RIP-seq analysis: Where do we stand right now?

[sjuzhet]

I'm a wet-lab scientist trying to dive into the world of computational biology, so I'm still fairly inexperienced when it comes to analyzing high throughput sequencing data. I feel I've got a solid grasp on ChIP-seq and RNA-seq and the options available to me.

However, RIP-seq experiments still elude me. From what I hear, ChIP-seq peak callers such as macs2 are unsuitable for RIP-seq for reasons beyond my understanding. I find tools such as Piranha and RIPseeker, and I wonder why they never really gained much popularity - is there something inherently flawed in their algorithm that prevents one from trusting their results?

I've tried aligning with tophat, bowtie, and BWA, I've tried de novo transcriptome assembly with Trinity. I've counted reads with HTseq and Sailfish, I've called peaks with Piranha and macs2. But, since I have no real expertise on this matter, I have no idea which results to trust. Is there anything I can do before I turn to the wet lab?

What is the best way to approach RIP-seq data?

[Brian Bushnell]

So... in circumstances such as these it is useful to know what kind of data you have, from what platforms, and from what kinds of organisms? And what do you hope to accomplish?

Bear in mind that if you focus on, say, Human genetics with 2x101bp Illumina reads, for a single individual with no known diseases, and ask a generic question... the answers you get may be focused on single-cell bacterial genomics, which are mostly unrelated. Or more likely, you won't get any answers unless you clarify the question.

[sjuzhet]

Fair enough - but you nearly nailed it with your guess. I've got 2x101 Hi-Seq reads from mammalian tissue culture. I'm interested in identifying protein-bound RNAs potentially responsible for regulating said protein's activity in modulating gene expression. To that end, I've got ChIP-seq and RNA-seq reads as well - but I feel comfortable enough with those results.

[Brian Bushnell]

I suggest that you perform a few basic steps before analyzing your data:

1) Adapter-trimming. Whether or not this is useful depends on the insert-size distribution of your library, and I avoid saying "it can't hurt" because many operations can cause problems, but usually adapter-trimming is useful and I have yet to encounter a situation where it (conservative adapter-trimming) is detrimental.
2) Trim the last 1bp of your reads. Illumina's base-callers use information from bases to the left and right to determine a given base. For some platforms, such as NextSeq, the last base plummets in accuracy (since it has no subsequent base), regardless of the claimed Q-score; and for all Illumina platforms I have tested, the last base - unless it was already trimmed (which for 101bp reads, it wasn't) - has terrible accuracy compared to non-last-bases of similar quality values. BBDuk has an "ftm" (force-trim-modulo) flag designed for Illumina reads that have an extra base; I set it at "ftm=5" because Illumina reads typically have a length that is a multiple of 5, or a multiple of 5 plus 1; if they have the extra 1, it needs to be discarded.
3) Other steps, like contaminant/spike-in removal or quality-trimming, may be useful but depend on specifically what you do with the data.

I have not studied RIP-seq so I don't know about its idiosyncrasies, but I recommend BBMap if you want RNA alignments to the whole genome with the highest possible sensitivity and specificity. I'm the developer, by the way. But I would not recommend it if it did not outperform everything else I have tested for RNA-seq genome-alignment accuracy.

When aligning to a transcriptome, it's a bit easier... you don't necessarily need to do mapping. Which are you doing?

[dariober]

I find tools such as Piranha and RIPseeker, and I wonder why they never really gained much popularity

RIP-Seq is not a very popular technique in the first place, it seems to me. So it's not surprising that the tools to analyze it are not popular either and there are no benchmarks out there... (P.S. I've never worked on RIP-Seq)