parameter prediction model for coagulation factor VIII based on a pre-trained language algorithm, the parameter prediction module 830 inputs the first training dataset into the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters. This operation may include operations 831) and 832).

[0095] In operation 831), a data serialization operation is performed on the first training data to obtain a second training dataset in text form.

[0096] In operation 832), the second training dataset is input into the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

[0097] Furthermore, in the example, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII may include a BERT-based BGE encoder and a linear projection layer.

[0098] For example, methods used to adjust model parameters may include backpropagation and / or gradient descent.

[0099] According to an embodiment of this disclosure, the parameter tuning module 840 can perform the following: adjust the model parameters of a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII based on the truth label and the predicted value, to obtain a trained pharmacokinetic parameter prediction model for coagulation factor VIII.

[0100] It should be noted that the operations performed on the above structural frames can be compared with those in the reference section. Figure 1 The related content is similar, so I will not repeat it here.

[0101] Next, refer to Figure 9 A detailed description of a device 900 for predicting pharmacokinetic parameters of coagulation factor VIII according to an embodiment of the present disclosure.

[0102] Figure 9 This is a structural block diagram illustrating a device 900 for predicting pharmacokinetic parameters of coagulation factor VIII according to an embodiment of the present disclosure.

[0103] Reference Figure 9 The pharmacokinetic parameter prediction device 900 for coagulation factor VIII according to embodiments of the present disclosure may include an acquisition module 910 and a prediction module 920.

[0104] According to embodiments of this disclosure, the acquisition module 910 can perform the following: acquire clinical data to be analyzed for children with hemophilia A (e.g., severe hemophilia A).

[0105] For example, the clinical data to be analyzed includes demographic information associated with the child, medication records, and data on the activity of factor VIII 1 in the child's blood before administration and the activity of factor VIII 2 in the child's blood after administration.

[0106] According to embodiments of this disclosure, the second coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points.

[0107] According to an embodiment of this disclosure, the prediction module 920 can perform the following: input the clinical data to be analyzed into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII, and obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

[0108] As an example, the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is trained using the training method for the pharmacokinetic parameter prediction model for coagulation factor VIII described above.

[0109] It should be noted that the operations performed on the above structural frames can be compared with those in the reference section. Figure 7 The related content is similar, so I will not repeat it here.

[0110] Figure 10 This is a block diagram illustrating a computing device 1000 according to an embodiment of the present disclosure.

[0111] Reference Figure 10 The computing device 1000 according to embodiments of the present disclosure may include a processor 1010 and a memory 1020. The processor 1010 may include (but is not limited to) a central processing unit (CPU), a digital signal processor (DSP), a microcomputer, a field-programmable gate array (FPGA), a system-on-a-chip (SoC), a microprocessor, an application-specific integrated circuit (ASIC), etc. The memory 1020 may store computer-executable instructions to be executed by the processor 1010. The memory 1020 includes high-speed random access memory and / or a non-volatile computer-readable storage medium. When the processor 1010 executes the computer-executable instructions stored in the memory 1020, the training method for the pharmacokinetic parameter prediction model for coagulation factor VIII as described above and the prediction method for the pharmacokinetic parameters of coagulation factor VIII as described above can be implemented.

[0112] The training method for the pharmacokinetic parameter prediction model of coagulation factor VIII and the prediction method for the pharmacokinetic parameters of coagulation factor VIII according to embodiments of this disclosure can be written as a computer program / instructions to form a computer program product and stored on a computer-readable storage medium. When the computer program / instructions are executed by a processor, the training method for the pharmacokinetic parameter prediction model of coagulation factor VIII and the prediction method for the pharmacokinetic parameters of coagulation factor VIII as described above can be implemented. When the instructions in the computer-readable storage medium are executed by a processor of an electronic device / server, the electronic device / server is enabled to execute the training method for the pharmacokinetic parameter prediction model of coagulation factor VIII and the prediction method for the pharmacokinetic parameters of coagulation factor VIII as described above. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disc storage, hard disk drive (HDD), solid-state drive (SSD), card storage (such as multimedia cards, secure digital (SD) cards, or ultra-fast digital (XD) cards), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, and any other device configured to store computer programs and any associated data, data files, and data structures in a non-transitory manner and to provide the computer programs and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the computer programs. In one example, the computer programs and any associated data, data files, and data structures are distributed across a networked computer system, such that the computer programs and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner through one or more processors or computers.

