Retrosynthesis Pathway Search With MCTS and Selectivity Filters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-aided synthesis planning (CASP) systems are inefficient in designing complex reaction pathways for chemical compounds, requiring substantial time and creativity from chemists and lacking effective algorithms for enumerating and selecting optimal pathways.
Innovation Solution
A method utilizing a Monte-Carlo Tree Search (MCTS) algorithm, single-step reaction enumeration, and pathway scoring mechanisms to efficiently generate, filter, and rank reaction pathways for synthesizing target molecules, incorporating reaction selectivity filters and data from databases to optimize the synthesis process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual synthesis planning is performed by chemists studying literature, then the synthesis pathway can be designed with high creativity and chemical intuition, but the process requires substantial time and effort
Solution Approach 1:
The system enables self-service by automatically performing synthesis planning tasks that previously required chemists to manually study literature and design pathways. The computer system independently enumerates reactions, scores pathways, and generates synthesis protocols without human intervention in the core planning functions.
Solution Approach 2:
The patent replaces the mechanical system of manual literature review and pathway design with an automated computer-based system using machine learning models, reaction enumeration algorithms, and scoring mechanisms to perform synthesis planning.
2Reliability
If computer-aided synthesis planning systems use hand-coded rules and historical knowledge databases, then the systems can support planning with existing knowledge, but they lack effective algorithms for enumerating and selecting optimal pathways
Solution Approach 1:
The system changes the parameters of pathway evaluation by introducing a scoring mechanism that assigns numerical scores to reaction pathways based on multiple criteria including reaction reliability, pathway length, and synthetic feasibility. This enables automated comparison and selection of optimal pathways.
Solution Approach 2:
The patent implements feedback through the scoring mechanism that evaluates generated pathways and uses reaction outcome predictions to assess pathway quality. The system learns from training data and improves its pathway selection capability through iterative optimization.
3Manufacturing precision
If the system enumerates all possible reaction pathways, then complete coverage of chemical space is achieved, but the computational complexity and time required increase substantially
Solution Approach 1:
The system applies partial action by using reaction selectivity filters that focus the enumeration on chemically reasonable pathways. Rather than exhaustively exploring all possible reactions, the filters prune the search space to include only pathways that meet predetermined chemical feasibility criteria.
Solution Approach 2:
The patent segments the synthesis planning process into distinct modules: reaction enumeration, pathway construction, scoring, and filtering. This segmentation allows each component to be optimized independently and facilitates efficient processing of complex synthesis problems.
Data Source
AI summary
A synthesis protocol for a reaction pathway of a target molecule can be determined by: providing target compound data; performing a chemical synthesis search for at least one reaction pathway for the target compound; processing the target compound data through a single-step reaction enumeration algorithm to obtain at least one reaction step of the least one reaction pathway; processing at least one reaction step with the at least one reaction pathway scoring mechanism model to obtain a reaction step score; constructing reaction pathways based on at least one reaction step and at least one reaction step score; providing a selectivity filter having a selectivity criteria; filtering the reaction pathways so that reactions violating the selectivity criteria is filtered out; ranking the reaction pathways; and providing the reaction pathway ranking.