Chemical Structure Search With Diversity-Preserving Evolutionary Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for searching a compound with a desired physical property value, such as Bayesian molecular design with a chemical language model, face inefficiencies due to the search quickly falling into local minima, reducing structural diversity and slowing the search process.
Innovation Solution
A method that includes inputting chemical structures and property values, calculating candidate structures, and adopting or rejecting them based on both approaching the target property value and increasing structural diversity, using probabilistic processes to escape local minima.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Bayesian inference with particulate filter is used to update structures, then the physical property value approaches the target value, but the structural diversity is reduced and the search falls into local minimum
Solution Approach 1:
The patent dynamically adjusts the exploration-exploitation balance during the search process. Initially, the algorithm emphasizes exploration by accepting structures that increase diversity even if they temporarily worsen property values. As the search progresses and diversity suffices, it shifts to exploitation, accepting only structures that improve property values. This dynamic adjustment prevents premature convergence while ensuring eventual convergence to high-quality solutions.
Solution Approach 2:
The patent changes the acceptance criteria parameters based on the search state. The first adoption process uses a threshold based on property value improvement, while the second adoption process uses a threshold based on diversity increase. These parameter thresholds are applied in a staged manner, allowing the algorithm to transition from diversity-focused search to precision-focused search as needed.
2Reliability
If sampling with replacement is performed based on weight, then structures with better property values are selected, but the search efficiency is lowered due to repeated sampling of similar structures
Solution Approach 1:
The patent extracts and enforces the diversity constraint separately from the property value optimization. By introducing a dedicated diversity evaluation (second adoption process) that operates independently from the property value weighting, the algorithm prevents the collapse into sampling only similar high-weight structures. This separation allows efficient property value optimization while maintaining structural diversity through the dual-adoption mechanism.
3Measurement precision
If the search focuses on improving physical property value, then the target value is approached, but the search gets trapped in local minimum and cannot escape
Solution Approach 1:
The patent performs preliminary diversity preservation actions early in the search process through the second adoption process. By proactively maintaining a diverse population of candidate structures before the search converges, the algorithm creates multiple potential pathways forward. This preliminary diversity maintenance acts as a safety mechanism that prevents premature trapping in local minima, allowing the search to escape and continue improving property values over time.
Data Source
AI summary
An object of the present invention is to provide a method for searching a compound, a program for searching a compound, a recording medium, and a device for searching a compound, which are capable of efficiently searching a structure of a compound. In the method for searching a compound according to the first aspect, because of performing a first adoption process to determine based on whether or not a physical property value of a candidate structure approaches a target value of a physical property value due to a change in chemical structure, in a case where the candidate structure is not adopted as a result of the first adoption process, performing a second adoption process to determine based on whether or not a structural diversity increases, and in a case where the candidate structure is not adopted as a result of the first adoption process and the second adoption process, performing a rejection process to reject the change in chemical structure and return to the chemical structure before the change, it is possible to improve the structural diversity to promote escape from local minimum, and efficiently search for the structure of the compound having a desired physical property value (target value).