Molecular Structure Confidence Scoring via Image-Graph Rendering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing molecular structure prediction models lack confidence scoring, leading to inaccurate data accumulation in academic databases due to erroneous predictions, making it difficult to determine reliable structural formulas.
Innovation Solution
A system that includes a processor to obtain molecular structure images, perform image rendering, and determine confidence using a second learning model that combines image and graph features to provide a confidence score, storing only graphs with a predetermined level of accuracy in a database.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a molecular structure prediction model is used to extract structural formulas from images, then the productivity of creating academic databases is improved, but the reliability of the data deteriorates due to erroneous predictions without confidence information
Solution Approach 1:
The patent implements a feedback mechanism by introducing a confidence model that evaluates the predicted molecular structure graph and outputs a confidence score. This feedback loop allows the system to assess its own predictions, enabling automatic filtering of low-confidence results before database storage, thus maintaining high reliability while preserving productivity.
Solution Approach 2:
The confidence model serves as an intermediary between the molecular structure prediction model and the database storage system. It evaluates the predicted structural formulas and provides confidence scores, acting as a mediator that determines which predictions are reliable enough for database inclusion, thereby resolving the contradiction between productivity and reliability.
2Reliability
If confidence measurement is added to the prediction model, then the reliability of predicted data is improved, but the device complexity increases due to additional processing steps
Solution Approach 1:
The patent segments the system into two distinct modules: a molecular structure prediction model that generates structural formulas from images, and a separate confidence model that evaluates the predicted graphs. This segmentation allows each module to specialize in its function, improving reliability through dedicated evaluation while managing complexity through modular design.
Solution Approach 2:
The confidence model acts as an intermediary component that adds reliability assessment without fundamentally altering the core prediction model. By inserting this evaluation layer between prediction and storage, the system achieves higher reliability while isolating the complexity increase to a dedicated module rather than propagating it throughout the entire system.
3Quantity of substance
If all predicted structural formulas are stored in the database, then the quantity of data is improved, but the purity of the database deteriorates due to inclusion of erroneous predictions
Solution Approach 1:
The confidence model provides feedback scores for each predicted structural formula, enabling the system to automatically filter predictions based on confidence thresholds. This feedback mechanism ensures that only high-confidence predictions are stored in the database, maintaining data purity while still accumulating a substantial quantity of reliable structural formulas over time.
Solution Approach 2:
The patent changes the parameter of data inclusion from binary (store or not store) to continuous (confidence-based filtering). By introducing confidence scores as a filtering parameter, the system can dynamically adjust which predictions are stored based on their reliability, thereby maintaining both adequate data quantity and high data purity in the database.
Data Source
AI summary
A system, a computer program, a device, and a method for measuring confidence of a molecular structure prediction model. The method includes obtaining a first molecular structure image, obtaining a first molecular structure graph using the molecular structure prediction model, performing image rendering on the first molecular structure image based on the first molecular structure graph, and determining confidence of the first molecular structure graph based on the image rendering result and the first molecular structure graph.