|
Abstract:
|
The development of sound statistical algorithms for mass spectrometry data will advance biomarker discovery, post-translational modification characterization, bacterial identifications and more. In this presentation, we will talk about two important developments: 1) a likelihood-based bacterial identification approach for bi-microbial mass spectrometry data and 2) a missing data imputation approach for quantitative mass spectrometry data. For mass-spectrometry-based bimicrobial identifications, we propose a two-component mixture model that utilizes information about a co-existence of two species in one mass spectrum. An observed bi-microbial mass spectrum is fitted by multiple candidate probabilistic models that are constructed using two mono-microbial reference mass spectra. Then, a posterior probability is used to measure our confidence about bacterial identifications. In this study, we demonstrate the superior performance of the proposed approach compared model-free approaches using two datasets. For the missing data imputation study, we implement a penalized EM algorithm approach to impute missing peptide abundances and evaluate its performance using standard mixture datasets.
|
