Bioinformatics in tear proteome

Tear is a complex fluid containing rich mixture of biomolecules. Current mass spectrometry based proteomic techniques are able to capture the expression of thousands of proteins from small tear volumes, opening possibility for data driven identification of predictive biomarkers. We have developed a bioinformatic approach for biomarker discovery that effectively finds associations between time course proteomic data and clinical phenotypic data. We first construct a feature matrix that contains the data and clinical information as well as various computationally derived features, for example pathway enrichments scores across thousands of pathways and various biologically motivated activity scores for master regulators. These feature matrices that can contain categorical, discrete or continuous data are then computationally analyzed to identify the strongest statistical associations with the phenotypes of interest. Cross validation strategy is utilized to make sure the identified associations are robust. Our approach has been applied to data from clinical patient samples to identify new candidate biomarkers.