Reaction-Based Chemical Compound Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning-based de novo drug design strategies, particularly structure generating schemes, face challenges such as generating infeasible molecules and limited applicability due to data-driven artifacts and reliance on specific training sets, while reaction-based models require known template compounds and are not trained end-to-end.
Innovation Solution
A reaction-based scheme using machine learning, including reinforcement learning and genetic algorithms, that explores synthetically accessible chemical space without needing template compounds, enabling efficient generation of chemical compounds with desired properties and providing synthesis methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If structure generating schemes are used to generate chemical compounds, then the generation speed and exploration of chemical space are improved, but the synthetic accessibility and feasibility of generated molecules deteriorate
Solution Approach 1:
The patent introduces a reaction-based model as an intermediary between the generative model and the final compound selection. This model uses known chemical reactions and commercially available reactants to guide the generation process, ensuring that generated compounds are synthetically accessible while maintaining high generation speed through machine learning acceleration
Solution Approach 2:
The patent changes the generation parameters from purely structure-based to reaction-based parameters. By incorporating reaction templates, reactant availability, and synthetic pathways as generation parameters, the system maintains high productivity while ensuring synthetic feasibility through the constraints of known chemical reactions
2Ease of manufacture
If reaction-based models are used to ensure synthetic accessibility, then the synthetic feasibility of generated compounds is improved, but the requirement for known template compounds increases system complexity
Solution Approach 1:
The patent creates a universal reaction-based framework that can handle multiple drug discovery tasks without requiring task-specific template compounds. The system uses a library of general chemical reactions and commercially available reactants that can be applied across different target proteins and disease indications, reducing the need for task-specific customization and lowering overall system complexity
3Ease of manufacture
If existing reaction-based models like DINGOS or PathFinder are used, then the synthetic accessibility is improved, but the requirement for known template lead compounds limits adaptability
Solution Approach 1:
The patent performs preliminary action by pre-compiling a comprehensive library of chemical reactions and commercially available reactants before the actual drug discovery process. This pre-prepared knowledge base enables the system to immediately generate synthetically accessible compounds for any target without requiring pre-existing template lead compounds, thereby improving both synthetic accessibility and adaptability
4Measurement precision
If data-driven estimation is used to compensate for infeasible molecules, then the scoring accuracy is improved, but the computation time increases
Solution Approach 1:
The patent performs preliminary action by using the reaction-based model to pre-filter and generate only synthetically accessible compounds. This eliminates the need for subsequent retrospective synthetic feasibility analysis and complex post-generation scoring, thereby maintaining high scoring accuracy while significantly reducing computation time
Data Source
AI summary
A system and method for generating libraries of chemical compounds having desired and specific properties by formulating a reaction-based mechanism that may be powered by several algorithms including but not limited to genetic algorithm, expert iteration algorithms, planning methods, reinforcement learning and machine learning algorithms. The system and method may also provide the process steps by which these optimized products S′ may be synthesized from the reactants R1,R2 and further enables a rapid and efficient search of the synthetically accessible chemical space.