Protein Mutation Prediction Using 3D Structure and Amino Acid Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting virus mutation struggle to accurately reflect structural influences between distant amino acids and differences in amino acid properties, leading to degraded accuracy in predicting virus mutations.
Innovation Solution
A non-transitory computer-readable recording medium stores an information processing program that trains a regression model using machine-learning to predict amino-acid sequences after mutation by inputting data into a machine-learning model, incorporating features derived from the three-dimensional structure and properties of a virus.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional time-series analysis or LSTM methods are used to predict amino-acid sequences, then the prediction process is simple, but the accuracy is degraded due to inability to reflect structural influences between distant amino acids
Solution Approach 1:
The patent transitions from one-dimensional time-series analysis to three-dimensional structural analysis by incorporating the spatial coordinates and geometric relationships of amino acids in the protein structure. This allows the model to capture structural influences between distant amino acids that cannot be detected by sequential methods alone.
Solution Approach 2:
The patent segments the protein structure into individual amino acid units with their respective three-dimensional coordinates and properties. Each amino acid is treated as a discrete entity with features including spatial position, chemical properties, and structural role, allowing independent analysis and feature extraction for each unit.
2Measurement precision
If conventional methods ignore three-dimensional structure features, then the computational process is fast, but the prediction accuracy is degraded
Solution Approach 1:
The patent performs preliminary extraction of three-dimensional structure features and amino acid properties before the main prediction process. By pre-processing and organizing spatial coordinates, structural roles, and chemical properties of amino acids in advance, the model can efficiently access this information during prediction without excessive computational overhead.
Solution Approach 2:
The patent replaces conventional time-series analysis mechanisms with a machine learning model that directly processes three-dimensional structural features. This substitution enables the system to capture complex spatial relationships and structural influences that mechanical sequential processing cannot detect.
Data Source
AI summary
An computer-readable recording medium having stored therein an information processing program causes a computer to execute a process including: obtaining a third feature based on statistical information, the statistical information being obtained by prediction of each of amino acids included in a protein corresponding to input data including a first feature related to a three-dimensional structure of a protein of a virus and a second feature related to a property originated from the three-dimensional structure, the prediction being performed by inputting the input data into a machine-learning model, and training a regression model that predicts an amino-acid sequence of the virus after mutation using the second feature and the third feature as an input feature.