HapTree-X: An Integrative Bayesian Framework for Haplotype Reconstruction from Transcriptome and Genome Sequencing Data

Identifying phase information is biomedically important due to the association of complex haplotype effects, such as compound heterozygosity, with disease. As recent next-generation sequencing (NGS) technologies provide more read sequences, the use of diverse sequencing datasets for haplotype phasing is now possible, allowing haplotype reconstruction of a single sequenced individual using NGS data. Previous haplotype reconstruction studies have ignored differential allele-specific expression in whole transcriptome sequencing (RNA-seq) data; however, intuition suggests that the asymmetry in this data (i.e. maternal and paternal haplotypes of a gene are differentially expressed) can be exploited to improve phasing power. In this paper, we describe a novel integrative maximum-likelihood estimation framework, HapTree-X, for efficient, scalable haplotype assembly of an individual genome using transcriptomic and genomic NGS read datasets, which makes use of differential allele-specific expression. Our advance includes the first method for haplotype assembly that uses differential expression, newly allowing the use of reads that cover only one SNP. We evaluate the performance of HapTree-X on real sequencing read data, both transcriptomic and genomic, from NA12878 (1000 Genomes Project and Gencode) and demonstrate that HapTree-X increases the number of SNPs that can be phased and sizes of phased-haplotype blocks, without compromising accuracy. We prove theoretical bounds on the precise improvement of accuracy as a function of coverage which can be achieved from differential expression-based methods alone. Thus, the advantage of our integrative approach substantially grows as the amount of RNA-seq data increases.