Estimation of Pharmacokinetic Parameters Using Deep Learning
Using machine learning and deep learning algorithms, specifically neural networks, addresses the inefficiencies and inaccuracies of traditional pharmacokinetic modeling by accurately predicting parameters like AUC and Cmax, enhancing predictive capabilities.
Patent Information
- Application Number
- JP2022557912
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2020-03-23
- Filing Date
- 2021-03-23
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2041-03-23
AI Technical Summary
Traditional mathematical modeling methodologies for pharmacokinetics and pharmacodynamics require significant computational resources and expert knowledge, making them time-consuming and labor-intensive, and existing methods for estimating pharmacokinetic parameters, such as the trapezoidal rule, are inaccurate due to dependence on data completeness and critical assumptions.
Employing machine learning or deep learning algorithms, particularly neural networks, to predict pharmacokinetic parameters by training on simulated data sets, allowing for improved accuracy in estimating parameters like AUC and Cmax through learning underlying data patterns.
The deep learning approach significantly enhances the accuracy of pharmacokinetic parameter estimation, reducing errors and improving predictive capabilities compared to conventional methods.
Smart Images

Figure 0007577756000001 
Figure 0007577756000002 
Figure 0007577756000003
Abstract
Description
[Technical field]
[0001] Field This description relates generally to systems and methods for estimating or predicting the pharmacokinetic properties of therapeutic agents. More specifically, there is a need for machine learning based systems and methods for accurately estimating or predicting the pharmacokinetic properties of therapeutic agents administered to a patient. [Background technology]
[0002] background The development of new drugs (e.g., therapeutic drugs) is driven by advances in many fields. Such fields include drug discovery, biotechnology, and in vivo and in vitro pharmacological / toxicological characterization techniques. Before a new therapeutic drug goes from a molecule or protein in the laboratory to a new product in the hospital / clinic or local pharmacy, various questions need to be answered regarding the efficacy, administration, safety, and side effects associated with the therapeutic drug. Answering these types of questions typically involves a series of clinical trials carefully designed to study various aspects of the new drug candidate.
[0003] Pharmacokinetics (PK) and pharmacodynamics (PD) are scientific fields related to therapeutic development that typically involve mathematical modeling. In general terms, PK (or pK) is described as "what the body does to the drug," and PD (or pD) is often described as "what the drug does to the body." More specifically, PK can focus on modeling how the body acts on a drug once it has been administered and subjected to the four bodily processes of absorption, distribution, metabolism, and excretion or elimination (ADME). This can be achieved by modeling the body concentrations in various regions of the body in general or as a function of time. PD aims to link these modeled drug concentrations to specific drug effects via PD models specifically designed to evaluate those effects. Thus, PK / PD modeling can link systemic drug concentration dynamics to the resulting drug effects over time. Such modeling allows for the description and prediction of the time course of various physiological effects (e.g., tumor cell counts, platelet counts, neutrophil counts, etc.) in response to various dosing regimens.
[0004] Traditional mathematical modeling methodologies for PK / PD evaluation can require iterative model evaluation and refinement, with human judgment involved at various steps in the loop. This can be time and labor intensive. Examples of such existing mathematical algorithms include expectation maximization, genetic algorithms, and scatter search. These techniques may be optimization-based, which can mean in practice that the scientist creating the model performs many function and gradient evaluations involving significant trial and error. Thus, effectively using these existing mathematical techniques to model PK and PD requires a significant amount of know-how and computational time. The know-how prerequisites and computational resource requirements represent significant obstacles along the path towards widespread adoption of PK, PD and PK / PD modeling for non-expert users. Summary of the Invention
[0005] overview In one or more embodiments, a method is provided for predicting at least one pharmacokinetic parameter of a drug administered to a subject. The one or more processors train a neural network based on a simulated training data set by the one or more processors. The simulated training data set includes a simulated time series concentration data set and a simulated value for the pharmacokinetic parameter corresponding to the simulated time series concentration data set. The one or more processors receive a time series concentration data set of the drug obtained from the subject. The one or more processors predict a value for the pharmacokinetic parameter using the time series concentration data set and the trained neural network.
[0006] In one or more embodiments, a system for predicting at least one pharmacokinetic parameter of a drug administered to a subject is provided. The system comprises a data store for storing a time series concentration dataset of the drug obtained from the subject, a computing device communicatively connected to the data store, and a display system communicatively connected to the computing device. The computing device comprises a pharmacokinetic data simulation engine and a pharmacokinetic prediction engine. The pharmacokinetic data simulation engine is configured to generate a simulated training dataset of the one or more drugs. The simulated training dataset includes a simulated time series concentration dataset and simulated values for the pharmacokinetic parameters corresponding to the simulated time series concentration dataset. The pharmacokinetic prediction engine is configured to train a neural network using the simulated training dataset and predict values for the pharmacokinetic parameters based on the time series concentration dataset using the trained neural network. The display system is configured to display a report including the predicted values for the pharmacokinetic parameters.
[0007] In one or more embodiments, a method is provided for training a neural network to predict at least one pharmacokinetic parameter of a drug. The one or more processors receive simulation parameters of a compartment-based pharmacokinetic model. The one or more processors generate a simulated training data set using the compartment-based pharmacokinetic model. The simulated training data set includes simulated time series concentration data sets of the one or more drugs and simulated values for the pharmacokinetic parameters corresponding to the simulated time series concentration data sets. The compartment-based pharmacokinetic model is encoded by the received simulation parameters. The one or more processors then train the neural network with the simulated training data set.
[0008] In one or more embodiments, a system for training a neural network to predict pharmacokinetic parameter values of a drug is disclosed. The system includes a data store, a computing device, and a display system. The data store stores simulation parameters of a compartment-based pharmacokinetic model. The computing device is communicatively connected to the data store and hosts a pharmacokinetic data simulation engine and a pharmacokinetic prediction engine. The pharmacokinetic data simulation engine is configured to generate a simulated training data set using the compartment-based pharmacokinetic model. The simulated training data set includes simulated time-series concentration data sets of one or more drugs and simulated values for pharmacokinetic parameters corresponding to the simulated time-series concentration data sets. The pharmacokinetic prediction engine is configured to train the neural network using the simulated training data collection.
[0009] In one or more embodiments, a method is provided for predicting at least one pharmacokinetic parameter of a drug administered to a subject. One or more processors receive a time series concentration dataset of the drug obtained from the subject. The one or more processors predict a value for the pharmacokinetic parameter based on the time series concentration dataset and a neural network trained using a simulated training data set.
[0010] In one or more embodiments, a system for predicting at least one pharmacokinetic parameter of a drug administered to a subject is provided. The system includes a data store for storing a time series concentration dataset of the drug obtained from the subject, a computing device communicatively connected to the data store, and a display system communicatively connected to the computing device. The computing device is configured to train a neural network using a simulated training dataset and predict a value for the pharmacokinetic parameter based on the time series concentration dataset and the trained neural network. The display system is configured to display a report including the predicted value for the pharmacokinetic parameter. [Brief description of the drawings]
[0011] For a more complete understanding of the principles disclosed herein, and the advantages thereof, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, in which:
[0012] [Figure 1] 1 is a visualization graph showing a standard concentration time series plot 100, illustrating examples of exemplary PK parameters.
[0013] [Diagram 2] FIG. 1 is a plot of concentration over time.
[0014] [Diagram 3] FIG. 3 is a diagram of the workflow of the trapezoidal sum calculation performed on the time series concentration plot of FIG. 2.
[0015] [Figure 4] FIG. 4 is a block diagram of a pharmacokinetic (PK) evaluation system 400 in accordance with one or more exemplary embodiments.
[0016] [Diagram 5] 1 is a depiction of an exemplary NN system and a method of using it to process time-series concentration data to predict PK parameters, according to various embodiments.