[0113] According to the embodiments of this disclosure, the training method and apparatus for predicting the pharmacokinetic parameters of coagulation factor VIII, and the prediction method and apparatus for the pharmacokinetic parameters of coagulation factor VIII, through the machine learning-based prediction method for the pharmacokinetic parameters of FVIII in hemophilia A children and the corresponding model training method, under the condition of significantly reducing the number of blood collections from patients, by using background information such as FVIII activity at predetermined blood collection time points and patient dosing records, can accurately predict two important PK parameters, including the recovery rate and half-life of FVIII.

[0114] On the other hand, the training method and apparatus for predicting pharmacokinetic parameters of coagulation factor VIII, and the method and apparatus for predicting pharmacokinetic parameters of coagulation factor VIII, according to embodiments of the present disclosure, provide beneficial assistance for subsequent PK assessment and personalized dosing regimens by achieving accurate prediction of PK parameters.

[0115] On the other hand, the training method and apparatus for predicting the pharmacokinetic parameters of coagulation factor VIII, and the method and apparatus for predicting the pharmacokinetic parameters of coagulation factor VIII, according to embodiments of the present disclosure, provide a data-efficient, accurate, and clinically applicable alternative for the designation of individualized treatment plans by combining advanced artificial intelligence methods with real pediatric datasets. This is also the first time that a language model has been applied to the pharmacokinetic prediction of severe hemophilia A in children.

[0116] While some embodiments of this disclosure have been disclosed and described, those skilled in the art will understand that modifications and variations may be made to these embodiments without departing from the concept and spirit of this disclosure, which is defined by the claims and their equivalents.

Claims

1. A training method for a pharmacokinetic parameter prediction model for coagulation factor VIII, characterized in that, The training method includes: Acquire historical clinical data for children with hemophilia A, including historical demographic information associated with the child, historical medication records, and historical factor VIII activity data in the child's blood before administration and in the child's blood after administration. Determine the truth labels of the pharmacokinetic parameters corresponding to the historical clinical data, and store the historical clinical data and the truth labels in tabular form as a first training dataset; The first training dataset is input into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters. The model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII are adjusted based on the true value label and the predicted value to obtain the trained pharmacokinetic parameter prediction model for coagulation factor VIII. The second set of historical coagulation factor VIII activity data includes only coagulation factor VIII activity data collected at three sampling time points. Among them, pharmacokinetic parameters include recovery rate and half-life, and The three sampling time points include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

2. The training method according to claim 1, characterized in that, The predetermined pharmacokinetic parameter prediction models for coagulation factor VIII include pharmacokinetic parameter prediction models for coagulation factor VIII based on linear regression algorithms, pharmacokinetic parameter prediction models for coagulation factor VIII based on random forest and XGBoost algorithms, pharmacokinetic parameter prediction models for coagulation factor VIII based on neural network algorithms, and / or pharmacokinetic parameter prediction models for coagulation factor VIII based on pre-trained language algorithms.

3. The training method according to claim 1, characterized in that, The predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is a pre-trained language algorithm-based pharmacokinetic parameter prediction model for coagulation factor VIII, and the step of inputting the first training dataset into the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters includes: Perform a data serialization operation on the first training data to obtain a second training dataset in text form; The second training dataset is input into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters.

