Oliver Serang

Joined Program: 2006
Previous Degree: B.S. Computer Engineering, North Carolina State University
Noble Lab
orserang (at) u.washington.edu

Research:
I have been working on the protein identification problem from shotgun proteomics. We originally phrased this problem as a linear optimization, where we modeled our belief that each protein is present in the sample with a weight, and then tried to minimize the sum of these weights such that the peptide-spectrum match scores, weighted by the proteins that contain them is greater than some threshold. By changing the threshold from 0 to larger values, we can force the optimization to include more proteins in the sample. The false positive rate and the true positive rates can be estimated by using a decoy database and purified cell lysate. Using our approach and this validation technique, we achieved similar results to ProteinProphet.

In solving the linear optimization problem, we developed a fast, novel method for solving linear programs. The method is (to our knowledge) the first randomized algorithm to perform well, and is competitive with well-known, efficient, and widespread methods.

For the protein identification problem, we have designed a well-motivated statistical approach that explicitly states all assumptions. The approach finds the true posterior belief that each protein is actually in the sample. We hope to implement this approach, and hypothesize that it will perform better than available methods.