We have a biological condition that we believe decreases splicing fidelity in c elegans. In our preliminary RNA seq data using 2x50 bp, 22 million reads per sample and 5 replicates, we found only 2 examples of intron retention, but we verified one of them using PCR.
We believe we can push the biological insult further to decreases splicing fidelity, and we want to do another RNA seq to find the targets that show different alternative splicing and mediate our phenotype, but I’m not an expert in this and would greatly appreciate opinions.
I read that for detection of these alternative splicing events, its best to use longer, paired-end reads, (2x125bp), and that you want at least 40 million reads per replicate. I was thinking to do 6 replicates per cell on the 2500 HiSeq in high-output mode with 2x125bp v4 chemistry, to give us ~37 million reads per sample.
Then I read about the pacbio sequel system, which gives you fewer (~0.25M) longer reads (10kb).
I have two questions really: (1) does anyone have an opinion about using PacBio versus HiSeq? and (2) if HiSeq, what specific reads versus replicates trade-off is best for detection of alternative splicing?
From your above description, I have understood, You want to do RNA sequence for studying SPLICING regions.
HiSeq 2500 can give maximum of 150*2 bidirectional sequencing, where as PacBio can give a stretch of 6kb to 12kb.
Though RNA Seq warrants cDNA conversion which can be maximum of 800bp based on the mode of cDNA conversion.
I would suggest Pacbio is better than Hiseq, if consider for the range of RNA seq you expected to do.
So you can decide, if your target region has more than 400bp or below to it.
If its below 400bp, then Hiseq 2500 is more than enough.
Further, library preparation plays major role, while performing RNA Seq, since, during preparation itself, we may miss the region or not covered required x coverage.
All the best.
Murugan N
For alternative splicing analysis it is advisable for paired end 150bp sequencing with 80 -100 million reads. This is a very high requirement and often you achieve around ~50 million reads. For rare transcripts this might not be sufficient.
On the other hand PacBio, though provides less read depth, has the advantage of long transcripts. A study in Nature biotechnology reported PacBio circular consensus sequencing libraries with poly-A+5′G cap enrichment method, could capture full lengths of expressed transcripts with the majority (90%) showing complete transcript sequences in the 1–2 kb range or even longer. I have attached the paper for your reference.
I personally have experience with Hiseq2500 but never had done comparative studies in PacBio system. But it essentially does put forward that if your target transcripts are few kbs long and your are interested in pre-mRNA species that you want to capture in length, then PacBio is more promising.
But if your target region is small (ideal is 60 million reads, since alternative splicing is not detectable with less reads. Hopefully this helps!!
Hi, an interesting and important question, and I would be interested what you decided in the end.
For (1) – like Sushree said, it depends on what kind of alterations you expect. If you are interested in "local" events, i.e. whether a single intron is retained, or a cassette exon excluded/included, etc., then Illumina is preferred, since you need depth and accuracy for this. If however you would like to see how more distant events correlate (e.g.: whether retention of the first intron is correlated with inclusion of cassette exon number 7, or so), then you need PacBio (or Nanopore).
I would like to add another thought regarding the read length: whereas I agree with the depth that she proposed (50-60 million reads per sample), I would suggest only 1x75 or 1x150. The reason is that you need, for local events, only sufficient alignements on both sides, e.g. whether a read spans an intron (proper splicing), or overlaps the intron-exon junction (intron retention). For sufficient alignements, 20-25nt per side are enough. If you use paired-end sequencing, you need statistical models that, depending on the insert size of the library, tell you whether for example an exon is included. E.g. if read 1 maps to exon 5 and read 2 maps to exon 7, then, based on the insert size and the lengthes of exons 5, 6 and 7, you can assign a probability to whether exon 6 is included or not. In my experience, the value of paired-end sequencing is somehow less than what everybody is saying…
Especially if you go for paired-end sequencing, make sure that you shorten the fragmentation time. Standard Illumina TruSeq fragmentation of 8 minutes gives fragments around 120-210nt, if you sequence 2x125bp then large parts of your data will be redundant since the two reads will overlap.
(2) – this depends on the variability of your samples. Tools like APsli (https://bioconductor.org/packages/release/bioc/html/ASpli.html) assign p-values to splicing events. You could use the data you have, and see how the p-values change when you do not use all replicates (i.e. compare p-values when using all five replicates, p-values when looking at all combinations of four replicates etc.). From this curves of p-values, you can estimate the sufficient amount of replicates.
- Badges
- Science method