Genomic Feature Selection for Cancer Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current classification models for cancer diagnosis face challenges due to the 'curse of dimensionality' in genomic data, where the large number of features relative to the number of samples leads to inaccurate and computationally expensive analysis, especially in distinguishing between similar cancer types like lung cancer subtypes.
Innovation Solution
A method is introduced to generate a lower-dimensionality vector for training classification models by pre-processing genomic data, assigning genetic identifier values, and performing dimensionality reduction using these values to create a more manageable and informative vector, which can include clustering and parameterization of elements to represent clusters effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical regression methods are used to analyze genomic data with many features, then classification accuracy may improve, but the number of training samples required increases significantly
Solution Approach 1:
The patent extracts and selects only the most relevant genetic features from the large set of genomic data features. By identifying and retaining key features that are most predictive of cancer types while discarding redundant or less informative features, the method reduces the feature dimensionality to a manageable subset that maintains classification accuracy with fewer training samples
Solution Approach 2:
The patent introduces an intermediary feature selection and dimensionality reduction process between the raw genomic data and the classification model. This intermediary step uses various techniques (filter methods, wrapper methods, embedded methods) to transform the high-dimensional genomic data into a lower-dimensional representation that is more suitable for statistical regression analysis with limited training samples
2Productivity
If the number of features in genomic data is reduced, then computational cost decreases, but information loss may occur
Solution Approach 1:
The patent applies local quality by treating different features differently based on their importance and relevance to cancer classification. Rather than uniformly reducing all features, the method identifies specific high-value features that should be retained and applies selective dimensionality reduction techniques that preserve locally important information while reducing overall dimensionality
Solution Approach 2:
The patent performs preliminary feature selection and evaluation before the main classification task. By pre-identifying and ranking features based on their predictive power, biological relevance, and redundancy, the method prepares the data in advance to ensure that only the most informative features are retained, minimizing information loss before computational classification begins
3Adaptability or versatility
If more features are included in the analysis, then classification completeness improves, but the complexity of the analysis increases
Solution Approach 1:
The patent segments the large set of genomic features into meaningful groups or categories (e.g., by biological pathway, gene function, chromosomal location, or disease relevance). This segmentation allows the analysis to be organized into manageable subsets, reducing overall complexity while maintaining comprehensive coverage of important features through structured, modular analysis
Data Source
Figure 1(a)~1(b)
Figure 2
Figure 3
AI summary
The invention relates to methods for generating a lower-dimensionality vector for training a classification model and uses thereof.