[0017] [Figure 6] FIG. 1 illustrates an exemplary high-level workflow of how a NN model is trained during a learning phase and can then implement what it has learned to predict PK parameters, according to various embodiments.
[0018] [Figure 7] FIG. 13 is a plot illustrating percent validation error in accordance with an example embodiment.
[0019] [Figure 7] 1 is a schematic diagram of a system for predicting PK parameter values of an agent (e.g., a drug, a biologic, etc.) administered to a subject (i.e., a patient), according to various embodiments.
[0020] [Figure 8] FIG. 11 is a plot illustrating an example of training and validation loss as a function of the number of training iterations, according to an example embodiment;
[0021] [Figure 9] FIG. 1 illustrates an exemplary training workflow in accordance with various embodiments.
[0022] [Figure 10] 1A-1C are a series of plots of various forms of compartment modeling in accordance with various embodiments.
[0023] [Figure 11]1 shows a set of plots illustrating improved accuracy of estimating AUC and Cmax using the neural network approach described herein compared to the traditional trapezoidal rule in accordance with an exemplary embodiment.
[0024] [Figure 10] 1 is an exemplary flow chart illustrating a method for predicting a PK parameter value of an agent (e.g., a drug, a biologic, etc.) administered to a subject (i.e., a patient), according to various embodiments.
[0025] [Figure 11] 1 is a plot showing improved accuracy of AUC and Cmax estimation using the novel NN approach versus the traditional trapezoidal rule, according to various embodiments.
[0026] [Figure 12] 1 is an exemplary flow chart illustrating a method for predicting a PK parameter value of an agent (i.e., a drug or a biologic) administered to a subject (i.e., a patient), according to various embodiments.
[0027] [Figure 13] 1 is an exemplary flow chart illustrating a method for training a neural network to predict a pharmacokinetic parameter value of a pharmaceutical agent (i.e., a drug or a biologic), according to various embodiments.
[0028] [Figure 14] 1 is an exemplary flow chart illustrating a method for predicting a PK parameter value of an agent (e.g., a drug, a biologic, etc.) administered to a subject (i.e., a patient), according to various embodiments.
[0029] [Figure 15] FIG. 1 is a block diagram illustrating a computer system upon which an embodiment of the present teachings may be implemented.
[0030] It should be understood that the drawings are not necessarily drawn to scale, and that objects within the drawings are not necessarily drawn to scale relative to each other. The drawings are representations intended to bring clarity and understanding to the various embodiments of the apparatus, systems, and methods disclosed herein. Wherever possible, the same reference numbers are used throughout the drawings to refer to the same or similar parts. Furthermore, it should be understood that the drawings are not intended to limit the scope of the present teachings in any way. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0031] Detailed Description I. Overview Pharmacokinetic / pharmacodynamic (PK / PD) principles have become a well-established quantitative framework for understanding the dose-concentration-effect relationships of various therapeutic agents and for selecting appropriate protocols (e.g., doses, schedules, etc.) for such therapeutic agents. Current methods for estimating pharmacokinetic properties (e.g., as described in Section III below) are basic. For example, some currently available methods use non-compartmental analysis (NCA) techniques that numerically infer pharmacokinetic parameters. One such method estimates the area under the curve (AUC) formed by the amount of therapeutic agent in a subject's blood over time. The AUC can be calculated using the trapezoidal rule, where the AUC is calculated as a sum of trapezoids. These types of techniques are simple and fast.
[0032] However, such currently available methodologies may be inaccurate in various circumstances because their accuracy depends on important assumptions or assumptions. These assumptions include, for example, that all relevant measurements are available and that the time intervals between measurements are appropriately short and condensed. Thus, approximation of AUC using the trapezoidal rule may pose limitations in terms of the accuracy of estimating pharmacokinetic parameters. Furthermore, the estimation variability of the linear trapezoidal rule is strongly dependent on data completeness. A large amount of missing values leads to a decrease in estimation accuracy when missing provisional values are imputed by linear interpolation.
[0033] Thus, recognizing and taking into consideration the above-mentioned problems, this specification describes various exemplary embodiments of systems and methods for accurately estimating or predicting the pharmacokinetic properties of a therapeutic drug administered to a patient. However, it should be understood that while the systems and methods disclosed herein make specific reference to their application in pharmacokinetics, they are equally applicable to other similar fields, such as toxicokinetics and toxicodynamics.
[0034] The embodiments described herein provide methods and systems for predicting pharmacokinetic parameters with a higher level of accuracy when compared to conventional methods. The use of machine learning (ML) or deep learning (DL) helps enable this improved accuracy through pharmacokinetic modeling. For example, in deep learning systems (e.g., including one or more neural network models), the prediction of pharmacokinetic parameters is driven by learning features of the underlying data. Furthermore, deep learning systems allow for the exploitation of local patterns in the data to make inferences. Furthermore, in deep learning systems, the prediction accuracy of the deep learning system increases as the amount of data used to train the deep learning system increases.
[0035] In various embodiments, one or more mathematical models are applied to generate simulated data, and the simulated PK data of one or more existing or potential future agents (e.g., chemotherapeutic agents or toxins, biotherapeutic agents or toxins, etc.) is used to train one or more machine learning (ML) or deep learning (DL) algorithms (e.g., convolutional neural networks, etc.) to generate predictions or to measure one or more PK parameters (e.g., AUC, C max , T max , t 1 / 2 Methods are provided for providing estimates of
[0036] In various embodiments, a system for predicting or estimating PK parameters can include a data store for storing measured (or "real-world" / clinical) PK data and one or more computers / servers capable of hosting and executing software code including a PK data simulation engine and a PK prediction engine. The PK data simulation engine can be configured to apply one or more mathematical models to simulate PK data that can be used to train an ML or DL algorithm. The PK prediction engine can be configured to use the simulated PK data to train an ML or DL algorithm to predict or estimate PK parameters using the measured PK data stored in the data store. II. Definition
[0037] However, the disclosure is not limited to these exemplary embodiments and applications or the manner in which the exemplary embodiments and applications operate or are described herein. Additionally, the figures may show simplified or partial views, and the dimensions of the elements in the figures may be exaggerated or out of proportion. Additionally, when the terms "on", "attached to", "connected to", "coupled to" or similar terms are used herein, an element (e.g., a material, layer, substrate, etc.) can be "on", "attached to", "connected to", or "coupled to" another element, regardless of whether the element is directly on, directly attached to, connected to, or coupled to the other element, or whether there are one or more intervening elements between the element and the other element. Furthermore, when a reference is made to a list of elements (e.g., elements a, b, c), such reference is intended to include any one of the listed elements by itself, any combination of less than all of the listed elements, and / or all combinations of the listed elements. The division of sections herein is merely for ease of review and is not intended to limit any combination of the elements described.
[0038] Unless otherwise defined, scientific and technical terms used in connection with the present teachings described herein shall have the meanings commonly understood by those skilled in the art.Furthermore, unless otherwise required by context, singular terms shall include the plural and plural terms shall include the singular.In general, the nomenclature and techniques utilized in connection with chemistry, biochemistry, molecular biology, pharmacology and toxicology are described herein and are well known and commonly used in the art.
[0039] As used herein, "substantially" means sufficient to function for the intended purpose. Thus, the term "substantially" allows for minor, slight variations from an absolute or perfect state, dimension, measurement, result, etc., that would be expected by one of ordinary skill in the art, but that do not significantly affect overall performance. When used in reference to a numerical value, or a parameter or characteristic that can be expressed as a numerical value, "substantially" means within 10 percent.
[0040] The term "plurality" means two or more.
[0041] As used herein, the term "plurality" can be 2, 3, 4, 5, 6, 7, 8, 9, 10 or more.
[0042] As used herein, the phrase "area under the curve" (AUC) can refer to the area under the curve that represents the variation in therapeutic (drug) concentration in the plasma of a subject as a function of time after administration.
[0043] As used herein, "maximum concentration" (C max The phrase "maximum (or peak) serum concentration" can refer to the maximum (or peak) serum concentration that a therapeutic agent (drug) achieves in a particular compartment or test area of the body after the drug is administered and before a second dose is administered.
[0044] As used herein, "minimum concentration" (C min ) or "trough concentration" (C trough The phrase "minimum (or peak) serum concentration" can refer to the minimum (or peak) serum concentration that a therapeutic agent (drug) achieves in a particular compartment or test area of the body after the drug is administered and before a second dose is administered.
[0045] As used herein, the "time to maximum concentration" (T max The term ) can refer to the time to reach maximum concentration after a therapeutic (drug) dose is administered before the next dose is administered.
[0046] As used herein, "half-life" (t 1 / 2 ) can refer to the time it takes for the concentration of a therapeutic agent (drug) in plasma to reach half of its steady-state value.
[0047] As used herein, the term "mean residence time" (MRT) can refer to the average time that a therapeutic agent (drug) remains at a site of action.
[0048] As used herein, the phrase "compartment model" or "compartment modeling" can refer to a mathematical modeling technique used to predict the PK parameters (indicative of ADME) of synthetic or natural therapeutic agents (drugs) in test subjects by modeling the concentration of the therapeutic agent in different regions of the body. Within these mathematical models, different regions of the body can be divided into portions called compartments in which the therapeutic agent can be assumed to act in the same way.
[0049] As used herein, the phrase "non-compartmental model" (NCA) can refer to a model-independent technique (meaning that it does not rely on assumptions regarding body compartments) used to predict the PK parameters (indicative of ADME) of a therapeutic agent (drug) administered to a test subject. NCA allows for the calculation of the PK parameters of a therapeutic agent (drug) from the measured drug concentration time course.
[0050] As used herein, "biologic" or "macrotherapeutic polymer" can refer to proteins and other biological macromolecules that have a therapeutic effect.
[0051] As used herein, a "therapeutic compound" or "small molecule therapeutic" can refer to any organic compound that affects biological processes and has a relatively low molecular weight of less than 900 daltons.
[0052] As used herein, "artificial neural network" or "neural network" (NN) can refer to a mathematical algorithm or computational model. A neural network can predict an output of a received input using one or more layers of nonlinear units. Some neural networks include one or more hidden layers in addition to an output layer. The output of each hidden layer is used as an input to the next layer in the network, i.e., the next hidden layer or the output layer. Each layer of the network generates an output from the received input according to the current values of each parameter set.
[0053] A neural network can process information in two ways. When it is being trained, it may be in "learning mode", and when it actually uses what it has learned, it may be in "inference (or prediction) mode". A neural network can learn through a feedback process called backpropagation, which allows the network to adjust the weight coefficients of individual nodes in the intermediate hidden layers (modify its behavior) so that the output matches that of the training data. In other words, a neural network can receive training data (training examples) and automatically learn how to arrive at the correct output, even when presented with a new range or set of inputs. Examples of types of neural networks include, but are not limited to: Feedforward Neural Networks (FNNs), Recurrent Neural Networks (RNNs), Modular Neural Networks (MNNs), Convolutional Neural Networks (CNNs), ResNet (Residual Neural Networks), etc. III. Traditional methods for predicting PK parameters
[0054] FIG. 1 is a visualization graph showing a standard time series concentration plot 100, illustrating examples of typical PK parameters. As shown herein, various types of PK parameters (e.g., C max 102, t 1 / 2 104, C min106, AUC108, etc.) are shown in a time series concentration plot 100 for the drug after the drug is administered to a test subject. max 102 is the value that indicates the maximum concentration of the drug in the subject's blood. max 102 and this maximum value T max It is often important to know the time it takes to reach the T. With an intravenous bolus dose, the maximum concentration is achieved immediately after the drug is injected into the bloodstream, i.e., the T max = 0. For parenteral (oral) administration, the concentration does not peak until some time later due to an absorption step. In general, the time it takes to reach a maximum value can be calculated by differentiating the expression for C(t) of the system with respect to t, setting the derivative to 0, and finally calculating T max It can be found by solving
[0055] In plot 100, t 1 / 2 104 or half-life represents the biological half-life of a drug in a subject after administration. That is, the half-life of a drug is the time it takes for the drug to lose half of its maximum concentration. min 106 is the minimum concentration of drug in the blood after administration. This PK parameter is often used in bioavailability and bioequivalence studies.
[0056] AUC108 or "area under the curve" is a value that indicates the overall therapeutic drug exposure of a subject. AUC108 can be evaluated using most types of models (compartmental or non-compartmental) and can be determined graphically based on a time series of concentration measurements.
[0057] 2 is an illustration of a time series concentration plot 200. As shown herein, the time series concentration plot 200 includes time series concentration data 202 measured from blood samples taken from a test subject at various time intervals occurring after administration of a drug. The AUC 204 of the time series concentration plot 200 consists of a series that can be estimated using the conventional "trapezoidal rule," where the AUC 204 below the time series concentration plot 200 is mathematically subdivided into a series of AUC segments 208 that are trapezoids bracketed by the time intervals over which the drug concentrations are measured. The AUC 204 is calculated as a trapezoidal sum calculation.
[0058] FIG. 3 is a diagram of a workflow 300 of a trapezoidal sum calculation performed on the time series concentration plot 200 of FIG. 2. Data 302 includes time points at which concentration measurements were made. Plot 304 represents a trapezoid-based representation of the time series concentration plot 200 of FIG. 2, in which the AUC 204 is divided into a series of trapezoids. Equation 306 identifies one way in which the trapezoidal sum calculation can be performed to provide an estimate of the AUC. This AUC estimate is therefore based on an approximation of the total area by summing the individual subareas. This method of numerical integration generally involves the summing of the last observations in the n-1 sections, t n Assume that the trajectory is divided into
[0059] Although this conventional approach can be simple and fast, its accuracy depends on the critical assumption that all relevant measurements are available and that the time intervals between measurements are suitably short and condensed. Thus, as mentioned above, this type of approach has limitations with respect to its accuracy. IV. Estimation of PK parameters using deep learning neural networks
[0060] In contrast to the imprecise methods described in Section III, various embodiments of the present disclosure provide methods and systems for accurately predicting one or more pharmacokinetic parameter values of a drug administered to a subject. These methods and systems can be implemented via computer software, hardware, firmware, or combinations thereof.
[0061] 4 is a block diagram of a pharmacokinetic (PK) evaluation system 400 according to one or more exemplary embodiments. The PK evaluation system 400 can be used to evaluate the PK effect resulting from administration of a drug (e.g., a therapeutic drug). In various embodiments, the PK evaluation system 400 is trained based on observed data and then used to predict the PK effect over time (including times beyond the time that the observed data is provided or available). As previously mentioned, the PK effect is the dose effect or the body's effect on the drug (e.g., drug concentration) in the body.
[0062] The PK evaluation system 400 can be used in a variety of environments, including, but not limited to, a research environment, a clinical trial environment, a drug development environment, a hospital environment, or some other type of environment. The PK evaluation system 400 can receive and process input data 401 to generate a report 402 describing and / or including these PK-based information. The input data 401 can include, for example, laboratory measurements obtained from the subject's plasma, measurements from a continuous monitoring device (e.g., in a hospital or home environment), or some other type of data. The report 402 can include, for example, a predicted PK time course for a particular number of days or weeks into the future based on a current dosing regimen. In some cases, the report 402 can also include at least one of a corresponding PK time course predicted based on a particular dosing interruption and / or adjustment to the current dosing regimen. The report 402 can be generated for an individual subject of interest or for an entire cohort of subjects in a clinical trial. In one or more examples, the report 402 can include one or more recommended actions based on the predicted PK time course.
