Computer-implemented methods and systems for identifying a species from mass spectra
One aspect of the invention provides a computer-implemented method of identifying a species from mass spectra. The computer-implemented method includes: loading a data set including a species and associated mass spectra into memory on a computer; for each of the associated mass spectra, pre-processing to identify ranks of intensities of an in-bin peak across bins defining ranges of mass-to-charge ratio; and training a species binary classifier for each of a plurality of species using at least the ranks of intensities. Another aspect of the invention provides a system for identifying a species from mass spectra. The system includes: a processor; and computer-readable memory containing instructions to: implement an interface programmed to receive mass spectra; store the mass spectra in the computer-readable memory; and invoke execution of any of the methods described herein on the processor.
1. A computer-implemented method of identifying a species and phenotype from mass spectra, the computer-implemented method comprising:
(a) loading a data set into memory on a computer, the data set comprising mass spectra;
(b) pre-processing the mass spectra to identify at least ranks of intensities across bins defining ranges of mass-to-charge ratio;
(c) applying species binary classifiers against at least the ranks of intensities;
(d) identifying one or more species based on step (c);
(e) calculating a False-Discovery Rate (FDR) for the one or more species using decoy members;
(f) if the FDR is less than a pre-defined threshold, applying phenotype binary classifiers against at least the ranks of intensities; and
(g) identifying one or more phenotypes based on step (f);
wherein the species binary classifiers and the phenotype binary classifiers are trained using a machine-learning method, the machine-learning method comprising:
loading a data set into memory on a computer, the data set comprising:
a species and associated mass spectra; and
a phenotype for each member of the data set;
for each of the associated mass spectra, pre-processing to identify ranks of intensities of an in-bin peak across bins defining ranges of mass-to-charge ratio;
training a species binary classifier for each of a plurality of species using at least the ranks of intensities; and
training a phenotype binary classifier for each of a plurality of phenotypes.
2. The computer-implemented method of claim 1 , wherein the machine learning method further comprises:
introducing decoy members into the data set.
3. The computer-implemented method of claim 2 , wherein the decoy members include shuffled ranks of intensities.
4. The computer-implemented method of claim 1 , wherein the mass spectra were generated using negative ion MALDI-TOF-MS (Matrix-Assisted Laser Desorption/Ionization—Time Of Flight—Mass Spectrometry) analysis.
5. The computer-implemented method of claim 1 , wherein the intensities are relative intensities.
6. The computer-implemented method of claim 1 , wherein the species are bacterial species.
7. The computer-implemented method of claim 1 , wherein the species are selected from the group consisting of: mouse, primate, human, mammal, and animal.
8. The computer-implemented method of claim 1 , wherein the mass spectra are glycolipid mass spectra.
9. The computer-implemented method of claim 1 , wherein the mass spectra are selected from the group consisting of: protein/peptide mass spectra and metabolite mass spectra.
10. The computer-implemented method of claim 1 , wherein the applying step identifies a plurality of species and phenotypes for the single mass spectrum.
11. A system for identifying a species from mass spectra, the system comprising:
a processor; and
computer-readable memory containing instructions to:
implement an interface programmed to receive mass spectra;
store the mass spectra in the computer-readable memory; and
invoke execution of a computer-implemented method of identifying a species and phenotype from mass spectra on the processor, the computer-implemented method comprising:
(a) loading the mass spectra;
(b) pre-processing the mass spectra to identify at least ranks of intensities across bins defining ranges of mass-to-charge ratio; and
(c) applying species binary classifiers against at least the ranks of intensities;
(d) identifying one or more species based on step (c);
(e) calculating a False-Discovery Rate (FDR) for the one or more species using decoy members;
(f) if the FDR is less than a pre-defined threshold, applying phenotype binary classifiers against at least the ranks of intensities; and
(g) identifying one or more phenotypes based on step (f);
wherein the species binary classifiers and the phenotype binary classifiers are trained using a machine-learning method, the machine-learning method comprising:
loading a data set into memory on a computer, the data set comprising:
a species and associated mass spectra; and
a phenotype for each member of the data set;
for each of the associated mass spectra, pre-processing to identify ranks of intensities of an in-bin peak across bins defining ranges of mass-to-charge ratio;
training a species binary classifier for each of a plurality of species using at least the ranks of intensities; and
training a phenotype binary classifier for each of a plurality of phenotypes.