The Taub Faculty of Computer Science Events and Talks

Tandem mass spectrometry (MS/MS) has emerged as the tool of choice for high-throughput proteomics analysis. In a typical MS/MS experiment, a protein mixture sample is digested to peptides and is chromatographically separated. Homogenous sets of peptide molecules are then selected by the mass spectrometer and fragmented via collision induced dissociation. The outcome of this process is recorded as a mass spectrum, which is a list of fragment masses and their corresponding intensities. The algorithmic challenge that arises is how to identify peptide amino acid sequences from such spectra. To perform this task successfully, sequencing algorithms need rely on accurate models for peptide fragmentation. However, peptide fragmentation is a complex process that can involve several competing chemical pathways. Though a lot of theoretical research and experiments have been performed with the goal understanding the peptide "fragmentation rules", incorporating this knowledge into algorithms has proven to be a difficult task.

In this talk we will present a novel approach for modeling peptide fragmentation that uses a machine learning boosting algorithm called ranking (Fruend et al. 2003) to create detailed models that reflect many of these chemistry-based "fragmentation rules". Our experimental results demonstrate that incorporating these models into algorithms significantly improves the peptide identification rate attained both with de novo sequencing and database searches.

This talk will be self-contained, no background in proteomics or machine learning is required.

Host: Zohar Yakhini