CNN Omics Data Mapping via Spatial Ordering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for analyzing gene expression data, such as transcriptomics, lack an easily interpretable visual representation of the whole dataset for all samples, making it difficult to understand perturbations in individual genes and gene sets, and do not provide a meaningful ordering that is necessary for effective processing by convolutional neural networks (CNNs).
Innovation Solution
The development of an apparatus and method that uses convolutional neural network (CNN) layers to map omics input data to N-dimensional data, transforming it into spatially ordered data, which can then be processed by a task-specific model to obtain a task-specific output, allowing for a visually interpretable representation of omics data as a two-dimensional image format that preserves the relationships between biomolecules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If gene expression data is stored and analyzed in traditional tabular formats (e.g., spreadsheets with genes as rows and samples as columns), then the data can be easily stored and processed by conventional statistical methods, but the data lacks a meaningful spatial ordering that enables visual interpretation and effective processing by convolutional neural networks
Solution Approach 1:
The patent applies dimensionality change by transforming one-dimensional gene expression data into two-dimensional spatial representations. Specifically, it uses manifold learning techniques to embed high-dimensional transcriptome data into 2D space, creating visual maps where spatial proximity reflects functional similarity. This allows the data to be both visually interpretable and suitable for CNN processing while preserving the underlying biological relationships
2Loss of information
If differential gene expression analysis and gene set enrichment analysis are performed on transcriptomic datasets, then perturbations in individual genes and gene sets can be understood, but an easily interpretable visual representation of the whole dataset for all samples is not provided
Solution Approach 1:
The patent segments the high-dimensional transcriptome data into meaningful spatial regions through manifold learning visualization. By embedding genes into 2D space based on expression similarity, it creates visual clusters that represent functional groups or pathways. This segmentation allows simultaneous visualization of individual gene perturbations and overall dataset patterns, maintaining information completeness while improving visual interpretability
Solution Approach 2:
The patent introduces manifold learning embedding as an intermediary transformation between traditional tabular data and visual representation. This intermediary step converts numerical expression values into spatial coordinates that preserve relationships, enabling both quantitative analysis and visual interpretation without losing information from the original dataset
3Ease of operation
If gene expression data is ordered alphabetically by gene name or by chromosome position, then the data is systematically organized, but this ordering has little functional significance and does not reflect biological relationships
Solution Approach 1:
The patent changes the ordering parameter from arbitrary criteria (alphabetical or chromosomal position) to functionally meaningful criteria based on expression similarity. By using manifold learning to compute spatial coordinates that reflect biological relationships, the data maintains systematic organization while the ordering now captures functional relationships, allowing both ease of operation and preservation of functional information
Data Source
AI summary
An apparatus comprises processing circuitry configured to: acquire convolutional neural network (CNN) layers; and train a first mapping layer which connects to an input layer of the CNN layers, and a second mapping layer which connects to an output layer of the CNN layers, wherein the first mapping layer maps omics input data to N-dimensional data, and wherein the second mapping layer receives further N-dimensional data that is output by the CNN layers and maps the further N-dimensional data to omics output data; wherein the training of the first mapping layer and the second mapping layer comprises fixing parameters of the CNN layers and minimizing a loss function, wherein the loss function is dependent on the input omics data and the output omics data.