Expert-in-the-Loop AI Polymer Generation for Synthetic Viability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for designing new polymers generate a large number of candidates, many of which are not synthetically viable or compliant with industrial constraints, leading to inefficiencies and bottlenecks in the discovery process.
Innovation Solution
A computer-implemented method and system that uses machine learning to accelerate polymer discovery by training models to generate new materials with desired features, incorporating expert knowledge and constraints to improve synthetic viability and compliance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If computational screening and generative modeling are used to accelerate polymer design, then the number of candidate materials generated increases, but the proportion of synthetically viable candidates decreases
Solution Approach 1:
The system performs preliminary filtering of candidate materials using multiple criteria (synthetic accessibility, polymerization robustness, regulatory compliance, feed-stock availability) before presenting them to experts. This pre-screening action ensures that only high-quality, synthetically viable candidates advance to the review stage, resolving the contradiction by improving reliability while maintaining productivity through automated evaluation.
Solution Approach 2:
The system incorporates feedback loops where expert reviews of candidate materials are used to refine and retrain the machine learning models. This continuous feedback mechanism improves the accuracy of synthetic viability predictions over time, ensuring that generated candidates better align with practical synthesis constraints while maintaining high generation rates.
2Adaptability or versatility
If a large number of candidate materials are generated, then the coverage of material space increases, but the time required for expert review increases
Solution Approach 1:
The system extracts and evaluates multiple critical criteria (synthetic accessibility, polymerization robustness, regulatory compliance, feed-stock availability) separately from the main expert review process. By taking out these evaluation tasks and automating them through machine learning models, the system maintains comprehensive material space coverage while significantly reducing the time burden on experts.
Solution Approach 2:
The system performs preliminary evaluation and ranking of candidates based on multiple automated criteria before presenting them to experts. This pre-processing action filters out low-quality candidates and prioritizes promising ones, allowing experts to focus their time on reviewing a smaller, higher-quality subset while still maintaining broad material space coverage through the automated screening of the full candidate set.
3Speed
If existing generative approaches are used, then the speed of candidate generation increases, but the consideration of practical constraints decreases
Solution Approach 1:
The system integrates multiple evaluation functions (synthetic accessibility assessment, polymerization robustness prediction, regulatory compliance checking, feed-stock availability verification) into a single unified framework. This multi-functional approach maintains fast candidate generation speeds while simultaneously considering all practical manufacturing constraints through parallel automated evaluations.
Solution Approach 2:
The system performs preliminary assessments of practical constraints (synthetic viability, polymerization robustness, regulatory compliance, feed-stock availability) during the candidate generation process itself, rather than as separate post-processing steps. This preliminary action ensures that speed is maintained while comprehensive consideration of manufacturing ease is incorporated into the generation criteria.
Data Source
AI summary
Candidate material for polymerization can be received. One or more desired features in the candidate material can be identified. A machine learning model can be trained to generate a new material having one or more of the desired features. Permissively, the candidate material can be determined from running a machine learning classification model that ranks a plurality of material as candidates. Permissively, the generated new material can be input to the machine learning classification model, for the machine learning classification model to include in ranking the plurality of material as candidates.