Blind Source Separation in NMR Spectroscopy via Sparse Component Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for blind source separation in spectroscopy and spectrometry, such as ICA and NMF, require a number of linearly independent mixtures to be greater than or equal to the number of pure components and assume statistical independence among components, which are not always met in real-world applications, especially with complex chemical compounds and biomolecules.

Innovation Solution

The use of sparse component analysis (SCA) combined with the detection of single component points (SCPs) to estimate the number of pure components and concentration matrix, even when the number of mixtures is less than the number of components, by transforming data into wavelet or analytical continuation domains to increase sparseness and apply a direction-based detection criterion.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If standard BSS methods (ICA or NMF) are used for blind source separation, then the separation can be performed under certain assumptions, but the method fails when the number of mixtures is less than the number of pure components

Engineering Contradiction:
Improvereliability of blind source separationVSAvoidapplicability to underdetermined cases
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The invention transforms the data from the original domain to a transformed domain (e.g., wavelet domain, frequency domain, or other appropriate domains) where the pure components exhibit sparser representations. This parameter transformation enables the application of sparsity constraints that make the underdetermined blind source separation problem solvable, allowing extraction of more pure components than the number of available mixtures.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The invention introduces a new dimension by transforming the data into a different domain (spatial, frequency, time-frequency, or other transformed domains). This dimensional transformation creates additional structure and sparsity patterns that were not visible in the original domain, enabling the separation of more components than mixtures by exploiting the sparsity in the transformed domain.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Ease of manufacture

If ICA algorithms are used, then statistical independence assumption can be made, but this assumption is not met in many spectroscopic applications

Engineering Contradiction:
Improvesimplicity of algorithm implementationVSAvoidvalidity of statistical independence assumption
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The invention changes the parameter representation by transforming data into domains where components exhibit sparsity rather than relying on statistical independence. The sparsity constraint (many coefficients being zero or near-zero) is a more realistic assumption for spectroscopic data than statistical independence, as spectral components often overlap but can be separated in transformed domains where each component occupies distinct regions.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If NMF methods are used, then nonnegativity and sparseness requirements are imposed, but these requirements are not satisfied simultaneously in majority of spectroscopic applications

Engineering Contradiction:
Improvepurity of extracted componentsVSAvoidapplicability to real spectroscopic data
Core Design Contradiction:
Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

The invention transforms the data into appropriate domains where the pure components naturally exhibit both nonnegativity and sparsity properties. In the transformed domain, the sparsity constraint can be satisfied while maintaining nonnegativity, allowing NMF to effectively separate more components than mixtures without requiring the components to be simultaneously nonnegative and sparse in the original domain.

Inventive Principle:
Principle #35Parameter changes

4Measurement precision

If library-based approach is used for identification, then reference matching can be performed, but this approach is ineffective when library content is limited or pure component spectra are unavailable

Engineering Contradiction:
Improveaccuracy of compound identificationVSAvoidcapability to handle unknown compounds
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The invention extracts pure component spectra directly from the mixture spectra through blind source separation in transformed domains, eliminating the need for external library references. By separating the mixed spectra into their constituent pure components using sparsity constraints and domain transformation, the method enables identification of unknown compounds without requiring pre-existing library spectra.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP2476008B1Underdetermined blind extraction of components from mixtures in 1d and 2d NMR spectroscopy and mass spectrometry by means of combined sparse component analysis and detection of single component points
Publication Date: 2015.04.29 RUDJER BOSKOVIC INST
  • EP2476008B1 patent drawingFigure 1
  • EP2476008B1 patent drawingFigure 2A~2B
  • EP2476008B1 patent drawingFigure 2C~2D

AI summary

The present invention relates to the blind extraction of pure components from mixtures recorded in the fields of ID or 2D NMR spectroscopy and mass spectrometry. Specifically, the invention is related to the application of sparse component analysis in combination with detection of single component points to blind decomposition of NMR spectroscopy or mass spectrometry data X into pure components S and concentration matrix A, when the number of pure components S is greater than the number of mixtures X. NMR mixtures are transformed into wavelet domain, since pure components int the wavelet domain are sparser than in the recording domain. By means of a direction based criterion single component points (SCPs) of the mixtures in the wavelet domain are detected where only one pure component is active. These SCPs are used for the estimation of the unknown number of pure components and for the estimation of the concentration matrix by means of data clustering methods. The pure components are estimated in frequency domain, e.g. by means of linear programming. Mass spectrometry mixture signals are extended by analytical continuation which is necessary to obtain a complex signal as is required by direction based SCPs detection criterion. The estimated pure components are ranked using a negentropy-based criterion.