Regression Framework for Gene Expression Data Transformation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods struggle to achieve cross-platform compatibility of gene expression data, limiting the integration of legacy microarray studies with new Next Generation Sequencing (NGS) systems due to platform-specific dynamic ranges and requiring re-profiling of samples for combined analysis.

Innovation Solution

A data-driven, regression-based framework transforms gene expression measurements across different platforms using primary and categorical models, eliminating the need for re-profiling and enabling the use of legacy data with newer platforms by adapting gene signatures and transforming input data or parameters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If quantile normalization is applied to combine expression data from heterogeneous platforms, then the expression distributions across platforms can be unified, but the method assumes that expression distributions only differ by a sample-specific scaling factor which limits its applicability to extreme expressions

Engineering Contradiction:
Improveexpression distribution unificationVSAvoidplatform compatibility range
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent transforms expression data by applying platform-specific parameter adjustments through regression models. Instead of assuming uniform scaling factors, the method learns platform-specific transformation parameters from paired training data, allowing accurate transformation across diverse platforms including those with extreme expressions. This resolves the contradiction by changing from fixed normalization parameters to learned transformation parameters that adapt to each platform's characteristics.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If gene-by-gene factor analysis with EM algorithms is used to obtain unified expression measures from multiple platforms, then a unified expression measure can be obtained, but genes with extreme expressions are excluded from the analysis

Engineering Contradiction:
Improveunified expression measureVSAvoidnumber of genes included
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent employs dynamic regression models that adapt to each gene's expression characteristics rather than applying a static factor analysis framework. The transformation parameters are learned individually for each gene from paired training data, allowing genes with extreme expressions to be included with appropriate platform-specific adjustments. This dynamic approach replaces the rigid EM algorithm framework, resolving the contradiction between measurement precision and gene inclusion quantity.

Inventive Principle:
Principle #15Dynamics

3Measurement precision

If a modified RMA algorithm is applied to probe-region expressions to obtain gene-level expressions, then gene-level expressions can be obtained from RNA-Seq data, but the method is rigid in terms of platform choice (RNA-Seq for input and RMA for output)

Engineering Contradiction:
Improvegene-level expression accuracyVSAvoidplatform flexibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal transformation framework that can handle multiple platform combinations through learned regression models. Instead of being restricted to RNA-Seq to microarray transformations, the method learns platform-specific parameters from paired training data of any platform combination, making the transformation process universally applicable. This multi-functional approach resolves the contradiction by making the system adaptable to various platform pairs while maintaining gene-level expression accuracy.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Loss of time

If legacy microarray data is integrated with new NGS data using existing methods, then some cross-platform compatibility can be achieved, but re-profiling of samples is still required for combined analysis

Engineering Contradiction:
Improvedata integration timeVSAvoidre-profiling cost
Core Design Contradiction:
Loss of timeVSLoss of energy

Solution Approach 1:

The patent performs preliminary transformation of legacy data to match the target platform's expression distribution before combination. By pre-transforming the legacy microarray data using learned platform-specific parameters, the method enables direct integration with NGS data without requiring re-profiling of samples. This preliminary action resolves the contradiction by eliminating the need for costly re-profiling while achieving accurate cross-platform integration.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3207133B1Cross-platform transformation of gene expression data
Publication Date: 2021.05.05 KONINKLIJKE PHILIPS NV
  • EP3207133B1 patent drawingFigure 1
  • EP3207133B1 patent drawingFigure 2
  • EP3207133B1 patent drawingFigure 3

AI summary

Data-driven generalized regression-based frameworks that support the transformation of measurements, applicable but not limited to gene expressions, from one platform to another over a wide dynamic range, with selected summary statistics / feature values as predictors for the model parameters. The framework consists of primary model training and transformation, and additional levels of categorical regression and transformation processes.