Multi-layer Network Feature Extraction for Omics Data Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in estimating disease-related variables, such as disease-free survival time, from omics data is exacerbated by the 'curse of dimensionality' due to a limited number of samples compared to the vast number of features in next-generation sequencing data, leading to over-adaptation and loss of generalization in machine learning models.
Innovation Solution
A method involving a multi-layer network approach where similarity values and weights are calculated between omics data nodes, with non-salient connections removed using backboning, and communities identified to extract relevant features using UMAP, reducing dimensionality and improving model generalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If machine learning models are trained using all available omics data features, then the completeness of information analysis is improved, but the model suffers from over-adaptation and loss of generalization capacity due to the curse of dimensionality
Solution Approach 1:
The patent extracts only the most relevant features from the vast omics data using feature selection techniques. This involves identifying and retaining a subset of features that have the highest predictive value while discarding redundant or less informative features, thereby resolving the contradiction between information completeness and model generalization.
Solution Approach 2:
The patent applies different processing strategies to different features based on their individual characteristics. Rather than treating all features uniformly, the system identifies specific features with high predictive power and applies targeted analysis methods to them, allowing the model to focus computational resources on the most informative data points.
2Reliability
If the number of features is reduced to solve the curse of dimensionality, then model generalization is improved, but the amount of information available for analysis is reduced
Solution Approach 1:
The patent merges multiple sources of information and integrates different types of omics data (genomics, transcriptomics, proteomics, metabolomics) into a unified feature set. This integration allows the model to capture complementary information from different biological layers while maintaining a manageable feature dimensionality.
Solution Approach 2:
The patent creates composite features by combining information from multiple omics layers. These composite features capture complex biological relationships and interactions that would be missed by analyzing individual omics data types separately, thereby preserving information completeness while reducing overall dimensionality.
3Adaptability or versatility
If all types of omics data (genomics, transcriptomics, proteomics, metabolomics) are integrated, then the comprehensiveness of disease analysis is improved, but the complexity of data processing and model training increases
Solution Approach 1:
The patent segments the integration process into distinct stages, each handling a specific type of omics data or a specific analytical task. This modular approach allows the system to process different data types independently before integrating them, reducing the overall computational complexity while maintaining comprehensive disease analysis capabilities.
Solution Approach 2:
The patent transforms multi-omics data from multiple dimensions (different data types) into a unified feature space. By projecting diverse omics data into a common dimensional framework, the system enables comprehensive analysis while simplifying the integration process and reducing processing complexity.
Data Source
AI summary
Solutions for estimating a variable of interest associated to a given disease as a function of omics data of a patient are provided. A first and second dataset of omics data is received. A multi-layer network is generated. Non-salient intra-omics connections and inter-omics connections of the multi-layer network are pruned. A plurality of communities of the multi-layer network are identified and one or more respective features are determined for each community. A training dataset is generated and a classifier configured for estimating the value of the variable of interest using the training dataset.