[NGS - analysis of gene expression data] Machine Learning + RNAseq data

[Chuckytah]

Hello,

i'm a portuguese student girl of master degree in bioinformatics

Do you advise me some good websites/articles for theese topics:
--‐ Next generation sequencing for gene expression measurement (RNA‐seq)
--‐ data analysis challenges in RNA--‐Seq data
--‐ applications of machine learning in classification of RNAseq data
--‐ identification and critical analysis available tools


I wanted to focus more in "applications of machine learning in
classification of RNAseq data" do you know some pratical case that i could
present to my classmates?

Thanks a lot,

Inês Martins
University of Minho, Portugal

[steven]

Hi Inês,

Here is a recent review about RNA-seq challenges in bioinformatics:
http://www.nature.com/nmeth/journal/...d=NMETH-201106

However, no real "machine learning" method comes to my mind about RNA-seq.. I do not recall Cufflinks or Scripture are trained on some learning dataset. Are they?

[altodor]

Storey paper

There are excellent references in this paper:

http://www.nature.com/nbt/journal/v29/n4/full/nbt.1831.html?WT.ec_id=NBT-201104

John Storey

H. Craig Mak
Nature Biotechnology 29, 331–333 (2011) doi:10.1038/nbt.1831
Published online 08 April 2011
John Storey provides his take on the importance of new statistical methods for high-throughput sequencing.

Good luck!

[Simon Anders]

By "machine learning", do you mean clustering and classification? To my k nowledge, not much has been done there yet so far. This is because, typically, you want to have large data sets with tens, better hundreds, of samples, to bring ML techniques to fruitful use and then, microarrays are still preferred as they are still cheaper. So, have a look at what people have done for microarray studies. I'm sure people will pretty soon start thinking about how to adapt these methods to RNA-Seq, so stay tuned.

[Chuckytah]

thank you all guys

[Chuckytah]

In the topic: Next generation sequencing for gene expression measurement

i'm doing RA (relative abundances)... I have the number of EST in one gene and i divide it by the total number of EST in the sample... is this right?

[sphil]

It's not enough to normalise with the number of est in sample. let me referr you to RPKM (PMID: 18516045) or packages like deg-seq, rna-seqc.

[steven]

In the topic: Next generation sequencing for gene expression measurement

i'm doing RA (relative abundances)... I have the number of EST in one gene and i divide it by the total number of EST in the sample... is this right?

I am confused about the terminology. What is your data/experiment exactly? ESTs are not reliable and should not be used to infer expression values. If you are doing some Digital Gene Expression or SAGE experiment then it is possible. With full RNA-seq too but methods differ: with SAGE tags you do not expect the length of the transcript to matter as the goal is to get all the reads of a given transcript to originate from a unique position. For Whole Transcriptome Sequencing RNA-seq reads should originate from all over the transcripts. Therefore the number of reads is expected to correlate with the size of the transcript, so a normalization may be required (see RPKM above). Note that if you are comparing gene expressions between different conditions this size-normalization is not required.