Kernels and kernel methods for spectral data

• US 7,617,163 B2
• Filed: 10/09/2002
• Issued: 11/10/2009
• Est. Priority Date: 05/01/1998
• Status: Expired due to Term

• Current Assignee: Curtis Anderson, Health Discovery Corporation, James Roberts, Joe Mckenzie, Jules B. Paderewski, Julian N. Stern, Memorial Health Systems Incorporated, Timothy P. O'Hayer
• Original Assignee: Health Discovery Corporation
• Inventors: Weston, Jason Aaron Edward, Chapelle, Olivier, Elisseeff, Andre, Ben-Hur, Asa
• Primary Examiner(s): Vincent; David R
• Assistant Examiner(s): Wong; Lut
• Application Number:
  US10/267,977
• Publication Number:
  US 20050228591A1
• Time in Patent Office: 2,589 Days
• Field of Search: 706/25, 706/20, 706/26, 706/12, 706/45
• US Class Current: 706/12
• CPC Class Codes: G06F 18/21355 nonlinear criteria, e.g. embedding a manifold in a Euclidean space
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06V 10/761 Proximity, similarity or dissimilarity measures
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods

Abstract

Support vector machines are used to classify data contained within a structured dataset such as a plurality of signals generated by a spectral analyzer. The signals are pre-processed to ensure alignment of peaks across the spectra. Similarity measures are constructed to provide a basis for comparison of pairs of samples of the signal. A support vector machine is trained to discriminate between different classes of the samples. to identify the most predictive features within the spectra. In a preferred embodiment feature selection is performed to reduce the number of features that must be considered.

Claims

1. A method for analysis of data contained in a plurality of spectra generated from mass spectrographic measurement of protein samples corresponding to different biological conditions, wherein the different biological conditions have associated different levels of protein expression, the method comprising:

downloading the plurality of spectra into a computer system comprising a processor and a storage device, wherein the processor is programmed to execute at least one support vector machine and performs the steps of;

aligning the plurality of spectra, comprising;

selecting a first example spectrum as a baseline example;

sliding each spectral peak of a second example spectrum one at a time along a plurality of peaks within the baseline example;

applying a scoring function to obtain a similarity score between each spectral peak of the second example spectrum and the peaks within the baseline example, the similarity score being determined according to the relationship

2. The method of claim 1, wherein the proteins are extracted from serum samples and the different biological conditions comprise normal, benign and prostate cancer.

3. The method of claim 1, further comprising, before or after the step of aligning, normalizing each of the plurality of spectra.

4. The method of claim 1, wherein the kernel comprises a polynomial kernel of degree 2.

5. The method of claim 1, wherein the at least one support vector machine comprises a multi-class support vector machine.

6. The method of claim 1, further comprising, before or after the step of aligning, smoothing the plurality of spectra by averaging over a window.

7. The method of claim 1, further comprising, before or after the step of aligning, extracting pre-determined peaks from the plurality of spectra.

8. The method of claim 1, wherein the scoring of similarity is made noise invariant by discarding a lower portion of the plurality of spectra.

9. The method of claim 1, wherein the scoring of similarity is made noise invariant by using repeated measurements of the protein samples.

10. The method of claim 1, further comprising storing the listing in the storage device.

11. The method of claim 1, wherein the feature selection algorithm is selected from SVM-recursive feature elimination and l₀-norm Minimization.

12. A method for analysis of mass spectrographic data generated from protein samples comprising different biological conditions, the method comprising:

downloading a plurality of spectra comprising the mass spectrographic data into a computer system comprising a processor and a storage device, wherein the processor is programmed to execute at least one support vector machine and performs the steps of;

aligning spectra within the mass spectrographic data, comprising;

selecting a first example spectrum as a baseline example;

sliding each spectral peak of a second example spectrum one at a time along a plurality of peaks within the baseline example;

applying a scoring function to obtain a similarity score between the second example spectrum and the baseline example, the similarity score being determined according to the relationship

13. The method of claim 12, wherein the mass spectrographic data are generated by a surface emitting laser desorption ionization system.

14. The method of claim 12, further comprising, before or after the step of aligning, normalizing each spectrum.

15. The method of claim 12, wherein the kernel comprises a polynomial kernel of degree 2.

16. The method of claim 12, wherein the at least one support vector machine comprises a multi-class support vector machine.

17. The method of claim 12, further comprising, before or after the step of aligning, smoothing the plurality of spectra by averaging over a window.

18. The method of claim 12, further comprising, before or after the step of aligning, extracting pre-determined peaks from the plurality of spectra.

19. The method of claim 12, wherein the scoring of similarity is made noise invariant by discarding a lower portion of the plurality of spectra.

20. The method of claim 12, wherein the scoring of similarity is made noise invariant by using repeated measurements of the protein samples.

21. The method of claim 12, further comprising storing the listing in the storage device.

22. The method of claim 12, wherein the feature selection algorithm is selected from SVM-recursive feature elimination and l₀-norm Minimization.

23. A method for analysis of protein mass spectrographic data contained in a plurality of spectra generated from a plurality of protein samples comprising a plurality of different medical conditions, the data residing in a storage device, the method comprising:

aligning the plurality of spectra using a processor by constructing a similarity measure for comparing pairs of spectra with a baseline spectrum, the similarity measure being determined according to the relationship

24. The method of claim 23, further comprising storing the listing in the storage device.

25. The method of claim 23, wherein the feature selection algorithm is selected from SVM-recursive feature elimination and l₀-norm Minimization.