Treating paired end data as single end

[tirohia]

I have a colleague who is recommend that we treat paired end data as single end data so that we get two technical replicates - it's an exploratory experiment, there's one replicate of 4 different cultivars of apple in two conditions.

Treating the PE data as SE seems ... wrong to me, but I can't immediately bring any objections to mind. Is there a reasonable reason that this should or shouldn't be done?

Cheers
Ben.

[dpryan]

This is wrong because you would be deluding yourself into believing that you have (approximately) twice as much evidence supporting coverage of a gene/region/whatever-you're-looking-at than you actually do. Remember that the point of technical replicates is to increase the read count, which wouldn't actually be done by treating mates separately since they are not, in actuality, separate...the mapping position of one will restrict that of the other (or at least it should). Also, if you don't have a well characterized genome (at least the last apple genome I looked at was far from well characterized) then dropping the mate from alignments might also tank you alignment rate/reliability.

[tirohia]

Thanks for that.

The best I had been able to come up was that it might impact on how well the reads are aligned to the reference genome. Which is similar to you last point I believe.

Cheers
Ben

[Brian Bushnell]

Thanks for that.

The best I had been able to come up was that it might impact on how well the reads are aligned to the reference genome. Which is similar to you last point I believe.

Cheers
Ben

No, it's much more than that. It's more like treating the passengers on the left and right side of a cruise ship as technical replicates. Obviously, both had the exact same journey.

Using the first and last half of the file from a single library as technical replicates is absurd, but considering the left and right reads from a single run as independent is infinitely worse - it's a travesty, since they're physically linked.

[dpryan]

No, it's much more than that. It's more like treating the passengers on the left and right side of a cruise ship as technical replicates. Obviously, both had the exact same journey.

I'm totally stealing that analogy!

[Chipper]

Tell him that you then also will split each read in two to get four replicates...

[woodydon]

Guess your colleague is tell you to split the read pairs into single reads as replicates. I think it depends on your questions for technical replicates.

[Brian Bushnell]

I think it depends on your questions for technical replicates.

Not really. There are no useful questions that could be answered correctly by considering those as technical replicates. Also, read2 has lower quality than read1, so they will have 100% correlation bias plus different expression levels, which is perfect. Everything will be differentially expressed by the same amount.

[woodydon]

Not really. There are no useful questions that could be answered correctly by considering those as technical replicates.

I agree with you. I am just wondering what his/her colleague is thinking.

Woody

[dpryan]

Tell him that you then also will split each read in two to get four replicates...

Careful, someone will read that and think it's a good idea!

[tirohia]

Colleague is new to bioinformatics following the lead of a non-bioinformatician they trust. (I'm the lone systems biology PhD student in the midst of molecular biologists and plant pathologists).

The experiment is a very loose (everyone acknowledges that) sneak peek to try and give some insight into what a properly designed experiment might be looking at. Still. Didn't seem right. As others seem to be doing, I'm am now officially stealing the cruise ship analogy.

Ta muchly.

[westerman]

If you are doing a 'sneak peek' then you really don't need technical replicates. Your statistics will not be correct and your statisticians will be horrified but the stats shouldn't be that far off. Besides, tech reps are over-rated. :-)

http://seqanswers.com/forums/showthread.php?t=8873