Computational analysis of epitope-specific T-cell repertoires.

Experimental advances in flow cytometry, single-cell analysis, and high-throughput sequencing are leading to ever-increasing volumes of data on epitope-specific alpha/beta T-cell responses. Paired sequence data, in which the link between the alpha and beta chain sequences is retained, permit a more complete analysis than can be achieved with unpaired data: they allow reconstruction of molecular models for complete receptors, for example, and they enable a more precise assessment of TCR sharing across individuals. We are developing new algorithms for analyzing paired, epitope-specific TCR sequence data with the goal of investigating the mapping between TCR sequence and peptide-MHC recognition, using large, epitope-specific sequence datasets for model viral infections in mice. Our tools facilitate a variety of common repertoire analyses including landscape visualization (by multidimensional scaling or hierarchical clustering), CDR3 pattern detection, convergent evolution analysis, and detection of gene segment enrichment and alpha/beta pairing preferences. A null model for the rearrangement process allows us to assign significance scores to sequence motifs and patterns of TCR sharing across individuals. 3D structural data on TCR:pMHC complexes inform our clustering algorithms, and we use molecular modeling to aid in interpreting recurring TCR motifs. We are also pursuing machine learning approaches for selecting features predictive of binding. We believe these analysis and visualization tools represent an important first step toward developing predictive computational models of epitope recognition. We also expect that they may be useful to other researchers faced with large TCR repertoire datasets.