An Algorithm for the Discovery of Robust Protein Biomarkers with Application to the Design of Classification Protein Arrays
The application of high throughput proteomics to the discovery of protein expression patterns enabling accurate, sensitive and specific prediction of disease class and progression holds great promise. Typically, tissue samples are obtained from populations of patients sharing a specific disease diagnosis and two distinct patient populations ("classes") are selected for study. Second, protein expression patterns are measured in each of these samples. Third, having measured protein expression profiles in each patient and class, machine learning methods are used to train a decision-making algorithm. Lastly, proteins expression patterns are analyzed to discriminate disease from control samples. Biomarker discovery and validation is a complex process that can give false results because of the possible artifacts introduced when interpreting the results from 2D gel or mass spectroscopy experiments. The inventors use a robust analytical method, TSP (Top Scoring Pair) that was originally developed for nucleic acid microarray experiments to give more reliable biomarkers with smaller sample sizes. A machine learning algorithm defines the two protein markers that have the largest difference in expression between the control and diseased states. Description (Set) Proposed Use (Set) This invention can be used with a variety of proteomic methods including mass spectrometry, 2D gel electrophoresis, and DIGE (Difference Gel Electrophoresis) to identify possible biomarker proteins that change their expression levels in control and diseased states. This is a general purpose approach to the design of diagnostic protein arrays for a broad range of diseases in which the disease manifests itself through changes in protein expression in a tissue (such as a tumor, blood, etc) that could be procured through patient biopsy
Winslow, Raimond L.
Type of Offer:
« More Diagnostic Patents