3D Molecular Vector Representation for Unlabeled Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI techniques are inadequate for generating new compounds with desired properties without relying on labeled data, requiring expensive cycles of synthesis and testing, and lack the ability to correlate molecular structures with potential properties.
Innovation Solution
An AI engine composed of multiple modules using machine learning models to predict molecular properties and generate candidate molecules based on targeted properties, employing unsupervised, invertible, physics-based vector representations in 3D space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If AI techniques are used to predict molecular properties, then the number of iterations in the design-make-test cycle is reduced, but the accuracy of property prediction without labeled data is insufficient
Solution Approach 1:
The patent introduces a physics-based vector representation as an intermediary that bridges molecular structure and properties. This representation encodes geometric and physical information in a continuous vector space, enabling accurate property prediction without requiring labeled training data. The vector representation acts as a mediator that translates structural information into predictive features while maintaining physical interpretability.
Solution Approach 2:
The patent performs preliminary encoding of molecular structures into physics-based vector representations before property prediction. By pre-processing structural information into meaningful vector embeddings that capture geometric and physical characteristics, the system prepares the data in advance for accurate prediction, reducing the need for iterative testing while maintaining high precision.
2Reliability
If traditional AI techniques are used for compound generation, then the ability to learn from labeled data is improved, but the capability to generate new compounds with desired properties from unlabeled data is limited
Solution Approach 1:
The patent creates a universal physics-based vector representation framework that functions effectively whether labeled or unlabeled data is available. The same vector encoding mechanism works for both supervised learning from labeled datasets and unsupervised generation from unlabeled structures, making the system adaptable to different data conditions while maintaining reliability in property prediction.
Solution Approach 2:
The patent changes the parameter representation from discrete molecular descriptors to continuous physics-based vector fields. This parameter transformation enables the system to capture subtle geometric and physical variations that are critical for property prediction, allowing accurate generation of novel compounds even without labeled training data by leveraging physical principles encoded in the vector representation.
3Measurement precision
If expensive synthesis and testing cycles are used, then the accuracy of determining targeted properties is improved, but the time and cost efficiency deteriorates
Solution Approach 1:
The patent replaces the mechanical synthesis-testing-measurement cycle with a computational physics-based prediction system. Instead of physically synthesizing compounds and measuring their properties through expensive experiments, the system uses physics-based vector representations to computationally predict properties with high accuracy, substituting physical experimentation with theoretical modeling that maintains precision while eliminating time and cost losses.
Data Source
AI summary
The Artificial Intelligence engine can perform one or more operations. A query can be submitted to the Artificial Intelligence engine to search directly for a set of targeted properties for an unnamed molecule having the set of targeted properties. An indication of a structure of one or more candidate molecules found to have the set of targeted properties with the Artificial Intelligence engine is generated by applying one or more machine learning algorithms. The indication of the structure of the one or more candidate molecules found to satisfy the set of targeted properties in 3-dimensional space is supplied to a user in response to the query for the set of targeted properties to the Artificial Intelligence engine.