Predicting V(D)J recombination using conditional random fields

Abstract

V(D)J gene segments undergo combinatorial recombination in the T-cells and B-cells to provide humans and other vertebrates with a large number of antibodies required for immunity. Each such recombination further undergoes mutations in their DNA sequences so that they can recognize diverse antigens. Predicting the combination of gene segments which formed a particular antibody is an essential task for studying disease propagation and analysis. We propose a model based on conditional random fields (CRFs) for predicting the boundary positions between V-D-J gene segments. We train the CRFs by generating synthetic gene recombinations using all of the alleles of the V, D and J gene segments. The alleles corresponding to a read can be determined by mapping the segmented reads to the DNA sequences of the gene segments using softwares like BLAST and usearch. We test our method on simulated dataset as well as real data of Stanford-S22 individual. © 2012 Springer-Verlag.