Treatment selection for lung cancer patients using mass spectrum of blood-based sample
A test for predicting whether a non-small-cell lung cancer patient is more likely to benefit from an EGFR-I as compared to chemotherapy uses a computer-implemented classifier operating on a mass spectrum of a blood-based sample obtained from the patient. The classifier makes use of a training set which includes mass spectral data from blood-based samples of other cancer patients who are members of a class of patients predicted to have overall survival benefit on EGFRI-Is, e.g., those patients testing VS Good under the test described in U.S. Pat. No. 7,736,905. This class-labeled group is further subdivided into two subsets, i.e., those patients which exhibited early (class label “early”) and late (class label “late”) progression of disease after administration of the EGFR-I in treatment of cancer.
1 . A method of generating a class label for a sample;
a) generating mass spectra of a development set of samples;
b) with the aid of a computer, generating a classifier from the mass spectra of the development set of samples;
c) obtaining a set of feature-dependent noise characteristics from the mass spectra of the development set of samples;
d) generating a mass spectrum of the sample;
e) generating a set of noisy feature value realizations of feature values of the mass-spectrum of the sample;
f) applying the classifier generated in step b) to the noisy feature value realizations and collating the results of the applying step;
g) generating statistical data on the results collated in step f); and
h) using the statistical data generated in step g) to determine a class label for the sample.
2 . The method of claim 1 , wherein the sample comprises a blood-based sample and wherein the development set of samples are in the form of a set of blood-based samples.
3 . The method of claim 1 , wherein the samples are obtained from a human with a disease.
4 . The method of claim 3 , wherein the disease is cancer.
5 . The method of claim 1 , wherein the noisy feature value realizations include both additive and multiplicative feature dependent noise characteristics.