Student: Vincent Xue

Lab: Keating

Title:  Modeling and Designing Bcl-2 Family Protein Interactions Using High-Throughput Interaction Data 

Protein-protein interactions (PPIs) play a major role in cellular function, mediating signal processing, and regulating enzymatic activity. Given the importance of protein interactions in the cell, understanding how proteins interact is essential for prediction of new binding partners, and for engineering new function. Analysis of protein mutants is an effective means to understanding the determinants of protein interactions. Traditionally, biophysical study of protein interactions has been limited by the amount of data available, but recent advances in high-throughput sequencing have enabled rapid assessment of thousands of mutant variants. In the Keating lab, we have developed an experimental protocol that is able to rank peptides for their binding affinity to a designated receptor. This technique, called SORTCERY, takes advantage of cell sorting and deep-sequencing technologies to provide more binding data at higher resolution than has previously been achievable. SORTCERY has been applied to study Bcl-2-family protein interactions, which play a major role in regulating apoptosis. With these emerging techniques, and new wealth of data, there exists a need for computational methods to process and analyze the high-throughput datasets.  In this thesis I show how experimental data from SORTCERY experiments can be processed, modeled, and used to design novel peptides with select specificity characteristics. I describe the computational processing pipeline that I developed to curate the data and regression models that I constructed from the data to relate protein sequence to binding. I applied models trained on experimental data sets to study the peptide-binding specificity landscape of the Bcl-xl, Mcl-1, and Bfl-1 anti-apoptotic proteins, and I designed novel peptide sequences that selectively bind only one receptor or a pre-specified combination of receptors. The synergistic application of data driven models combined with high-throughput binding assays can bring us closer towards a complete understanding of protein interactions for rational design.