Hi - getting there, need some help

[abilslan]

Hi,

I'm Alan from Glasgow. I'm a bench scientist but with a fair amount of bioinformatics (network analysis, discrete models, neural nets). Programming hobbyist is the best way to describe it so far - I'm happy in C and Matlab but it's not my core job and I need to give a decent pitch to bring more computational stuff into the lab. I'm interested in RNAseq but not directly working on it at the moment (except with public stuff at home). I'm trying to convince the boss that it might be worth moving beyond arrays. But, I'll need to show some nice data to make the point. I just got myself a Linux box for the house (finally!) especially for this testing ground. I mainly have the tuxedo protocol running, I think. Of course, there are one or two questions...

Thanks.

[Heisman]

You didn't really ask a question. What exactly does your lab work on that would justify using RNA-seq? In terms of showing nice data I imagine you can use things already published and show how sequencing can accomplish things that arrays cannot.

[SNPsaurus]

Hi Alan,

Getting your hands on data and working through it yourself is a great idea. My personal bioinformatics 'trick' is to always take a quick look at the raw data. I've seen people struggle for a long time with a pipeline only to realize the reads are mostly PCR duplicates or the quality was poor.

It can be hard to justify changing platforms. But if RNA-Seq has features that are better than arrays, better to make the switch sooner rather than later, I think. I loved building an array spotter and the data that resulted, but once I saw the RNA-Seq data I dropped arrays immediately!

[abilslan]

Thanks Heisman. The lab works on pathways regulating cell immortality and senescence in cancer. We're especially interested in affecting these with small molecules and usually use arrays as input for pathway analysis to understand the transcriptional effects induced by a hit compound. This works well but I wonder if RNAseq could be more revealing. We're also involved in some plasma biomarker trials. One suggestion is to use miRNA arrays in one of the trials, but again I'm wondering if we shouldn't be trying RNAseq in at least a few patient samples to get a feel for whether the output could be useful in this context.

I was going to post my question in the RNAseq forum, but might as well quickly ask here. As a first run-through I'm trying to compare paired-end reads from 2 prostate cancer from sra. I'm up to the cuff-diff stage of the tuxedo protocol and all seems to have gone well (I have all the logs if needed).

The cuff-diff instructions say I need to include at least 2 sam files with the merged.gtf file as input. I'm just not sure exactly which sam files these should be. I tried the accepted_hits output from tophat to no avail. Should it be the raw files that are provided as input to tophat or something else?

[abilslan]

Thank you SNPsaurus. Wise words. I have a few horror stories of my own of the odd student here and there not quite grasping the need for QC in array analyses. Who would have guessed that slides covered in dried-on gunk will not produce robust results? Our lab has its own Agilent scanner and we do get through a lot of arrays, so I don't think we would make the switch completely overnight. But, I do think if I can get some decent analysis out of some public data it should highlight the potential power of RNAseq for some of our more critical experiments.

[abilslan]

Sorry, please ignore my question. Cuffdiff works perfectly well. It was late when I was trying this part of it and I just had a typo. However, the threads on this site have been very helpful getting to this point, and more questions may arise. Thanks.