Major histocompatibility complexes (MHC) guide immune response by presenting antigenfragments on a cell's surface and interacting with T-cell receptors (TCRs). In recent years, manyT-cell therapies have successfully engineered T-cells to target MHC-antigen complexesassociated with cancers and other diseases. However, most T-cell therapies require identifyinga TCR that interacts with an MHC-antigen complex of interest, a slow and expensive searchprocess. Our proposal aims to speed up this search process through a computational algorithmthat will predict whether a TCR will interact with an MHC allele of interest. Current screeningassays for low frequency TCRs have high false positive rates. Researchers can use our tool tocomputationally filter TCR candidates for interaction with a specific MHC allele before runningexpensive validation experiments. In this proposal, we will first validate our approach through aprototype algorithm that we will train on public TCR-MHC interaction data. We will then conductnew tetramer staining experiments that address two major challenges for developing analgorithm across multiple MHC alleles: the lack of interaction data for alleles other than A*02,and the limited antigen diversity in existing public data. These experiments will provideTCR-MHC data across 800 antigens for four common MHC alleles: A*01:01, A*02:01, A*11:01,and B*07:02. Finally, we will construct and validate computational algorithms for each MHCallele and evaluate the importance of various TCR components (e.g., alpha or beta chan,CDR3) in predicting TCR-MHC interaction. Our work will result in the first computational tool tohelp T-cell therapy developers filter TCR candidates based on MHC specificity. Beyond celltherapies, this tool will also help researchers track T-cells in diseases where MHC alleles play amajor role.
Public Health Relevance Statement: Narrative
We are developing a machine learning algorithm to identify T-cell receptors that interact with
MHC alleles of interest. T-cell therapies show enormous promise for late stage cancer patients
as well as millions of patients with resistant autoimmune conditions, but the T-cell identification
process for developing a new therapy often takes years and costs millions of dollars. Our tool
can significantly speed up this process by filtering out T-cell receptor candidates that do not
interact with an MHC allele of interest.
Project Terms:
