Deep Learning PK Parameter Prediction From Concentration Time Series

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for estimating pharmacokinetic properties of therapeutics are inaccurate and labor-intensive, relying on crucial assumptions and requiring significant computational resources, limiting their applicability to non-expert users.

Innovation Solution

Employing machine learning, specifically deep learning neural networks, to predict pharmacokinetic parameters by learning features from underlying data, exploiting local patterns for improved accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional mathematical modeling methodologies are used for PK/PD evaluation, then model evaluation and refinement can be performed, but the process becomes time and labor intensive requiring significant computational resources and expert know-how

Engineering Contradiction:
ImprovePK parameter estimation accuracyVSAvoidmodel evaluation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces conventional mathematical modeling methodologies (which require iterative optimization, function evaluations, and gradient calculations) with a machine learning-based system. The ML model is trained on simulated PK data and then used to directly predict PK parameters from concentration-time profiles, eliminating the need for time-consuming iterative mathematical optimization and reducing dependency on expert know-how.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If conventional mathematical algorithms such as expectation-maximization or genetic algorithms are used, then PK parameters can be estimated, but the process requires significant trial-and-error and computational time

Engineering Contradiction:
ImprovePK parameter estimation accuracyVSAvoidmodeling efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent performs preliminary action by generating extensive simulated PK data using compartmental models and training the machine learning system in advance. This pre-trained model can then rapidly predict PK parameters for new concentration-time profiles without requiring trial-and-error optimization during actual analysis, significantly improving productivity while maintaining accuracy.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If iterative optimization methods are used for PK modeling, then accurate parameter estimation can be achieved, but the process requires significant computational resources and expert know-how

Engineering Contradiction:
ImprovePK parameter accuracyVSAvoidmodeling complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent substitutes complex iterative optimization algorithms with a machine learning-based prediction system. The ML model captures the complex relationships between concentration-time profiles and PK parameters through training on simulated data, allowing accurate parameter estimation without requiring users to understand or implement complex optimization algorithms, thereby reducing modeling complexity.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

4Measurement precision

If conventional PK modeling approaches are used, then pharmacokinetic parameters can be calculated, but the methods are labor-intensive and require significant computational resources

Engineering Contradiction:
ImprovePK parameter estimation accuracyVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent performs computationally intensive work in advance by generating simulated PK data and training the machine learning model. Once trained, the model can rapidly predict PK parameters with minimal computational resources required during actual analysis, significantly reducing computational resource consumption while maintaining estimation accuracy.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4128248B1Estimating pharmacokinetic parameters using deep learning
Publication Date: 2025.12.10 GENENTECH INC
  • EP4128248B1 patent drawingFigure 1
  • EP4128248B1 patent drawingFigure 2
  • EP4128248B1 patent drawingFigure 3

AI summary

A method and system for predicting at least one pharmacokinetic parameter of an agent administered to a subject. One or more processors train, by one or more processors, a neural network based on a simulated training data collection. The simulated training data collection comprising a simulated time-series concentration dataset and a simulated value for a pharmacokinetic parameter that corresponds to the simulated time-series concentration dataset. The one or more processors receive a time-series concentration dataset of the agent obtained from a subject. The one or more processors predict a value for the pharmacokinetic parameter using the time-series concentration dataset and the neural network that has been trained.