[0063] The PK evaluation system 400 includes a computing platform 403, a data storage 404, and a display system 406. The computing platform 403 can take a variety of forms. In one or more embodiments, the computing platform 403 includes a single computer (or computer system) or multiple computers in communication with each other. In other examples, the computing platform 403 takes the form of a cloud computing platform. In some embodiments, the computing platform 403 is referred to as a computing device / analysis server. In some embodiments, the computing platform 403 can be a workstation, a mainframe computer, a distributed computing node (part of a "cloud computing" or distributed networking system), a personal computer, a mobile device, or the like.
[0064] The data storage 404 and the display system 406 are each in communication with the computing platform 403. The data storage 404 can be communicatively connected to the computing platform 403 in a variety of ways, such as, but not limited to, a network connection, which can be either a "hardwired" physical network connection (e.g., Internet, LAN, WAN, VPN, etc.) or a wireless network connection (e.g., Wi-Fi, WLAN, etc.).
[0065] In some examples, the data storage 404, the display system 406, or both may be considered part of or otherwise integrated with the computing platform 403. Thus, in some examples, the computing platform 403, the data storage 404, and the display system 406 may be separate components that communicate with one another, while in other examples some combination of these components may be integrated together.
[0066] The PK evaluation system 400 includes a data manager 408 and a prediction system 410 implemented on a computing platform 403. Each of the data manager 408 and the neural network system 410 is implemented using hardware, software, firmware, or a combination thereof. The data manager 408 may also be referred to as a pharmacokinetic (PK) data simulation engine. The prediction system 410 may also be referred to as a pharmacokinetic (PK) prediction engine.
[0067] It should be understood that the various components or engines of the PK assessment system 400 may be combined or collapsed into a single engine, component, or module depending on the requirements of a particular application or system architecture. Additionally, in various embodiments, the data storage 404, the data manager 408, the prediction system 410, or a combination thereof, may include additional engines or components as required by a particular application or system architecture.
[0068] In various embodiments, the data manager 408 provides input data 401 to the neural network system 410. This input data 401 may be obtained from the data storage 404, may be received from some other source, or may be a combination thereof. The input data 401 may have been generated via one or more samples from one or more subjects. For example, the sample analyzer 420 may be communicatively connected to the data storage 404 via one or more communication links (e.g., via a serial bus, a wireless network connection, etc.). The sample analyzer 420 may be configured to analyze a blood sample from a test subject, such as test subject 422, to determine a concentration of an agent (drug or biologic) in the blood. The time series concentration data may be generated as blood samples are drawn from the subject 422 at set time intervals and analyzed using the sample analyzer 420. In various embodiments, the time series concentration data may be stored in the data storage 404 for subsequent processing. In some embodiments, the time series concentration data is provided to the data manager 408 in real time or near real time as the input data 401.
[0069] The prediction system 410 is a machine learning or deep learning prediction system. In one or more embodiments, the prediction system 410 includes a neural network system 411. The neural network system 411 includes one or more neural network models.
[0070] When the neural network system 410 is being trained, the input data 401 takes the form of training data 416. After training, the neural network system 410 can actually be used to predict PK parameters, and in these instances the input data 401 may be referred to as test data 418. Thus, the type of input data 401 provided to the neural network system 410 can take different forms depending on whether the neural network system 410 is in training or prediction mode.
[0071] For example, the neural network system 410 can include a training module 412 and a prediction module 414. The training module 412 can be used when the neural network system 410 is being trained (e.g., in a training mode). As an example, the training module 412 trains the neural network system 410 using training data 416 received from the data manager 408. The prediction module 414 uses the neural network system 410 for prediction (e.g., in a prediction mode). For example, after the neural network system 410 is trained, the prediction module 414 can be actually used to predict the PK parameters 424 using the test data 418 received from the data manager 408.
[0072] In some embodiments, the data manager 408 simulates the training data 416 used by the prediction system 410. For example, the data manager 408 may be configured to use one or more compartment-based pharmacokinetic models to generate the training data 416 (e.g., a simulated training data set). The training data 416 (simulated training data set) can be comprised of multiple individual simulated time series concentration data sets of one or more modeled agents (drugs or biologics) and simulated values for one or more PK parameters corresponding to the multiple individual simulated time series concentration data sets. Examples of types of simulated pharmacokinetic parameters include, but are not limited to, AUC, C max , C min , C trough , T max , MRT, T last , or t 1 / 2 Includes.
[0073] In various embodiments, a compartment-based pharmacokinetic model can be custom configured by inputting one or more compartment model parameters prior to running a model simulation to generate simulated time-series concentration data. Examples of compartmental model parameters include, but are not limited to, administration scheme (e.g., subcutaneous, oral, intravenous, etc.), molecular class (i.e., small or large molecule), molecular species (e.g., drugs, mAbs, proteins, enzymes, etc.), compartmental model specific parameters (e.g., central compartment volume of distribution, absorption constants, desorption constants, etc.), modeled species (e.g., mammals, rodents, humans, non-human primates, etc.), modeled species demographics (e.g., age, weight, sex, etc.), baseline albumin, baseline tumor size, etc.
[0074] In various embodiments, the compartment-based pharmacokinetic model can be a one-compartment pharmacokinetic model. In the case of a one-compartment pharmacokinetic model, the compartment model parameters that can be set include, but are not limited to, the distribution volume of the central compartment, the absorption rate into the central compartment, the elimination rate from the central compartment, etc.
[0075] In various embodiments, the compartment-based pharmacokinetic model can be a two-compartment model. In the case of a two-compartment model, the configurable compartment model parameters include, but are not limited to, the distribution volume of the central compartment, the distribution volume of the peripheral compartment, the absorption rate into the central compartment, the elimination rate from the central compartment, the intercompartmental clearance between the central compartment and the peripheral compartment, etc.
[0076] In various embodiments, the compartment-based pharmacokinetic model can be a Michaelis-Menten two-compartment model. For the Michaelis-Menten PK two-compartment model, the compartment model parameters that can be set include, but are not limited to, the distribution volume of the central compartment, the distribution volume of the peripheral compartment, the absorption rate into the central compartment, the elimination rate from the central compartment, the intercompartmental clearance between the central and peripheral compartments, the Michaelis-Menten constant associated with the nonlinear clearance mechanism, etc.
[0077] In various embodiments, the training data 416 is transmitted in real-time or near real-time to the prediction system 410 for use in training the neural network system 411. In various embodiments, the training data 416 is generated and stored in the data storage 404, and then subsequently transmitted / obtained to the training neural network system 411.
[0078] The prediction system 410, in some embodiments, can be configured to initialize a self-organizing neural network, train the neural network using the training data 416, and predict one or more PK parameters 424 based on a time series concentration data set of the drug obtained from the subject (stored in the data storage unit 706). Examples of types of predicted OK parameters include, but are not limited to, AUC, C max , C min , C trough , T max , MRT, T last , or t 1 / 2 Includes.
[0079] After values for the PK parameters 424 are predicted, the PK parameters 424 can be displayed as a report 402 (e.g., results or summary) on the display system 406. The display system 406 in some embodiments may be implemented in a client terminal 426 communicatively connected to the computing platform 403. In various embodiments, the client terminal 426 can be a thin-client computing device. In various embodiments, the client terminal 426 can be a personal computing device having a web browser (e.g., Internet Explorer™, Firebox™, Safari™, etc.) that can be used to control the operation of the data storage 404, the data manager 408, the prediction system 410, or a combination thereof.
