AltHapAlignR: improved accuracy of RNA-seq analyses through the use of alternative haplotypes

Abstract

MotivationReliance on mapping to a single reference haplotype currently limits accurate estimation of allele or haplotype-specific expression using RNA-sequencing, notably in highly polymorphic regions such as the major histocompatibility complex.ResultsWe present AltHapAlignR, a method incorporating alternate reference haplotypes to generate gene- and haplotype-level estimates of transcript abundance for any genomic region where such information is available. We validate using simulated and experimental data to quantify input allelic ratios for major histocompatibility complex haplotypes, demonstrating significantly improved correlation with ground truth estimates of gene counts compared to standard single reference mapping. We apply AltHapAlignR to RNA-seq data from 462 individuals, showing how significant underestimation of expression of the majority of classical human leukocyte antigen genes using conventional mapping can be corrected using AltHapAlignR to allow more accurate quantification of gene expression for individual alleles and haplotypes.Availability and implementationSource code freely available at https://github.com/jknightlab/AltHapAlignR.Supplementary information Supplementary data are available at Bioinformatics online.

Expression estimation and eQTL mapping for HLA genes with a personalized pipeline

“…This may be explained by different techniques (RNA-seq vs. qPCR), the samples used, or the fact that imputation of expression was used for heterozygous genotypes in previous studies [13,18]. Even in a comparison with [33], which also reanalyzed the GEUVADIS dataset with an HLA-personalized approach, we observe only a moderate concordance in the ordering of lineages by expression level. This is likely due to several differences between our approach and theirs, including the strategy for alignment ([33] use the 8 MHC region haplotypes to guide quantification, while we directly use the entire known HLA diversity for this task), and the treatment of reads mapping to multiple loci or alleles ( [33] discard these reads, while we use a maximum likelihood approach to measure their contribution).…”

“…For both first and second steps of our pipeline a read can align to more than one allele or gene (due to their sequence level similarity). Instead of discarding such reads (as in [33]) or evenly splitting them among the compatible references (as in [30]), we use maximum likelihood estimates of expression obtained by an expectation-maximization (EM) algorithm, which is implemented within Salmon and kallisto. This procedure probabilistically assigns reads to each reference in the index, in a way that accounts for reads that align to more than one gene or allele [35,36].…”

“…For example, seq2HLA is a tool developed by [30] to provide in-silico HLA types and expression estimates, and later applied to demonstrate that different tumors types are associated with different HLA expression levels [31], and also to provide a large catalog of HLA expression in 56 human tissues and cell types [32]. AltHapAlignR [33] is another recently described algorithm which infers the MHC references which are the closest to the individual's MHC haplotypes, and maps reads to them. The authors reanalyzed the GEU-VADIS dataset [24], and provided comparisons with conventional read mapping, showing an improvement in accuracy with the HLA-tailored pipeline.…”

In-depth genetic analysis of 6p21.3 reveals insights into associations between HLA types and complex traits and disease

“…There were 146 expressed genes in the MHC interval (TPM >2 in at least 10 samples), of which 24 were HLA genes, including six HLA class I genes , 3 HLA class I pseudogenes, 11 HLA class II genes, MICA, MICB, TAP1 and TAP2. For each individual, we estimated expression levels of the 24 HLA genes by aligning reads to cDNA sequences specific for the HLA types called by HLA-VBSeq(Nariai et al, 2015) to avoid alignment biases (Aguiar et al, 2019;Gensterblum-Miller et al, 2018;Lee et al, 2018;Panousis et al, 2014) Supplementary File 4). The HLA class I genes tended to be expressed at high levels, consistent with their ubiquitous expression in all cell types, and the HLA class II genes tended to be expressed at lower levels, as expected due to their primary role in immune cells (Matzaraki et al, 2017).…”

Scavenger: A pipeline for recovery of unaligned reads utilising similarity with aligned reads

“…However, this approach is limited as there is no ground truth in real data to help with optimisation, and increasing the number of reads aligned will also result in an increase in the number of false positive reads. Another strategy for solving the false-negative non-alignment problem is by incorporating variation information during alignment, in the form of utilising alternate loci sequences within the reference genome [15] or integration of a single nucleotide polymorphism database to the reference [13], to help minimise the effect of divergence of the personal genome compared to the reference genome. This approach is also limited as it requires existing variation information, which may not be available in non-model organisms.…”