Regency EF
Finding Optimal Cutoff Value Based on Inflated Mixture Distributions and Its Application to T Cell Repertoire Sequencing Data (304094)
*Jason Baik, San Francisco State University
Tao He, San Francisco State University
Hai Yang, University of California, San Francisco
Li Zhang, University of California, San Francisco
Keywords: Local false discovery rate; T cell receptor; Mixture of distributions; Cut-off value
T cells represent a crucial component of the adaptive immune system and are thought to mediate the anti-tumoral immunity. Antigen-specific recognition is realized via T cell receptor (TCR), which is the product of somatic V(D)J gene recombination, plus some random addition/subtraction of nucleotides at recombination junctions. Next generation sequencing of TCR is used as a platform to profile TCR repertoire. The distribution of TCR data based on single-cell RNA sequencing typically follows some unknown distribution with inflation in the low abundance which might due to technical noise. We aim to find a cutoff value to filter the low abundance data and keep the enriched data, while controlling type I error via false discovery rate. We considered different inflated mixtures of distributions (inflated discrete mixtures, and inflated continuous mixtures) to model the abundance of clones in TCR repertoire in order to account for the excess ones in abundance and compared the cut-offs. We proposed a data-dependent procedure to select the best candidate mixture distribution and searched for the optimal cutoff. Both simulated and real data sets are used to illustrate the method.