Machine Learning Models for Synthetic Route Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current algorithms for planning multi-step chemical reactions to synthesize small organic molecules are inefficient, robust, and not optimized for determining efficient synthesis routes, limiting the development of drugs and related compounds.
Innovation Solution
A method utilizing deep-learning neural networks to predict single-step organic chemistry reactions and autonomously design synthetic pathways by imbibing all known single-step reactions, guided by machine learning models and proprietary chemical data intelligence to efficiently synthesize drug products, including the use of knowledge graphs and dynamic linear programming for optimizing routes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional algorithms are used for planning multi-step chemical reactions, then synthesis routes can be generated, but the process is tedious, time-consuming, and inefficient
Solution Approach 1:
The patent replaces traditional mechanical algorithmic approaches with machine learning models that have been trained on chemical reaction data. The ML models autonomously predict reaction outcomes and optimize synthesis routes, substituting the step-by-step mechanical planning process with intelligent prediction-based planning that significantly reduces time and improves efficiency
Solution Approach 2:
The machine learning models are trained on extensive chemical reaction datasets and enable the system to independently determine optimal synthesis routes without requiring manual intervention or traditional algorithmic processing. The system serves itself by using learned patterns from training data to automatically plan multi-step reactions, eliminating the need for tedious manual or algorithmic planning
2Reliability
If existing algorithms are used for determining synthesis routes, then routes can be produced, but they are not robust or optimized for efficiency
Solution Approach 1:
The machine learning models are pre-trained on extensive datasets of chemical reactions before being deployed for synthesis route determination. This preliminary training action equips the models with robust knowledge of reaction patterns, outcomes, and optimizations, enabling them to reliably determine efficient synthesis routes when applied to new molecular targets
Solution Approach 2:
The patent utilizes machine learning models that can evaluate and optimize multiple parameters simultaneously (reaction conditions, reagent selection, step sequencing) to determine the most efficient synthesis routes. The models adjust these parameters based on learned patterns from training data, achieving both robustness and optimization that traditional algorithms cannot provide
Data Source
AI summary
A subject matter of the present invention is system, method, and computer readable medium configured to synthesize a drug product. Single-step organic-chemistry reactions are provided as inputs to a first machine learning model. At least one synthetic route to a product is determined based on an output of the first machine learning model. A score for each of the at least one synthetic route is determined. The at least one synthetic route and its respective score are provided as inputs to a second machine learning model. At least one qualified route is determined based on an output of the second machine learning model.