[0080] FIG. 5 is a depiction of an exemplary NN system 500 and a method of using it to process time-series concentration data to predict PK parameters, according to various embodiments. The NN system 500 shown in FIG. 5 can execute an ML-based methodology for PK parameter estimation or prediction, whereby the NN model 510 is trained by a plurality of simulated PK training data sets (i.e., simulated time-series concentration data and their corresponding PK parameters) generated by PK compartment model simulation and NCA modeling to enable automation of the NCA PK parameter modeling task by the trained NN model 510. In various embodiments, the NN model 510 is a feed-forward neural network. In various embodiments, the NN model 510 is a recurrent neural network. In various embodiments, the NN model 510 is a modular neural network. However, it should be understood that essentially any type of NN model 510 topology can be used, so long as it can be trained by the simulated PK training data sets to output accurate estimates or predictions of the PK parameters 508 when fed with real (non-simulated) or real-time time-series concentration data 502.
[0081] In various embodiments, the NN model 510 is a deep learning NN model, which is a class of machine learning algorithms that use multiple (hidden) layers to progressively extract higher level features from raw inputs. In various embodiments, the DL NN model is a convolutional neural network. In various embodiments, the DL NN model is a residual neural network.
[0082] 5, the NN model 510 can be comprised of an input layer 504, one or more hidden layers 506, and an output layer 508. In various embodiments, the input layer 504 is configured to incorporate time series concentration data 502 so that the data can be processed by the NN model 510. In various embodiments, the input layer 504 can be configured to incorporate additional inputs other than the time series concentration data, such as, but not limited to, the therapeutic agent (drug) being modeled, subject demographic information, disease state, etc.
[0083] In various embodiments, the hidden layers 506 can be composed of two or more, five or more, four or more, five or more, or more than five layers of nodes. In various embodiments, the number of hidden layers 506 (in the NN 510 model) and / or the number of nodes in each hidden layer 506 is pre-set prior to training. In various embodiments, the number of hidden layers 506 (in the NN 510 model) and / or the number of nodes in each hidden layer 506 is automatically set by the NN model 510 when it is being trained.
[0084] In various embodiments, the output layer 508 is max , C min , C trough , T max , MRT, TW, or t 1 / 2 The PK parameter may be configured to output one or more types of PK parameters, including but not limited to:
[0085] At a high level, the NN model 510 operates by finding the correct mathematical operations (by "learning" using training data) to turn an input, either an input layer of input data or any previous layer, into an output of a subsequent layer, either an output layer of the output of the NN model 510 or the output of any given layer acting as a previous layer to another layer. This occurs whether the mathematical operations are linear or non-linear relationships. The NN model 510 "learns" by adjusting the weights (and any thresholds) of the nodes in the various layers of the NN model to improve the accuracy of the results. Learning is complete when additional observations (of the training data) are examined but do not usefully reduce the error rate or reduce the error rate to a predefined value. Complex NN models, having many layers, are referred to as "deep learning" neural networks.
[0086] After learning (training) and during model use (inference, prediction), the NN model 510 is responsible for relating the input layer / input data set to the output layer / output data set via a mathematical function and the previously trained weights. During inference, the weights are fixed, and the input data and numerical values of the weights along with the architecture of the layers of the NN model 510 (as introduced above) represent a mathematical function of how the input data and weights are calculated together to compute the output of the NN model 510.
[0087] 6 illustrates an exemplary high-level workflow of how a NN model can be trained during a learning phase and then implement what it has learned to predict PK parameters, according to various embodiments. As mentioned above, a neural network processes information in two ways: when it is being trained, it is in the learning phase 602, and when it actually executes what it has learned, it is in the inference phase 604.
[0088] As shown herein, during the learning phase 602, training data 606 is first pre-processed 608 and then iteratively fed to the NN model 610 via a backpropagation (which is a method of training) process that adjusts the weights (and any thresholds) of each of the nodes in the various layers of the NN model 610, thereby converging to predetermined (or preset) validation set error values 612 and / or test set error values 614.
[0089] In various embodiments, pre-processing 608 of the training data 606 is a normalization of the time series concentration data and correlated NCA PK parameters that make up the training data 606. In various embodiments, pre-processing 608 of the training data 606 removes low quality data sets from the training data 606.
[0090] After the error backpropagation process has repeated enough iterations such that the validation set error values 612 and the test set error values 614 converge to predefined (pre-set) values, the fully trained NN model 618 can be used in the inference stage 604. In the inference stage 604, clinically obtained time-series concentration data 616 (i.e., real-world clinical data) undergoes a preprocessing step 620, and then this preprocessed data is fed to the trained NN model 618, which processes the preprocessed data and outputs one or more predicted PK parameter values 622. Outputs include, but are not limited to, AUC, C max , C min , C trough , T max , MRT, T last or t 1 / 2 The PK parameters can be as follows:
[0091] In various embodiments, pre-processing 608 of the training data 606 is a normalization of the time series concentration data and correlated NCA PK parameters that make up the training data 606. In various embodiments, pre-processing 608 of the training data 606 removes low quality data sets from the training data 606.
[0092] 7 is a plot 700 illustrating percent validation error according to an example embodiment. inf and C max For values, the graph shows how the percent validation error, calculated as predicted versus actual values after running the validation data set through the NN model 610 during the training phase 602, decreases as the number of training iterations increases. Note, however, that the improvement to the accuracy of the predictions eventually levels off as a function of the number of training iterations suggesting that there is an optimal number of training iterations for the NN model 610, and that there are diminishing returns in prediction accuracy when overriding the NN model.
[0093] During the backpropagation process, a "forward pass" refers to the computation process that calculates the output layer values from the input data. The input data traverses all the nodes from the first layer to the last layer. A loss function is computed from the output values, and then a "backward pass" refers to the process of counting the weight changes (effectively learning) using a gradient descent algorithm (or similar). The computation is done from the last layer back to the first layer. The backward and forward pass cycle together make one iteration.
[0094] FIG. 8 is a diagram of a plot 800 illustrating an example of training and validation loss as a function of training iteration number according to an exemplary embodiment. The plot 800 shows training loss and validation loss for the initial NN model 610 and the fully trained NN model 61.8 of FIG. 6. The training loss is the error on the training set of data. The validation loss is the error after mimicking the validation set of data through the trained network. In general, the lower the loss, the better the model (unless the model is overfitting to the training data). The loss is an interpretation of how well the NN model 610 is performing on these two sets. Unlike accuracy, the loss is not a percentage. It is the sum of the errors made on each example in the training or validation set.
[0095] 9 is a diagram of an exemplary training workflow 900 according to various embodiments. The training workflow 900 is a high-level diagram of how simulated PK data can be generated and used to create a training data set.
[0096] As illustrated herein, the simulated concentration time series data 902 are generated using one or more PK compartment models, which involves first setting (via inputs) simulation parameters of the one or more PK compartment models and then running a simulation of the PK compartment models to generate the simulated concentration time series data 902 for one or more existing or potential future therapeutic agents.
[0097] Examples of PK compartment model simulation parameters include, but are not limited to, administration scheme (e.g., subcutaneous, oral, intravenous, etc.), molecular class (e.g., small or large molecule), molecular species (e.g., drugs, mAbs, proteins, enzymes, etc.), compartment model specific parameters (e.g., central compartment volume of distribution, absorption constants, desorption constants, etc.), modeled species (e.g., mammals, rodents, humans, non-human primates, etc.), modeled species demographics (e.g., age, weight, sex, etc.), baseline albumin, baseline tumor size, etc.
[0098] In various embodiments, a one-compartment model can be used to simulate the time series concentration data 902. For the PK1 compartment model, the compartment parameters that can be set include, but are not limited to, the volume of distribution of the central compartment, the rate of absorption into the central compartment, the rate of elimination from the central compartment, etc.
[0099] In various embodiments, a two-compartment model can be used to simulate the time series concentration data 902. For the PK2-compartment model, the compartment parameters that can be set include, but are not limited to, the volume of distribution of the central compartment, the volume of distribution of the peripheral compartment, the absorption rate into the central compartment, the elimination rate from the central compartment, the intercompartmental clearance between the central compartment and the peripheral compartment, etc.