4. The training method according to claim 1, characterized in that, The predetermined pharmacokinetic parameter prediction model for coagulation factor VIII includes a BERT-based BGE encoder and a linear projection layer, and Specifically, the backpropagation algorithm and / or gradient descent algorithm are used to adjust the model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII based on the truth labels and the predicted values.

5. A method for predicting pharmacokinetic parameters of coagulation factor VIII, characterized in that, The prediction method includes: Acquire clinical data to be analyzed for children with hemophilia A, including demographic information data associated with the children, medication record data, and data on the activity of the first coagulation factor VIII in the children's blood before administration and the activity of the second coagulation factor VIII in the children's blood after administration. The clinical data to be analyzed is input into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII to obtain the predicted values ​​of the corresponding pharmacokinetic parameters. The second coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points. The predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is trained using the training method for the pharmacokinetic parameter prediction model for coagulation factor VIII as described in any one of claims 1 to 4. Among them, pharmacokinetic parameters include recovery rate and half-life, and The three sampling time points include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

6. A training device for a pharmacokinetic parameter prediction model of coagulation factor VIII, characterized in that, The training device includes: The data acquisition module is configured to acquire historical clinical data for children with hemophilia A, including historical demographic information data associated with the child, historical drug administration records, and historical data on the activity of the first historical coagulation factor VIII in the child's blood before drug administration and the activity of the second historical coagulation factor VIII in the child's blood after drug administration. The data preprocessing module is configured to: determine the truth labels of the pharmacokinetic parameters corresponding to the historical clinical data, and store the historical clinical data and the truth labels in tabular form as a first training dataset; The parameter prediction module is configured to: input the first training dataset into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII, and obtain the predicted values ​​of the corresponding pharmacokinetic parameters. The parameter tuning module is configured to: adjust the model parameters of the predetermined pharmacokinetic parameter prediction model for coagulation factor VIII based on the truth label and the predicted value, thereby obtaining the trained pharmacokinetic parameter prediction model for coagulation factor VIII. The second historical coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points. The predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is trained using the training method for the pharmacokinetic parameter prediction model for coagulation factor VIII as described in any one of claims 1 to 4. Among them, pharmacokinetic parameters include recovery rate and half-life, and The three sampling time points include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

7. A device for predicting pharmacokinetic parameters of coagulation factor VIII, characterized in that, The prediction device includes: The acquisition module is configured to acquire clinical data to be analyzed for children with hemophilia A, including demographic information data associated with the children, medication record data, and data on the activity of the first coagulation factor VIII in the children's blood before administration and the activity of the second coagulation factor VIII in the children's blood after administration. The prediction module is configured to: input the clinical data to be analyzed into a predetermined pharmacokinetic parameter prediction model for coagulation factor VIII, and obtain the predicted values ​​of the corresponding pharmacokinetic parameters. The second coagulation factor VIII activity data includes coagulation factor VIII activity data collected at three sampling time points. The predetermined pharmacokinetic parameter prediction model for coagulation factor VIII is trained using the training method for the pharmacokinetic parameter prediction model for coagulation factor VIII as described in any one of claims 1 to 4. Among them, pharmacokinetic parameters include recovery rate and half-life, and The three sampling time points include 1 hour after administration, 3 hours after administration, and 24 hours after administration.

8. A computer program product, characterized in that, The computer program product includes a computer program / instruction that, when executed by a processor, implements the training method for the pharmacokinetic parameter prediction model for coagulation factor VIII as described in any one of claims 1 to 4, and the method for predicting pharmacokinetic parameters for coagulation factor VIII as described in claim 5.

9. A computing device, characterized in that, The computing device includes: at least one processor; at least one memory storing computer-executable instructions, wherein, when executed by the at least one processor, the computer-executable instructions cause the at least one processor to execute the training method for a pharmacokinetic parameter prediction model for coagulation factor VIII as claimed in any one of claims 1 to 4 and the prediction method for pharmacokinetic parameters of coagulation factor VIII as claimed in claim 5.

Images