Transfer Learning for Sensory Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting sensory properties, such as olfactory properties, face challenges due to the complexity of molecular structure relationships and limited available data, especially in domains like human scent perception and insect repellents, where traditional approaches like physics-based modeling are computationally expensive and require extensive data.
Innovation Solution
A machine-learned sensory prediction model is trained on a first sensory prediction task and used to make predictions for a second task, leveraging graph neural networks to process chemical structures and generate sensory embeddings that can be transferred and fine-tuned for tasks with limited data, enabling accurate predictions across different species and domains.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based modeling is used to predict sensory properties, then prediction accuracy can be improved, but computational cost and data requirements increase significantly
Solution Approach 1:
The patent replaces physics-based modeling (mechanical/computational system) with machine learning models that learn patterns directly from data. The system uses neural networks to predict sensory properties without requiring complex physics calculations, thereby reducing computational cost while maintaining prediction accuracy through data-driven approaches.
Solution Approach 2:
The patent creates simplified representations (molecular embeddings) that capture essential molecular features without requiring full physics-based analysis. These embeddings serve as compressed copies of molecular information that can be processed efficiently by machine learning models, reducing computational requirements while preserving predictive power.
2Manufacturing precision
If extensive experimental production is conducted to gather training data, then model training quality improves, but time and resource consumption increase
Solution Approach 1:
The patent performs preliminary actions by pre-training models on large public datasets before fine-tuning on smaller task-specific datasets. This preliminary training establishes strong baseline representations that reduce the amount of additional experimental data needed, thereby reducing time and resource consumption while maintaining model training quality.
Solution Approach 2:
The patent develops universal molecular embeddings that can be applied across multiple sensory prediction tasks. These embeddings capture general molecular features that transfer across different prediction tasks, reducing the need for task-specific experimental data collection and enabling efficient adaptation to new tasks with limited data.
3Adaptability or versatility
If traditional trial-and-error methods are used for molecular design, then flexibility in exploring chemical space is maintained, but productivity and efficiency decrease
Solution Approach 1:
The patent implements feedback loops where predicted sensory properties guide subsequent molecular design decisions. The system uses prediction results to identify promising molecular candidates and iteratively refine designs, replacing random trial-and-error with directed optimization that maintains flexibility while dramatically improving productivity.
Solution Approach 2:
The patent employs dynamic molecular optimization where the search strategy adapts based on prediction outcomes. The system dynamically adjusts which molecular features to explore and how to modify structures based on feedback from sensory predictions, enabling efficient navigation of chemical space while maintaining adaptability to different target properties.
Data Source
AI summary
A computer-implemented method for predicting whether a molecule will be a good mosquito repellent is disclosed. The method includes obtaining a machine-learned prediction model obtained by transfer learning. The model has been trained using a first, larger training dataset for an odour prediction task and with a second, smaller training dataset for predicting whether a molecule would function as a mosquito repellent. The method further includes obtaining input data that describes a chemical structure of a selected molecule, providing the input data that describes the chemical structure of the selected molecule as input to the machine-learned prediction model, receiving prediction data descriptive of whether the selected molecule would be a good mosquito repellent as an output of the machine-learned sensory prediction model and providing the prediction data as output.