[0100] In various embodiments, a Michaelis-Menten PK two compartment model can be used to simulate the time series concentration data 902. For the Michaelis-Menten PK two compartment model, compartment parameters that can be set include, but are not limited to, the volume of distribution of the central compartment, the volume of distribution of the peripheral compartment, the rate of absorption into the central compartment, the rate of elimination from the central compartment, the intercompartmental clearance between the central and peripheral compartments, the Michaelis-Menten constant associated with the nonlinear clearance mechanism, etc.
[0101] After the simulated time series concentration data 902 is generated, it is sparsely sampled 906 to form a subset of simulated time series concentration training data sets 908, each of which is further processed using non-compartmental analysis (NCA) 904 to calculate their corresponding simulated PK parameter values 910. Examples of types of simulated PK parameter values include, but are not limited to, AUC, C max , C min , C trough , T max , MRT, T last or t 1 / 2 Includes.
[0102] 10 is a series of plots 1000 for various forms of compartmental modeling according to various embodiments. The series of plots 1000 illustrates how compartmental modeling can be used to simulate time series concentration data combined with PK parameters predicted using NCA to form a training data set, according to various embodiments.
[0103] In the example shown in Figure 10, three different compartment models (i.e., 1 compartment 1010, 2 compartment 1012 and 2 compartment Michaelis-Menten 1014) were used to generate a simulated time series concentration dataset 1002. To show the effect of dosing scheme on the simulated time series concentration dataset 1002, the same compartment model simulations were run using different treatment dosing scheme settings (i.e., bolus 1006 and oral / subcutaneous 1008).
[0104] As described above, after the simulated time series concentration datasets 1002 are generated, they are further processed using NCA to calculate their corresponding simulated PK parameter values 1004. In various embodiments, the simulated time series concentration datasets are for one or more existing or potential future therapeutic agents. V. Experimental Results
[0105] The improved systems and methods disclosed herein were compared with conventional approaches for predicting pharmacokinetic parameter values of a drug administered to a subject.
[0106] FIG. 11 shows the AUC and C using the neural network approach described herein compared to the traditional trapezoidal rule according to an exemplary embodiment. max 11 shows a set of plots 1100 illustrating improved accuracy in estimating
[0107] As shown herein, AUC inf The % error histogram 1102 shows the AUC determined using the NN method disclosed herein and the traditional trapezoidal rule method. inf Figure 1 shows a plot of the % error (predicted vs. actual) for the values. AUC determined using the NN method inf AUC values were determined using the traditional trapezoidal rule method. inf It is clear both visually and numerically (average error rate) that the results are significantly more accurate than the AUC infIn this case, the average percentage error of the NN method was about 9.1, while the average percentage error of the trapezoidal rule was about 81.5.
[0108] The observed significant improvement in the accuracy of the NN method over the conventional method was determined using the NN method and the conventional common observation method. max Plot the %error for the value C max This is reinforced by the %error histogram 1104, which shows that the improvement in accuracy using the NN prediction method is proportional to the AUC inf All PK parameters (e.g., C min , C trough , T max , or t 1 / 2 This is important because it shows that the same applies to the determination of max For,the mean percent error of the NN method was about 2.2, and the,mean percent error observed vs. the mean percent error of the true,was about 88.8. VI. Exemplary Methods
[0109] FIG. 12 is an exemplary flowchart illustrating a method 1200 for predicting a PK parameter value of an agent (i.e., a drug or a biologic) administered to a subject (i.e., a patient), according to various embodiments.
[0110] In step 1202, a neural network is trained by one or more processors based on a simulated training data set, the simulated training data set including simulated time series concentration data sets and simulated values for pharmacokinetic parameters corresponding to the simulated time series concentration data sets. In some embodiments, step 1202 includes training a neural network based on the multiple individual simulated time series concentration data sets, which in turn can be used to simulate values for one or more pharmacokinetic parameters for each simulated time series concentration data set of the multiple individual simulated time series concentration data sets.
[0111] In one or more embodiments, the simulated value for the pharmacokinetic parameter corresponds to the entirety of the simulated time series concentration dataset. In one or more embodiments, the simulated value for the pharmacokinetic parameter corresponds to a particular portion of the simulated time series concentration dataset (e.g., one or more data points in the simulated time series concentration dataset).
[0112] In various embodiments, the neural network is a feedforward neural network (FNN). In various embodiments, the neural network is a recurrent neural network (RNN). In various embodiments, the neural network is a modular neural network (MNN). In various embodiments, the neural network is a convolutional neural network (CNN). In various embodiments, the neural network is a residual neural network (ResNet).
[0113] In various embodiments, the compartment-based pharmacokinetic model is a one-compartment model. In various embodiments, the compartment-based pharmacokinetic model is a two-compartment model. In various embodiments, the compartment-based pharmacokinetic model is a two-compartment Michaelis-Menten model.
[0114] A time series concentration dataset of a drug obtained from a subject is received by one or more processors in step 1204. The drug may include, for example, a small molecule therapeutic or a biologic.
[0115] In step 1206, values for the pharmacokinetic parameters are predicted by one or more processors using the time series concentration data set and the trained neural network. As mentioned above, non-limiting examples of the types of pharmacokinetic parameters whose values are predicted include AUC, C max , C min , C trough , T max , MRT, Tlast , or t 1 / 2 Includes.
[0116] FIG. 13 is an exemplary flowchart illustrating a method 1300 for training a neural network to predict a pharmacokinetic parameter value of an agent (i.e., a drug or a biologic), according to various embodiments.
[0117] In step 1302, simulation parameters of a compartment-based pharmacokinetic model are received by one or more processors. Examples of simulation parameters include, but are not limited to, administration scheme (e.g., subcutaneous, oral, intravenous, etc.), molecular class (i.e., small or large molecule), molecular species (e.g., drugs, mAbs, proteins, enzymes, etc.), compartment model specific parameters (e.g., central compartment volume of distribution, absorption constants, desorption constants, etc.), modeled species (e.g., mammals, rodents, humans, non-human primates, etc.), modeled species demographics (e.g., age, weight, sex, etc.), baseline albumin, baseline tumor size, etc.
[0118] In various embodiments, the compartment-based pharmacokinetic model is a one-compartment model. In various embodiments, the compartment-based pharmacokinetic model is a two-compartment model. In various embodiments, the compartment-based pharmacokinetic model is a two-compartment model with Michaelis-Menten clearance.
[0119] In step 1304, a simulated training data set is generated by the one or more processors using the compartment-based pharmacokinetic model. The simulated training data set includes a simulated time series concentration dataset (or a plurality of simulated time series concentration datasets) and a simulated value for at least one pharmacokinetic parameter corresponding to the simulated time series concentration dataset (or a plurality of simulated time series concentration datasets). In one or more embodiments, a simulated value for the one or more pharmacokinetic parameters is generated for each simulated time series concentration dataset of the plurality of simulated time series concentration datasets.
[0120] The compartment-based pharmacokinetic model is encoded by the received simulation parameters. Examples of the types of pharmacokinetic parameters for which simulated values are generated include, but are not limited to, AUC, C max , C min , C trough , T max , MRT, T last , or t 1 / 2 Includes.
[0121] In step 1306, the neural network is trained by one or more processors using a training data set simulated using one or more processors. The trained neural network can then be used to predict pharmacokinetic parameter values of a drug.
[0122] In various embodiments, the neural network is a feedforward neural network (FNN). In various embodiments, the neural network is a recurrent neural network (RNN). In various embodiments, the neural network is a modular neural network (MNN). In various embodiments, the neural network is a convolutional neural network (CNN). In various embodiments, the neural network is a residual neural network (ResNet).
[0123] FIG. 14 is an exemplary flowchart illustrating a method 1400 for predicting a PK parameter value of an agent (eg, a drug, a biologic, etc.) administered to a subject (i.e., a patient), according to various embodiments.
[0124] In step 1402, a drug time series concentration data set obtained from a subject is received by one or more processors.
