3D Molecular Graph Training for PROTAC Property Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods lack an effective way to accurately predict molecular properties of proteolysis targeting chimeras, which are crucial for drug discovery and protein design, particularly in areas like cancer treatment.

Innovation Solution

A model training method that constructs three-dimensional molecular graph data of proteolysis targeting chimeras, using atomic nodes and adjacency matrices, and trains a prediction model to predict properties like lipophilicity and solubility by minimizing deviations between predicted and actual molecular properties.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional methods are used for molecular property prediction, then the process is simple, but the accuracy is insufficient

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

Solution Approach 1:

The patent transitions from traditional two-dimensional molecular representations to three-dimensional molecular graph data, incorporating spatial coordinates and geometric information. This dimensional enhancement enables the model to capture spatial relationships and molecular conformations, significantly improving prediction accuracy for molecular properties while managing model complexity through structured data organization.

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

Solution Approach 2:

The patent transforms molecular data from simple connectivity information to comprehensive three-dimensional parameters including atomic coordinates, bond lengths, bond angles, and spatial relationships. By changing the parameter representation from 2D to 3D space, the model achieves higher prediction accuracy for molecular properties such as solubility, permeability, and binding affinity.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If complex prediction models are developed, then prediction accuracy improves, but computational efficiency decreases

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the molecular system into individual atomic nodes and molecular subgraphs, each processed independently through neural network layers. This segmentation allows parallel computation across different atomic and molecular components, maintaining high prediction accuracy while improving computational efficiency through distributed processing and reduced complexity per computational unit.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

By organizing molecular data in three-dimensional space with explicit coordinate information, the patent enables more efficient spatial queries and distance calculations compared to traditional methods. The structured 3D representation allows the model to process spatial relationships more efficiently, balancing accuracy with computational performance.

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

3Measurement precision

If three-dimensional molecular graph data is constructed, then prediction accuracy improves, but data processing complexity increases

Engineering Contradiction:
Improveprediction accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the three-dimensional molecular graph into atomic nodes with individual feature vectors and molecular subgraphs with structured relationship data. This segmentation simplifies data processing by breaking down complex 3D molecular structures into manageable units that can be processed independently and then aggregated, maintaining accuracy while reducing overall processing complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms raw molecular data into standardized three-dimensional parameters including atomic coordinates, bond vectors, and geometric features. By establishing consistent parameter representations and coordinate systems, the patent simplifies subsequent data processing operations while capturing the essential spatial information needed for accurate molecular property prediction.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4708301A1Model training method and apparatus, and molecular property information prediction method and apparatus
Publication Date: 2026.03.11 ZHEJIANG LAB
  • EP4708301A1 patent drawingFigure 1~2
  • EP4708301A1 patent drawingFigure 3~4
  • EP4708301A1 patent drawingFigure 5~6

AI summary

The present application provides a model training method and apparatus, and a molecular property information prediction method and apparatus. The method include: obtaining data of a designated proteolysis targeting chimera molecule; according to the data, constructing three-dimensional molecular graph data of the designated proteolysis targeting chimera molecule; inputting the three-dimensional molecular graph data of the designated proteolysis targeting chimera molecule into a prediction model to be trained, so that the prediction model predicts molecular property information of the designated proteolysis targeting chimera molecule; and according to a deviation between predicted molecular property information and actual molecular property information of the designated proteolysis targeting chimera molecule, training the prediction model.