Compound Property Prediction Using Edge-Propagation Molecular Graphs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing AI models for predicting compound properties, such as toxicity and solubility, face challenges due to the lack of a grid structure in chemical compounds, leading to lower accuracy in property prediction.

Innovation Solution

A method involving a multi-layer edge information propagation model is used to convert chemical structure information into a graph, perform multiple rounds of information propagation on edges to obtain propagation state information, and predict compound properties based on edge features using a neural network.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional AI models (CNN) are used for compound property prediction, then the model structure is simple, but the prediction accuracy is low due to lack of grid structure in chemical compounds

Engineering Contradiction:
Improveprediction accuracyVSAvoidmodel complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the chemical compound representation into distinct node features (atom properties) and edge features (bond properties). Each node and edge is independently processed and updated through separate message passing mechanisms, allowing the model to capture local structural characteristics while maintaining overall molecular context, thereby improving prediction accuracy without excessive complexity

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from traditional grid-based CNN architectures to a graph-based representation that operates in a different dimensional space. By representing molecules as graphs with nodes and edges rather than forcing them into grid structures, the model accommodates the inherent non-grid nature of chemical compounds while capturing structural relationships through edge-based message passing

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If edge information propagation is performed for multiple rounds, then the utilization of chemical bond features is improved, but the computational time increases

Engineering Contradiction:
Improvefeature utilization accuracyVSAvoidcomputational time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs a fixed number of rounds of edge information propagation (e.g., 3-5 rounds) rather than exhaustive propagation. This partial action is sufficient to capture the essential chemical bond features and their contextual relationships, achieving high prediction accuracy while avoiding the excessive computational time that would result from too many propagation rounds

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The message passing process continuously refines edge features across multiple propagation rounds, with each round building upon the previous round's results. This continuous refinement ensures that chemical bond features are thoroughly utilized and contextualized, improving feature utilization accuracy while the fixed number of rounds prevents computational time from becoming excessive

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP3992976B1Compound property prediction method and apparatus, and computer device and readable storage medium
Publication Date: 2026.02.25 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3992976B1 patent drawingFigure 1
  • EP3992976B1 patent drawingFigure 2a~2b
  • EP3992976B1 patent drawingFigure 2c~2d

AI summary

Disclosed are a compound property prediction method and apparatus, and a computer device and a readable storage medium. Specifically, the method comprises: obtaining chemical structure information of a target compound, the chemical structure information comprising atoms and chemical bonds; generating a chemical structure graph corresponding to the chemical structure information according to the chemical structure information, the chemical structure graph comprising nodes corresponding to the atoms and edges corresponding to the chemical bonds; constructing original node features of the nodes and original edge features of the edges; performing a plurality of rounds of information propagation on the edges according to the original node features of the nodes and the original edge features of the edges to obtain propagation state information of the edges after the plurality of rounds of information propagation; obtaining target features of the edges according to the propagation state information; and predicting properties of the target compound according to the target features of the edges, and outputting a property prediction result of the target compound.