Autoencoding Formulations for Faster Multi-Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing formulation development processes are labor-intensive and require extensive experimental efforts to identify formulations that meet multiple target parameters, such as environmental compatibility, bioavailability, and stability, without effective decision support tools for predicting the effects of ingredient additions or modifications.
Innovation Solution
A machine learning model is trained using composition and property data of numerous formulations to generate a compressed representation, predict properties, and minimize loss functions, enabling the identification of promising formulation candidates and predicting their properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional experimental formulation development is used, then formulations meeting multiple target parameters can be identified, but extensive experimental effort and time are required
Solution Approach 1:
The patent applies preliminary action by training a machine learning model in advance on extensive formulation data to learn complex composition-property relationships. This pre-trained model can then rapidly predict properties of new formulations without requiring extensive experimental trials, thus reducing development time while maintaining prediction accuracy.
Solution Approach 2:
The patent uses copying by creating a virtual model (machine learning model) that replicates the behavior and relationships of real formulation systems. This digital twin allows for in-silico experimentation and prediction, replacing numerous physical experiments with computational predictions, thereby significantly reducing time and resource requirements.
2Reliability
If traditional experimental formulation development is used, then formulations meeting multiple target parameters can be identified, but extensive experimental effort and resources are required
Solution Approach 1:
The patent replaces the mechanical/experimental system with an information-processing system. Instead of physically preparing and testing numerous formulation variants, the machine learning model processes composition data and predicts properties computationally. This substitution of physical experimentation with computational prediction dramatically improves development efficiency while maintaining reliability.
Solution Approach 2:
The patent applies parameter changes by transforming the formulation development process from physical experimentation to computational prediction. The machine learning model learns from historical data and uses this knowledge to predict properties based on composition parameters, enabling rapid evaluation of multiple formulations without physical synthesis and testing.
3Productivity
If machine learning model is trained to predict formulation properties, then extensive experimentation can be reduced, but training data requirements and model complexity increase
Solution Approach 1:
The patent applies universality by designing a machine learning model that can predict multiple formulation properties simultaneously from a single set of composition inputs. This multi-functional approach consolidates what would otherwise require separate models or experimental procedures, managing complexity while enhancing productivity across multiple property predictions.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Systems, methods, and computer programs disclosed herein relate to training and using a machine learning model to predict compositions and/or properties of formulations, in particular of active ingredient formulations.