[0125] In step 1404, values for the pharmacokinetic parameters are predicted by the one or more processors based on the neural network trained using the time-series concentration dataset and the simulated training dataset. The neural network can be trained, for example, but not limited to, via method 1300 of Figure 13. The neural network can be trained using a simulated training dataset that includes a plurality of simulated time-series concentration datasets of one or more drugs and a simulated value for one or more pharmacokinetic parameters for each simulated time-series concentration dataset of the plurality of simulated time-series concentration datasets.
[0126] In various embodiments, the neural network is a feedforward neural network (FNN). In various embodiments, the neural network is a recurrent neural network (RNN). In various embodiments, the neural network is a modular neural network (MNN). In various embodiments, the neural network is a convolutional neural network (CNN). In various embodiments, the neural network is a residual neural network (ResNet). VII. Computer Implemented Systems
[0127] FIG. 15 is a block diagram illustrating a computer system 1500 on which an embodiment of the present teachings can be implemented. In various embodiments of the present teachings, the computer system 1500 can include a bus 1502 or other communication mechanism for communicating information and a processor 1504 coupled to the bus 1502 for processing information. In various embodiments, the computer system 1500 can also include a memory, which can be a random access memory (RAM) 1506 or other dynamic storage device, coupled to the bus 1502 for determining instructions to be executed by the processor 1504. The memory can also be used to store temporary variables or other intermediate information during execution of instructions to be executed by the processor 1504. In various embodiments, the computer system 1500 can further include a read-only memory (ROM) 1508 or other static storage device coupled to the bus 1502 for storing static information and instructions for the processor 1504. A storage device 1510, such as a magnetic disk or optical disk, can be provided and coupled to the bus 1502 for storing information and instructions.
[0128] In various embodiments, the computer system 1500 can be coupled via the bus 1502 to a display 1512, such as a cathode ray tube (CRT) or liquid crystal display (LCD), for displaying information to a computer user. An input device 1514, including alphanumeric and other keys, can be coupled to the bus 1502 for communicating information and command selections to the processor 1504. Another type of user input device is a cursor control device 1516, such as a mouse, trackball, or cursor direction keys, for communicating directional information and command selections to the processor 1504 and for controlling cursor movement on the display 1512. This input device 1514 typically has two degrees of freedom in two axes, a first axis (i.e., x) and a second axis (i.e., y), that allow the device to specify a position in a plane. However, it should be understood that input devices 1514 that allow three-dimensional (x, y, and z) cursor movement are also contemplated herein.
[0129] Consistent with a particular implementation of the present teachings, results can be provided by computer system 1500 in response to processor 1504 executing one or more sequences of one or more instructions contained in memory 1506. Such instructions can be read into memory 1506 from another computer-readable medium or computer-readable storage medium, such as storage device 1510. Execution of the sequences of instructions contained in memory 1506 can cause processor 1504 to perform the processes described herein. Alternatively, hardwired circuitry can be used in place of or in combination with software instructions to implement the present teachings. Thus, implementation of the present teachings is not limited to any specific combination of hardware circuitry and software.
[0130] The terms "computer-readable medium" (e.g., data store, data storage, etc.) or "computer-readable storage medium" as used herein refer to any medium that participates in providing instructions to the processor 1504 for execution. Such media may take many forms, including but not limited to, non-volatile media, volatile media, and transmission media. Examples of non-volatile media may include, but are not limited to, optical, solid-state, and magnetic disks, such as the storage device(s) 1510. Examples of volatile media may include, but are not limited to, dynamic memory, such as the memory 1506. Examples of transmission media may include, but are not limited to, coaxial cables, copper wire, and fiber optics, including the wires that comprise the bus 1502.
[0131] Common forms of computer readable media include, for example, a floppy disk, a flexible disk, a hard disk, a magnetic tape or any other magnetic medium, a CD-ROM, any other optical medium, punch cards, paper tape, any other physical medium with a pattern of holes, RAM, PROM, and EPROM, flash EPROM, any other memory chip or cartridge, or any other tangible medium from which a computer can read.
[0132] In addition to computer-readable media, instructions or data can be provided as signals on a transmission medium included in a communication device or system to provide a sequence of one or more instructions to the processor 1504 of the computer system 1500 for execution. For example, a communication device can include a transceiver having signals indicative of instructions and data. The instructions and data are configured to cause one or more processors to implement the functions outlined in the disclosure herein. Representative examples of data communication transmission connections include, but are not limited to, a telephone modem connection, a wide area network (WAN), a local area network (LAN), an infrared data connection, an NFC connection, and the like. VIII. Exemplary Embodiments
[0133] Various exemplary embodiments are described herein.
[0134] In one or more embodiments, a method is provided for training a neural network to predict at least one pharmacokinetic parameter of a drug. The one or more processors receive simulation parameters of a compartment-based pharmacokinetic model. The one or more processors generate a simulated training data set using the compartment-based pharmacokinetic model. The simulated training data set includes simulated time series concentration data sets of the one or more drugs and simulated values for the pharmacokinetic parameters corresponding to the simulated time series concentration data sets. The compartment-based pharmacokinetic model is encoded by the received simulation parameters. The one or more processors then train the neural network with the simulated training data set.
[0135] In one or more embodiments, a system for training a neural network to predict pharmacokinetic parameter values of a drug is disclosed. The system includes a data store, a computing device, and a display system. The data store stores simulation parameters of a compartment-based pharmacokinetic model. The computing device is communicatively connected to the data store and hosts a pharmacokinetic data simulation engine and a pharmacokinetic prediction engine. The pharmacokinetic data simulation engine is configured to generate a simulated training data set using the compartment-based pharmacokinetic model. The simulated training data set includes simulated time-series concentration data sets of one or more drugs and simulated values for pharmacokinetic parameters corresponding to the simulated time-series concentration data sets. The pharmacokinetic prediction engine is configured to train the neural network using the simulated training data collection.
[0136] In one or more embodiments, a system for predicting a pharmacokinetic parameter value of a drug administered to a subject comprises: a data store for storing a time-series concentration dataset of the drug obtained from the subject; a computing device communicatively connected to the data store, the computing device including a pharmacokinetic prediction engine configured to train a neural network to predict a pharmacokinetic parameter value based on the time-series concentration dataset of the drug obtained from the subject, the neural network being trained using a simulated training dataset including a plurality of simulated time-series concentration datasets of one or more drugs as well as corresponding simulated pharmacokinetic parameter values; and a display communicatively connected to the computing device and configured to display a report including the predicted pharmacokinetic parameter values.
[0137] In one or more embodiments, a non-transitory computer-readable medium storing computer instructions for predicting a value of at least one pharmacokinetic parameter of a drug administered to a subject includes training, by one or more processors, a neural network based on a simulated training dataset, the simulated training dataset including a simulated time series concentration dataset and a simulated value for the pharmacokinetic parameter corresponding to the simulated time series concentration dataset; receiving, by the one or more processors, a time series concentration dataset of the drug obtained from the subject; and predicting, by the one or more processors, a value for the pharmacokinetic parameter using the time series concentration dataset and the trained neural network.
[0138] In one or more embodiments, a non-transitory computer-readable medium storing computer instructions for training a neural network to predict at least one pharmacokinetic parameter of a drug includes receiving simulation parameters of a compartment-based pharmacokinetic model; generating a simulated training data set using the compartment-based pharmacokinetic model, the simulated training data set including simulated time-series concentration data sets of one or more drugs and simulated values for the pharmacokinetic parameters corresponding to the simulated time-series concentration data sets; encoding the compartment-based pharmacokinetic model with the received simulation parameters; and training the neural network with the simulated training data set.
