Mapping Accuracy of Short Reads from Massively Parallel Sequencing and the Implications for Quantitative Expression Profiling

Background

Massively parallel sequencing offers an enormous potential for expression profiling, in particular for interspecific comparisons. Currently, different platforms for massively parallel sequencing are available, which differ in read length and sequencing costs. The 454-technology offers the highest read length. The other sequencing technologies are more cost effective, on the expense of shorter reads. Reliable expression profiling by massively parallel sequencing depends crucially on the accuracy to which the reads could be mapped to the corresponding genes.

Methodology/Principal Findings

We performed an in silico analysis to evaluate whether incorrect mapping of the sequence reads results in a biased expression pattern. A comparison of six available mapping software tools indicated a considerable heterogeneity in mapping speed and accuracy. Independently of the software used to map the reads, we found that for compact genomes both short (35 bp, 50 bp) and long sequence reads (100 bp) result in an almost unbiased expression pattern. In contrast, for species with a larger genome containing more gene families and repetitive DNA, shorter reads (35–50 bp) produced a considerable bias in gene expression. In humans, about 10% of the genes had fewer than 50% of the sequence reads correctly mapped. Sequence polymorphism up to 9% had almost no effect on the mapping accuracy of 100 bp reads. For 35 bp reads up to 3% sequence divergence did not affect the mapping accuracy strongly. The effect of indels on the mapping efficiency strongly depends on the mapping software.

Conclusions/Significance

In complex genomes, expression profiling by massively parallel sequencing could introduce a considerable bias due to incorrectly mapped sequence reads if the read length is short. Nevertheless, this bias could be accounted for if the genomic sequence is known. Furthermore, sequence polymorphisms and indels also affect the mapping accuracy and may cause a biased gene expression measurement. The choice of the mapping software is highly critical and the reliability depends on the presence/absence of indels and the divergence between reads and the reference genome. Overall, we found SSAHA2 and CLC to produce the most reliable mapping results.     

Large Scale Comparison of Gene Expression Levels by Microarrays and RNAseq Using TCGA Data

RNAseq and microarray methods are frequently used to measure gene expression level. While similar in purpose, there are fundamental differences between the two technologies. Here, we present the largest comparative study between microarray and RNAseq methods to date using The Cancer Genome Atlas (TCGA) data. We found high correlations between expression data obtained from the Affymetrix one-channel microarray and RNAseq (Spearman correlations coefficients of ∼0.8). We also observed that the low abundance genes had poorer correlations between microarray and RNAseq data than high abundance genes. As expected, due to measurement and normalization differences, Agilent two-channel microarray and RNAseq data were poorly correlated (Spearman correlations coefficients of only ∼0.2). By examining the differentially expressed genes between tumor and normal samples we observed reasonable concordance in directionality between Agilent two-channel microarray and RNAseq data, although a small group of genes were found to have expression changes reported in opposite directions using these two technologies. Overall, RNAseq produces comparable results to microarray technologies in term of expression profiling. The RNAseq normalization methods RPKM and RSEM produce similar results on the gene level and reasonably concordant results on the exon level. Longer exons tended to have better concordance between the two normalization methods than shorter exons.     

ShapeShifter: a novel approach for identifying and quantifying stable lariat intronic species in RNAseq data

Background: Most intronic lariats are rapidly turned over after splicing. However, new research suggests that some introns may have additional post-splicing functions. Current bioinformatics methods used to identify lariats require a sequencing read that traverses the lariat branchpoint. This method provides precise branchpoint sequence and position information, but is limited in its ability to quantify abundance of stabilized lariat species in a given RNAseq sample. Bioinformatic tools are needed to better address these emerging biological questions. Methods: We used an unsupervised machine learning approach on sequencing reads from publicly available ENCODE data to learn to identify and quantify lariats based on RNAseq read coverage shape. Results: We developed ShapeShifter, a novel approach for identifying and quantifying stable lariat species in RNAseq datasets. We learned a characteristic “lariat” curve from ENCODE RNAseq data and were able to estimate abundances for introns based on read coverage. Using this method we discovered new stable introns in these samples that were not represented using the older, branchpoint-traversing read method. Conclusions: ShapeShifter provides a robust approach towards detecting and quantifying stable lariat species.     

Selection of primers for optimal taxonomic classification of environmental 16S rRNA gene sequences

Microbial community profiling using 16S rRNA gene sequences requires accurate taxonomy assignments. ‘Universal'' primers target conserved sequences and amplify sequences from many taxa, but they provide variable coverage of different environments, and regions of the rRNA gene differ in taxonomic informativeness—especially when high-throughput short-read sequencing technologies (for example, 454 and Illumina) are used. We introduce a new evaluation procedure that provides an improved measure of expected taxonomic precision when classifying environmental sequence reads from a given primer. Applying this measure to thousands of combinations of primers and read lengths, simulating single-ended and paired-end sequencing, reveals that these choices greatly affect taxonomic informativeness. The most informative sequence region may differ by environment, partly due to variable coverage of different environments in reference databases. Using our Rtax method of classifying paired-end reads, we found that paired-end sequencing provides substantial benefit in some environments including human gut, but not in others. Optimal primer choice for short reads totaling 96 nt provides 82–100% of the confident genus classifications available from longer reads.     

Variations in 16S rRNA-based microbiome profiling between pyrosequencing runs and between pyrosequencing facilities

Pyrosequencing of 16S rRNA gene amplicons on the 454 FLX Titanium platform has been widely used to analyze microbiomes in various environments. However, different results may stem from variations among sequencing runs or among sequencing facilities. This study aimed to evaluate these variations between different pyrosequencing runs by sequencing 16S rRNA gene amplicon libraries generated from three sets of rumen samples twice each on the 454 FLX Titanium system at two independent sequencing facilities. Similar relative abundances were found for predominant taxa represented by large numbers of sequence reads but not for minor taxa represented by small numbers of sequence reads. The two sequencing facilities revealed different bacterial profiles with respect to both predominant taxa and minor taxa, including the most predominant genus Prevotella, the family Lachnospiraceae, and the phylum Proteobacteria. Differences in primers used to generate amplicon libraries may be a major source of variations in microbiome profiling. Because different primers and regions of 16S rRNA genes are often used by different researchers, significant variations likely exist among studies. Quantitative interpretation for relative abundance of taxa, especially minor taxa, from prevalence of sequence reads and comparisons of results from different studies should be done with caution.     

Uniform approximation is more appropriate for Wilcoxon Rank-Sum Test in gene set analysis

Gene set analysis is widely used to facilitate biological interpretations in the analyses of differential expression from high throughput profiling data. Wilcoxon Rank-Sum (WRS) test is one of the commonly used methods in gene set enrichment analysis. It compares the ranks of genes in a gene set against those of genes outside the gene set. This method is easy to implement and it eliminates the dichotomization of genes into significant and non-significant in a competitive hypothesis testing. Due to the large number of genes being examined, it is impractical to calculate the exact null distribution for the WRS test. Therefore, the normal distribution is commonly used as an approximation. However, as we demonstrate in this paper, the normal approximation is problematic when a gene set with relative small number of genes is tested against the large number of genes in the complementary set. In this situation, a uniform approximation is substantially more powerful, more accurate, and less intensive in computation. We demonstrate the advantage of the uniform approximations in Gene Ontology (GO) term analysis using simulations and real data sets.