mRNA Quantification via eXpress

eXpress is a tool that allows mRNA quantification using a set of transcripts as a reference (this is opposed to popular RNA-Seq tools like TopHat, which align reads to a genome and have to model gaps caused by exon junctions).

Using transcripts rather than genomic chromosomes as a reference sequences is actually how I imagined RNA-Seq analysis would be conducted, before I learned about standard practices.  In fact, samtools provides an 'idxstats' function that can be used to calculate normalized RPKM expression values.  So, I was curious if the extra modeling done by eXpress is really any better than this simple sort of RPKM calculation: having a more complicated model can potentially improve accuracy, but more complicated models can also leave extra room for things to go wrong, can lead to over-fitting, etc.  For example, I have used eXpress on some de novo assembly data, and I actually found that normal de novo programs seemed to provide better results than those specifically designed for RNA-Seq data (however, to be clear, I think the results of this blog post emphasize that the problem was with the assembly and not the mRNA quantification, as I would have expected).

The short answer is "Yes" - I think it is better to use eXpress over idxstats for calculating RPKM/FPKM values.

To illustrate this, first take a look at the correlations between the eXpress FPKM values and the RPKM values calculated using idxstats:

The correlation isn't horrible, but you can see a non-trivial amount of genes whose expression levels have consistently lower in eXpress than idxstats.  However, this by itself doesn't really prove one options is better than the other option.  Because I feel comfortable with the gene-level mRNA quantification levels from cufflinks (and the RSEM-like algorithm implemented in Partek; for example, see Figure 5 in this paper or click here to see a direct correlation between these two results), I decided to see how the results compared when using different tools for a transcript-based reference (eXpress, idxstats) versus a genomic/chromosome-based reference (cufflinks, Partek).

Again, you see these outliers if you compare the idxstats results to cufflinks (or to Partek - click here for those results):

However, you don't see these outliers when comparing eXpress to cufflinks (or to Partek - again, click here for those results):

So, eXpress clearly provides more robust results than the simpler idxstats comparison.  You can also see this in box plot below, showing the correlation coefficients for all the mRNA quantification strategies that I tested.

Of course, systematic differences between mRNA quantification methods should (at least partially) be corrected when identifying differentially expressed genes between two groups (because the differences affect both groups).  However, there are some certain circumstances when the mRNA quantification levels may want be used in isolation, such as for ranking the most highly expressed genes in a sample (as was the case for the de novo assembly data that I worked with).  In this situations, I would definitely recommend a tool like eXpress over trying to calculate RPKM values from tools like idxstats.

FYI, here are some details on the methodology for this comparison:

• MiSeq samples from GSE37703 were used for these comparisons.

• Correlations were calculated using log2(FPKM/RPKM + 0.1) expression values.

• eXpress and idxstats were run on Bowtie2 alignments of the same set of RefSeq transcripts (downloaded from the UCSC Genome Browser, with duplicated gene IDs removed).  The Partek EM algorithm used a set of RefSeq sequences used by the vendor and cufflinks used the genes.gtf file downloaded from iGenomes on the TopHat website.  Only commonly represented gene symbols were used for calculating correlations.  Only genes declared "solvable" by eXpress were considered for calculating correlations.  As an example, click here to view a venn diagram of overlapping gene symbols for SRR493372.

P.S. It looks like you may have to be signed into Google Docs to view the image previews properly.  However, you can always download the files to view them locally.

RNA-Seq Differential Expression Benchmarks

I recently published a paper whose primary purpose was to serve as a reference for the protocol that I use for RNA-Seq analysis (see main paper and supplemental figures).

The aspect of the paper that I think is most interesting to the genomics community is a comparison of statistical tools for defining differentially expressed genes, which had the greatest influence on the resulting gene lists (at least among the comparisons that I make in the paper).  So, I will review those relevant figures in this blog post.

The plots below show the robustness of the gene lists produced by a given algorithm.  In other words, the higher the "common" line on the graph, the more robust the gene lists (i.e. the higher the proportion of genes commonly called by multiple algorithms).  Most readers will probably not be as interested in the x-axis (rounding factor for RPKM values), and it only changes the gene lists for Partek and sRAP.

Analysis of Patient Cohort (Tumor versus Normal).  1-factor is just tumor versus normal, while 2-factor also includes patient ID (pairing tumor and normal samples).  cuffdiff results not shown because no genes were defined with FDR < 0.05.  sRAP not shown because gene list was very small (see Figure S3 from the paper)

Analysis of Cell Line Comparison (Mutant versus WT)

To be fair, I will certainly admit robustness is not the same as accuracy.   Uniquely identified genes may be true positives that represent a lower false negative rate.  However, this did correspond to some circumstantial evidence I've seen with other datasets where cuffdiff and edgeR have given some weird results.  The results from this paper don't actually contain the clearest examples of this, but you can take a look at the GAGE4 stats to see an example where I would at least argue that edgeR provides inflated statistical significance.

Overall, I think Partek works the best (which is what I use for COH customers), but I was also pleased with DESeq (and sRAP, but I am obviously biased).  In fact, these comparisons support earlier observations that DESeq is conservative in defining lists of differential expressed genes (Robles et al. 2012).

However, my main goal is not to simply tell you what is the single best solution.  In fact, the cell line comparison above also had paired microarray data, and I would say the concordance between the two technologies was roughly similar for most algorithms:

RNA-Seq versus Microarray Gene lists.  "Microarray DEG" = proportion of differentially expressed genes in microarray data also present in RNA-Seq gene list.  "RNA-Seq DEG" = proportion of differentially expressed genes in RNA-Seq data also present in microarray gene list.

The similarity in microarray concordance kind of reminds me of Figure 2a from Rapport et al. 2013, which compares RNA-Seq gene lists to ~1000 qPCR validated genes.  However, I think properly determining accuracy can be difficult.  For example, look at the differences between the qPCR results in Figure 2a and the ERCC spike-ins in Figure S5 for that same paper.

Instead, these are the main take-home points I would like to emphasize:

1) Simple methods comparing RPKM values (in this case, rounded and log₂ transformed) for defining differentially expressed genes can work at least as well as more complicated methods that are unique for RNA-Seq analysis (at least for gene-level comparisons).  For example, one claim against count-based methods in general (including edgeR, DESeq, etc.) is that there can be confounding factors, such as changes in splicing patterns.  Although I agree this is a theoretical problem that probably does occur to some extent, it doesn't seem to be a major factor influencing concordance with microarray data, qPCR validation, etc.

2) There is probably not a solution that works best in all situations. In this paper, you can see the results look very different with the patient versus cell line datasets.  For practical reasons, a lot of benchmarks will probably use cell line datasets.  However, it is not safe to assume performance for large patient cohorts will be comparable to cell line data (or patient data with little or no biological replicates).