Kernels and kernel methods for spectral data

• US 20050228591A1
• Filed: 10/09/2002
• Published: 10/13/2005
• Est. Priority Date: 05/01/1998
• Status: Active Grant

• 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: Curtis Anderson, Garry L. Carls, Glynn Bergeron, James Roberts, Joe Mckenzie, John E. Matthews, Jules B. Paderewski, Julian N. Stern, Memorial Health Systems Incorporated, Peter J. Farley, Simpson K. Russell, Timothy P. O'Hayer
• Inventors: Chapelle, Olivier, Weston, Jason, Ellisseeff, Andre, Hur, Asa
• Granted Patent: US 7,617,163 B2
• Time in Patent Office: Days
• US Class Current: 702/19
• 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 preprocessed 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 a plurality of samples comprising a plurality of different sample classes, the method comprising:

aligning the plurality of spectra;

constructing a similarity measure for comparing pairs of samples;

training at least one support vector machine to discriminate between the plurality of different sample classes, wherein the at least one support vector machine comprises a kernel;

processing the plurality of spectra using the at least one support vector machine;

identifying at least one predictive feature within the plurality of spectra.

2. The method of claim 1, further comprising, after aligning the plurality of spectra, preprocessing the plurality of spectra to select features.

3. The method of claim 1, wherein the step of aligning comprises sliding each spectrum of the plurality of spectra along a plurality of different feature values against a baseline example and rating a similarity between the spectra and the baseline example.

4. The method of claim 1, wherein the data is selected from the group consisting of DNA sequences, protein sequences, documents, graphs, signals, microarray expression profiles, spectra, images, spatio-temporal data and relational data.

5. The method of claim 1, wherein the plurality of spectra comprises mass spectrograms of proteins.

6. The method of claim 5, wherein the proteins are from serum samples for diagnosis of prostate cancer.

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

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

9. The method of claim 1, wherein the at least one support vector machine comprises a multiclass support vector machine.

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

11. The method of claim 1, further comprising, before or after the step of aligning, extracting predetermined peaks from the plurality of spectra.

12. The method of claim 1, wherein the similarity measure is noise invariant.

13. The method of claim 12, wherein similarity measure if calculated from technical repeats of the plurality of spectra.