Protein Structure Prediction Using Sequence Feature Augmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing protein structure prediction methods face inefficiencies due to the need for large databases, leading to prolonged query times and reduced accuracy, and do not adequately account for positional information and biological evolution in amino acid sequences.

Innovation Solution

A method using a sequence feature augmentation model trained on both a first and second database, where the second database has a larger data scale, to enhance the initial sequence features, allowing for accurate prediction of protein structure information with reduced database query times.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a database with a larger data scale is used to extract sequence features, then the accuracy of sequence feature extraction is improved, but the query operation takes a long time, leading to lower prediction efficiency

Engineering Contradiction:
Improveaccuracy of sequence feature extractionVSAvoidquery operation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent pre-trains a sequence feature augmentation model using multi-sequence alignment data from a large database to learn evolutionary patterns and positional information. This preliminary action allows the model to generate augmented sequence features quickly during prediction without requiring real-time queries to the large database, thus resolving the contradiction between extraction accuracy and query time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses the sequence feature augmentation model to copy and generate augmented sequence features based on the initial sequence features extracted from a smaller database. This copying mechanism allows the system to leverage patterns learned from large-scale data without requiring actual access to the large database during prediction, achieving both accuracy and efficiency.

Inventive Principle:
Principle #26Copying

2Measurement precision

If a database with a larger data scale is used to extract sequence features, then the accuracy of protein structure information prediction is improved, but the prediction efficiency is reduced

Engineering Contradiction:
Improveaccuracy of protein structure information predictionVSAvoidprediction efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The sequence feature augmentation model is pre-trained in advance using data from a large database to capture evolutionary patterns and positional information. This preliminary training enables the model to generate accurate augmented sequence features during prediction without requiring real-time access to the large database, thus improving prediction accuracy while maintaining high prediction efficiency.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The sequence feature augmentation model acts as an intermediary between the initial sequence features and the final prediction. It processes the initial features to generate augmented features that incorporate evolutionary patterns learned from large-scale data, enabling accurate predictions without direct access to the large database during the prediction phase.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Device complexity

If traditional methods are used that do not account for positional information and biological evolution, then the method complexity is reduced, but the prediction accuracy is insufficient

Engineering Contradiction:
Improvemethod complexityVSAvoidprediction accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent transforms the initial sequence features into augmented sequence features by incorporating evolutionary patterns and positional information as additional parameters. The sequence feature augmentation model learns to generate these enhanced features during pre-training, improving prediction accuracy while maintaining relatively simple prediction-time operations.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The model copies and enhances sequence features by generating augmented versions that include positional and evolutionary information. This copying process enriches the feature representation without requiring complex real-time calculations during prediction, balancing accuracy and complexity.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4009328B1Method, device and apparatus for predicting protein structure information, and storage medium
Publication Date: 2025.06.25 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP4009328B1 patent drawingFigure 1~2
  • EP4009328B1 patent drawingFigure 3
  • EP4009328B1 patent drawingFigure 4

AI summary

A method, device and apparatus for predicting protein structure information, and a storage medium, which relate to the technical field of biological information. The method comprises: performing a sequence alignment query according to an amino acid sequence of a protein in a first database to obtain multi-sequence alignment data, performing feature extraction on the multi-sequence alignment data to obtain an initial sequence feature, processing the initial sequence feature through a sequence feature amplification model to obtain an amplified sequence feature of the protein, and then predicting the structure information of the protein according to the amplified sequence feature. The described solution can improve the prediction efficiency of the protein structure information under the condition that the prediction accuracy of the protein structure information is guaranteed on the basis of the AI-predicted protein structure information.