Method for detecting subpopulations in spectral analysis

• US 5,121,338 A
• Filed: 07/11/1991
• Issued: 06/09/1992
• Est. Priority Date: 03/10/1988
• Status: Expired due to Fees

• Current Assignee: ADVANCED RESEARCH & TECHNOLOGY INSTITUTE
• Original Assignee: Indiana University Foundation (Indiana University)
• Inventors: Lodder, Robert A.
• Primary Examiner(s): Dixon, Joseph L.
• Application Number:
  US07/730,394
• Time in Patent Office: 334 Days
• Field of Search: 364/554, 364/552, 364/496-499, 364/413.01, 364/413.05, 364/525, 250/338.1, 250/338.5, 250/339-343, 356/30, 356/36, 356/319, 356/328, 356/334, 382/6
• US Class Current: 702/30
• CPC Class Codes: G01N 21/359 using near infrared light

Abstract

A method for using spectral analysis to detect subpopulations is disclosed. According to this method, a training set of spectra of a first plurality of samples is obtained and a bootstrap distribution is formed. A test set of spectra of a second plurality of samples is then obtained and a second bootstrap distribution is formed. First and second univariate distributions are then formed from the bootstrap distributions. A quantile-quantile relationship of the training and test sets is then developed and a determination of whether the test set and training set are substantially identical is made. The plurality of test samples are thus used to calculate probability density contours inside a training-set spectral cluster and detect perturbations of those contours using a bootstrap procedure. False samples are detected as subclusters well inside the training set, and trace analyses using a very small number of wavelengths are facilitated.

Claims

1. A method for detecting nonhomogeneous samples comprising the steps of:

assembling a first plurality of known samples to provide a training set of known samples;

sequentially placing each of the training set of known samples in a spectroscopic apparatus for measuring a property of the sample as a function of an observable parameter;

measuring the property of the known samples to obtain a training set of spectra at a second plurality of values of the observable parameters;

storing said training set of spectra;

forming a first bootstrap distribution from the training set;

measuring the property of a third plurality of test samples at the second plurality of values of the observable parameters to obtain a set of test spectra;

storing said set of test spectra;

forming a second bootstrap distribution from the test set;

forming first and second univariate distributions from the first and second bootstrap distributions;

calculating a quantile-quantile relationship of the training and test sets; and

determining whether the test set was drawn from a population substantially identical to the population from which the training set was drawn based on the quantile-quantile relationship.

2. The method of claim 1, wherein the number of the first plurality of samples essentially equals the number of the third plurality of samples.

3. The method of claim 1, further comprising the steps of collecting a validation set of spectra of a plurality of samples from the population from which the training set was collected, forming a third bootstrap distribution from the validation set, forming a third univariate distribution from the third bootstrap distribution, calculating a quantile-quantile relationship of the training and validation sets, and validating the training set based on the quantile-quantile relationship of the training and validation sets.

4. The method of claim 1 wherein the determining step includes calculating a correlation coefficient from the quantile-quantile relationship.

5. The method of claim 1, wherein the spectroscopic apparatus is a spectrometer, and the observable parameter is wavelength.

6. The method of claim 5, wherein the spectrometer is a near-infrared spectrometer.

7. The method of claim 1, wherein the spectroscopic apparatus is a mass spectrometer, and the observable parameter is mass.

8. A method for detecting nonhomogeneous samples of pharmaceutical capsules comprising the steps of:

assembling a first plurality of known samples from a population of pharmaceutical capsules to provide a training set of known samples;

sequentially placing each of the training set of known samples in a spectroscopic apparatus for measuring a property of the sample as a function of an observable parameter;

measuring the property of the known samples to obtain a training set of spectra at a second plurality of values of the observable parameters;

storing the training set of spectra;

forming a first bootstrap distribution from the training set;

obtaining a test set of spectra at the second plurality of values of the observable parameters from a third plurality of samples of pharmaceutical capsules;

forming a second bootstrap distribution from the test set;

forming first and second univariate distributions from the first and second bootstrap distributions;

calculating a quantile-quantile relationship of the training and test sets; and

determining whether the test set was drawn from a population substantially identical to the population from which the training set was drawn based on the quantile-quantile relationship.

9. The method of claim 8, wherein the determining step includes calculating a correlation coefficient from the quantile-quantile relationship.

10. The method of claim 9, further comprising the steps of collecting a validation set of spectra of a plurality of samples from the population from which the training set was collected, forming a third bootstrap distribution from the validation set, forming a third univariate distribution from the third bootstrap distribution, calculating a quantile-quantile relationship of the training and validation sets, and validating the training set based on the quantile-quantile relationship of the training and validation sets.

11. The method of claim 8, wherein the spectroscopic apparatus is a spectrometer, and the observable parameter is wavelength.

12. The method of claim 11, wherein the spectrometer is a near-infrared spectrometer.

13. The method of claim 8, wherein the spectroscopic apparatus is a mass spectrometer, and the observable parameter is mass.