Tools and best practices for data processing in allelic expression analysis.
Castel SE, Levy-Moonshine A, Mohammadi P, Banks E, Lappalainen T. · 2015
Read Tools and best practices for data processing in allelic expression analysis. by Castel SE, Levy-Moonshine A, Mohammadi P, Banks E, Lappalainen T. (2015) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4574606.txt.
Introduction
Stephane E. Castel and four co-authors set standards for measuring which gene copy a cell uses, in Genome Biology in 2015. Each person carries two copies of most genes, and allelic expression analysis checks whether one copy works harder than the other. The approach feeds studies of gene regulation and RNA decay, but its read counts are riddled with technical traps.
Low-quality or double-counted reads, genotyping slips, mapping bias toward the reference copy, sample handling effects, and uneven sequencing depth can all fake an imbalance. The paper tests each error source and gives correction guidelines. With their quality controls, detection of true allelic differences improves. Tools for producing clean allelic data at scale are included.
Catalog record
| CREATOR | Castel SE, Levy-Moonshine A, Mohammadi P, Banks E, Lappalainen T. |
|---|---|
| YEAR | 2015 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Tools and best practices for data processing in allelic expression analysis By Stephane E Castel, Ami Levy-Moonshine, Pejman Mohammadi, Eric Banks, Tuuli Lappalainen Published in: Genome Biology DOI: 10.1186/s13059-015-0762-6 ABSTRACT Abstract Allelic expression analysis has become important for integrating genome and transcriptome data to characterize various biological phenomena such as cis-regulatory variation and nonsense-mediated decay. We analyze the properties of allelic expression read count data and technical sources of error, such as low-quality or double-counted RNA-seq reads, genotyping errors, allelic mapping bias, and technical covariates due to sample preparation and sequencing, and variation in total read depth. We provide guidelines for correcting such errors, show that our quality control measures improve the detection of relevant allelic expression, and introduce tools for the high-throughput production of allelic expression data from RNA-sequencing data. Electronic supplementary material The online version of this article (doi:10.1186/s13059-015-0762-6) contains supplementary material, which is available to authorized users. BACKGROUND Integrating genome and transcriptome data has become a widespread approach for understanding genome function. Allelic expression (AE; also called allele-specific expression or allelic imbalance) analysis is becoming an increasingly important tool for this, as it quantifies expression variation between the two haplotypes of a diploid individual distinguished by heterozygous sites (Fig. 1a). This approach can be used to capture many biological phenomena (Fig. 1b): effects of genetic regulatory variants in cis [1–8], nonsense-mediated decay triggered by variants causing a premature stop codon [9–12], and imprinting [13, 14]. Standard RNA-sequencing (RNA-seq) data capture AE only when higher expression of one parental allele is shared between individual cells (Additional file 1), as opposed to random monoallelic expression of single cells that typically cancels out when a pool of polyclonal cells is analyzed [15, 16]. [Figure: Allelic expression and its sources. a Schematic illustration of AE. b Biological sources of AE, with the x-axis denoting the approximate sharing of AE across tissues of an individual, and the y-axis having the estimated sharing of AE signal in one tissue across different individuals [5, 8, 12, 13, 15]. SNP single-nucleotide polymorphism] In this paper, we describe a new tool in the Genome Analyzer Toolkit (GATK) software package for efficient retrieval of raw allelic count data from RNA-seq data, and analyze the properties of AE data and the sources of errors and technical variation, with suggested guidelines for accounting for them. While most types of errors may be rare, they are easily enriched among sites with allelic imbalance, and can sometimes mimic the biological signal of interest, thus warranting careful analysis. Our focus is on methods for obtaining accurate data of AE rather than building a graphical user interface (GUI) pipeline [17] or downstream statistical analysis of its biological sources [9, 13, 18–20]. The example data in most of our analysis are the open-access RNA-seq data set of the lymphoblastoid cell lines (LCLs) of 1000 Genomes individuals from the Geuvadis project [5]. RESULTS AND DISCUSSION Unit of AE data The biological signal of interest in AE analysis is the relative expression of a given transcript from the two parental chromosomes. Typical AE data seek to capture this by counts of RNA-seq reads carrying reference and alternative alleles over heterozygous sites in an individual [heterozygous single-nucleotide polymorphisms (het-SNPs)], and this is the focus of our analysis unless mentioned otherwise. The Geuvadis samples with a median depth of 55 million mapped reads have about 5000 het-SNPs covered by ≥30 RNA-seq reads, distributed across about 3000 genes and 4000 exons (Fig. 2; Additional file 2). The exact number varies due to differences in sequencing depth, its distribution across genes, and individual DNA heterozygosity. About one half of these genes contain multiple het-SNPs per individual, which could be aggregated to better detect AE across the gene (Fig. 2d). However, alternative splicing can introduce true biological variation in AE in different exons, and incorrect phasing needs to be accounted for in downstream analysis [13]. Additionally, summing up data from multiple SNPs is not appropriate if the same RNA-seq reads overlap both sites. In the Geuvadis data, 9 % of the reads used in AE analysis in fact overlap more than one het-SNP (Figure S2d in Additional file 2), but this will become more frequent as read lengths increase [21]. In the future, better tools are needed to partition RNA-seq reads to either of the two haplotypes according to all het-SNPs that they overlap [22]. In fact, this could help to phase exonic sites separated by long introns. [Figure: Genomic coverage of AE data in Geuvadis CEU samples. a Cumulative distribution of RNA-seq read coverage per het-SNP (each line represents one sample). b, c The number of het-SNPs (b) and protein-coding genes (c) per sample as a function of coverage cutoff. d The number of protein-coding genes with AE data versus the number of het-SNPs they contain. Each line is the median for all samples at a specific coverage level] AE analysis of small insertions or deletions (indels) has proven to be technically very challenging and it is rarely attempted even though frameshift indels are an important class of protein-truncating variant. Alignment errors over indel loci are pervasive due to multiple mismatches of reads carrying alternative alleles, and lower genotyping quality adds further error [12]. In Rivas et al. [12] we describe the first approach for large-scale analysis of AE over indels, but further methods development is warranted for better sensitivity and computational scalability. In addition to classical AE analysis to detect differences in total expression level of two haplotypes, it is also possible to analyze allelic differences in transcript structure or splicing [allelic splicing (AS)] [5, 21]. These methods compare the exon distribution of reads and their mates carrying different alleles of a heterozygous site, and work increasingly well for longer total fragment lengths. In these analyses, the data structure is somewhat more complex than reference/non-reference read counts in AE, depending on the specific algorithm. While this paper focuses on classical AE analysis of SNPs, most of the quality analysis steps apply to indel AE and AS analyses as well. Tools to retrieve allele counts Allele counts are the starting point for all AE analyses, and many previous tools can retrieve these counts. However, they also perform other analyses that require additional input data and increase the runtime. Here we present simple tools that can be used to retrieve only allele counts, using the minimum required inputs in standard formats. We present two solutions: 1) a highly efficient Python tool that processes results from SAMtools mpileup, the framework used by the majority of existing AE analysis pipelines; and 2) an easy to use tool in the widely used GATK v.3.4 [23, 24] called ASEReadCounter, which does not require any additional setup, and includes a variety of easily customizable read processing options as well as professional maintenance and documentation, similar to other GATK tools. Both operate on aligned RNA-seq reads and count the reference and alternative allele reads that passed filters for mapping and base quality at each bi-allelic heterozygous variant. The GATK tool offers several additional options for processing RNA-seq reads: by default each read fragment is counted only once if the base calls are consistent at the site of interest, and duplicate reads are filtered (see below). Other options