[0139] In one or more embodiments, a non-transitory computer readable medium storing computer instructions for predicting at least one pharmacokinetic parameter of a drug administered to a subject includes receiving, by one or more processors, a time-series concentration dataset of the drug obtained from the subject, and predicting, by the one or more processors, a value for the pharmacokinetic parameter based on a neural network trained using the time-series concentration dataset and a simulated training data set.
[0140] In one or more embodiments, the neural network of one or more exemplary embodiments described herein is, in some embodiments, a convolutional neural network.
[0141] In one or more embodiments, a simulation parameter of one or more of the exemplary embodiments described herein is the class of the drug being simulated.
[0142] In one or more embodiments, one or more of the agents of the exemplary embodiments described herein is one of a small molecule therapeutic or a biologic.
[0143] In one or more embodiments, a simulation parameter of one or more exemplary embodiments described herein is a dosing scheme for a drug.
[0144] In one or more embodiments, the administration scheme of one or more of the exemplary embodiments described herein is one of intravenous administration of the agent or oral administration of the agent.
[0145] In one or more embodiments, the compartment-based pharmacokinetic model of one or more exemplary embodiments described herein includes at least one of a one-compartment model, a two-compartment model, or a Michaelis-Menten model.
[0146] In one or more embodiments, the data store and the computing device of one or more of the exemplary embodiments described herein are part of an integrated apparatus.
[0147] In one or more embodiments, the data store of one or more of the exemplary embodiments described herein is hosted by a device different from the computing device of one or more of the exemplary embodiments described herein.
[0148] In one or more embodiments, the data store and the computing devices of one or more of the exemplary embodiments described herein are part of a distributed network system. IX. Further Considerations
[0149] It should be understood that the flowcharts, diagrams, and accompanying disclosure described herein can be implemented using computer system 1300 as a standalone device or on a distributed network of shared computer processing resources, such as a cloud computing network.
[0150] The methodologies described herein can be implemented by various means depending on the application. For example, the methods can be implemented in hardware, firmware, software, or any combination thereof. In the case of a hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, electronic devices, other electronic units designed to perform the functions described herein, and / or combinations thereof.
[0151] In various embodiments, the methods of the present teachings may be implemented as firmware and / or software programs and applications written in conventional programming languages such as C, C++, Python, etc. When implemented as firmware and / or software, the embodiments described herein can be implemented on a non-transitory computer-readable medium having stored thereon a program for causing a computer to perform the above-described methods. It should be understood that the various engines described herein can be provided on a computer system, such as computer system 1100, whereby processor 1104 performs the analysis and decisions provided by these engines according to instructions provided by any one or combination of memory components 1106 / 1108 / 1110 and user input provided via input device 1114.
[0152] While the present teachings have been described in conjunction with various embodiments, it is not intended that the present teachings be limited to such embodiments. On the contrary, the present teachings encompass various alternatives, modifications, and equivalents, as will be appreciated by those skilled in the art.
[0153] In describing various embodiments, the specification may present a method and / or process as a particular sequence of steps. However, to the extent that the method or process does not depend on the particular order of steps described herein, the method or process should not be limited to the particular order of steps described, and one of ordinary skill in the art can readily appreciate that the order may be altered and still be within the spirit and scope of the various embodiments.
Claims
1. 1. A method for predicting at least one pharmacokinetic parameter of a drug administered to a subject, comprising: training, by one or more processors, a neural network based on a simulated training data set, the simulated training data set including a simulated time-series concentration data set of the drug and simulated values for pharmacokinetic parameters corresponding to the simulated time-series concentration data set of the drug; receiving, by the one or more processors, a time series concentration data set of the drug obtained from a subject; predicting, by the one or more processors, a value for the pharmacokinetic parameter using a time series concentration data set of the drug obtained from the subject and the trained neural network; A method comprising:
2. 2. The method of claim 1, wherein the simulated values for the pharmacokinetic parameters are generated using a compartment-based pharmacokinetic model.
3. 3. The method of claim 2, wherein the compartment-based pharmacokinetic model comprises at least one of a one-compartment model, a two-compartment model, or a Michaelis-Menten model.
4. The pharmacokinetic parameters are area under the curve (AUC), minimum concentration value (C min ), or the maximum density value (C max 10. The method of claim 1, further comprising at least one of:
5. The method of claim 1 , wherein the agent comprises a small molecule therapeutic or a biologic.
6. The method of claim 1 , wherein the neural network comprises a convolutional neural network.
7. 1. A system for predicting at least one pharmacokinetic parameter of a drug administered to a subject, comprising: a data store for storing a time series concentration data set of the drug obtained from a subject; a computing device communicatively connected to the data store, a pharmacokinetic data simulation engine configured to generate a simulated training data set for one or more drugs, the simulated training data set including a simulated time-series concentration data set for the drug and simulated values for pharmacokinetic parameters corresponding to the simulated time-series concentration data set for the drug; a pharmacokinetic prediction engine configured to train a neural network using the simulated training data set and to use the trained neural network to predict values for the pharmacokinetic parameters based on a time-series concentration data set of the drug obtained from the subject; A computing device comprising: a display system communicatively connected to the computing device and configured to display a report including the predicted values for the pharmacokinetic parameters; A system comprising:
8. 8. The system of claim 7, wherein the pharmacokinetic data simulation engine is configured to generate the simulated training data set using a compartment-based pharmacokinetic model.
9. 9. The system of claim 8, wherein the compartment-based pharmacokinetic model comprises at least one of a one-compartment model, a two-compartment model, or a Michaelis-Menten model.
10. The system of claim 7 , wherein the neural network comprises a convolutional neural network.
11. The system of claim 7 , wherein the agent comprises a small molecule therapeutic or a biologic.
12. The system of claim 7 , wherein the data store and the computing device are part of an integrated apparatus.
13. The system of claim 7 , wherein the data store is hosted by a device different from the computing device.
14. 1. A method for predicting at least one pharmacokinetic parameter of a drug administered to a subject, comprising: receiving, by one or more processors, a time series concentration data set of the drug obtained from a subject; predicting, by the one or more processors, a value for a pharmacokinetic parameter based on a time-series concentration dataset of the drug obtained from the subject and a neural network trained using a simulated training dataset, the simulated training dataset including a simulated time-series concentration dataset of the drug and a simulated value for the pharmacokinetic parameter corresponding to the simulated time-series concentration dataset of the drug; A method comprising:
15. The method of claim 14 , wherein the neural network is a convolutional neural network.
16. The pharmacokinetic parameters are area under the curve (AUC), minimum concentration value (C min ), or the maximum density value (C max 15. The method of claim 14, comprising at least one of:
17. The method of claim 14 , wherein the agent comprises a small molecule therapeutic or a biologic.
18. 1. A system for predicting at least one pharmacokinetic parameter of a drug administered to a subject, comprising: a data store for storing a time series concentration data set of the drug obtained from a subject; a computing device communicatively connected to the data store, the computing device configured to train a neural network using a simulated training dataset including a simulated time-series concentration dataset of the drug and a simulated value for a pharmacokinetic parameter corresponding to the simulated time-series concentration dataset of the drug, and to predict a value for the pharmacokinetic parameter based on the drug time-series concentration dataset obtained from the subject and the trained neural network; a display system communicatively connected to the computing device and configured to display a report including the predicted values for the pharmacokinetic parameters; A system comprising:
19. 20. The system of claim 18, wherein the neural network comprises a convolutional neural network.
20. The values of the pharmacokinetic parameters are the area under the curve (AUC), the minimum concentration value (C min ), or the maximum density value (C max 20. The system of claim 18, further comprising:
Citation Information
Patent Citations
Effective Pediatric Dosing for Treating ADHD with Methylphenidate
JP2020504763A
Residual Semi-Recurrent Neural Networks
JP2022526937A
Tool for in vitro-in vivo correlation
US20030078761A1
Neural network pattern recognition for predicting pharmacodynamics using patient characteristics
US20050216200A1
Multifunctional Neural Network System and Uses Thereof
US20110225112A1
