Molecular-Orbital Machine Learning for Fast Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current molecular simulation methods for designing and synthesizing molecules are computationally expensive and time-consuming, consuming significant resources and requiring months of wall-clock time, limiting their application in industrial innovation and development.
Innovation Solution
The use of molecular-orbital-based machine learning (MOB-ML) processes to predict molecular system properties, which enables faster computation and human efficiency by generating molecular-orbital-based features and using these features to synthesize molecules with specific properties, such as solubility and binding affinity, through machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based quantum mechanical methods are used to simulate molecular systems, then prediction accuracy is improved, but computational time and resource consumption increase dramatically
Solution Approach 1:
The patent creates a machine learning model that copies the behavior and predictive capabilities of expensive physics-based quantum mechanical methods. By training the model on data generated from these accurate but slow methods, the system learns to replicate their prediction accuracy while operating at much lower computational cost and speed, effectively creating a fast surrogate model that mimics the slow but accurate physics-based approaches.
2Measurement precision
If physics-based quantum mechanical methods are used to simulate molecular systems, then prediction accuracy is improved, but resource consumption increases significantly
Solution Approach 1:
The machine learning model serves as a computational copy that replicates the predictive accuracy of resource-intensive physics-based quantum mechanical methods. Once trained on a subset of data generated from these accurate methods, the model can make predictions without requiring the extensive computational resources and energy consumption of the original physics-based approaches, thereby maintaining accuracy while dramatically reducing resource usage.
3Reliability
If traditional molecular simulation methods are used, then molecular system properties can be determined, but the process is too slow for industrial application
Solution Approach 1:
The patent implements a machine learning model that copies the functionality of traditional molecular simulation methods for determining molecular system properties. This surrogate model maintains the reliability of property determination while achieving computational speeds thousands of times faster, enabling industrial-scale applications where rapid screening and optimization of molecular candidates are required.
Solution Approach 2:
The patent transforms the computational approach by changing from direct physics-based quantum mechanical calculations to a machine learning parameterized model. By representing molecular properties as functions of molecular orbital features processed through trained neural networks, the system achieves both reliability in property prediction and the computational speed necessary for industrial productivity.
Data Source
AI summary
Systems and methods for determining molecular structures based on molecular-orbital-based (MOB) features are described. MOB features can be utilized in combination with machine-learning methods to predict accurate properties, such as quantum mechanical energy, of molecular systems.