Li Jiang M.S. Thesis Proposal

Computer Science Master's student Li Jiang will present her M.S. Thesis Proposal in 213 Sudikoff at 4:00 pm Tuesday, September 30, 2014.

September 30, 2014
4:00 pm - 5:30 pm
Location
213 Sudikoff Lab
Sponsored by
Computer Science Department
Audience
Public
More information
Shannon Stearne

Improving MHC: Peptide Binding Prediction from Sparse Data with a Multi-Allele Graphical Model and Active Learning

Receptor–ligand interactions play an important role in controlling many biological systems. One prominent example is the binding of peptides to the major histocompatibility complex (MHC) molecules controlling the onset of cellular immune responses. Due to allelic variations, determination of the binding specificity for each variant experimentally is infeasible. Further ,we haven’t seen any method efficiently leveraging similarity of MHC variants to solve this task. Knowledge about one allele should also be informative to any similar alleles. A network composed of multiple alleles and measurement data has the potential to transfer these information. To maximize the utility of each data measurement and allele similarity, we propose to use graphical model to joint represent all allele’s pocket profile. In this graph we represent each item of allele’s pocket profile as node,causal relationship as directed edge and significant pocket similarity as undirected edge. In this way, not only alleles with rich measurements can get good regularization power but also alleles with insufficient data can get more reasonable extrapolation. Another benefit of using graphical model is an enriched modeling of pocket profile. Currently what we can get from pocket profile methods is just a specific value to describe the contribution of a pocket to a peptide residue. Richer representations, probability distribution for example, will enable modeling of belief over current knowledge. With the ability to capture the uncertainty, active learning techniques based on Uncertainty Sampling and Submodular Optimization can be applied to guide experiment design.

Location
213 Sudikoff Lab
Sponsored by
Computer Science Department
Audience
Public
More information
Shannon Stearne