Understanding the major functions of the cell requires accurate
measurement and characterization of its main biochemical actors, proteins.
Shotgun proteomics using liquid chromatography coupled with tandem mass
spectrometry offers the ability to obtain a comprehensive view of many
of a cell's proteins in a single experiment. A major step in shotgun
proteomics is the computational analysis of the tandem mass spectrum
to determine the originating peptide. This talk will present two methods
to improve this identification step using a machine learning.
The first method uses a support vector regressor to learn to predict peptide
chromatographic retention time. These predictions can in turn be used to
eliminate peptide identifications that are likely to be incorrect. The
second method presents a set of dynamic Bayesian networks that model
peptide fragmentation chemistry. These models provide insight into
peptide fragmentation chemistry as well as a further improvement in
discriminating correct from incorrect peptide identifications.