[bansal_raman]

Great, Thanks Xi

[Xi Wang]

Thanks Xi,
Do you think that the quality controls like box chart can help to determine if further normalization is required or not?

Yes. But it should be based on the assumption that the majority of genes are expressed at the same level in the two samples studied.

[bansal_raman]

Thanks Xi,
Do you think that the quality controls like box chart can help to determine if further normalization is required or not?

[Xi Wang]

Hi Xi Wang
I found your replies in this thread very informative. I have a further inquiry. For differential gene expression analysis (between two samples), do we need to transform and normalize the RPKM values (obtained for genes of individual samples)? I guess RPKM values are already obtained by normalization

Hi,

It has been realized that the further normalization is needed if the total numbers of expressed RNA molecules are different in two samples. See the reference:
Robinson MD, Oshlack A.A scaling normalization method for differential expression analysis of RNA-seq data.Genome Biol. 2010;11(3):R25. Epub 2010 Mar.

http://genomebiology.com/2010/11/3/R25

[bansal_raman]

Hi Xi Wang
I found your replies in this thread very informative. I have a further inquiry. For differential gene expression analysis (between two samples), do we need to transform and normalize the RPKM values (obtained for genes of individual samples)? I guess RPKM values are already obtained by normalization

[Xi Wang]

i think is a UCSC refFlat format. how can i created that archive

You can download the archive from UCSC genome browser. Eg, for hg18 refSeq genes, follow the link below:

http://hgdownload.cse.ucsc.edu/goldenPath/hg18/database/refFlat.txt.gz

[Xi Wang]

Hi Xi Wang, sorry to answer leate back. im already at home.
thanx for your helpe. i intersting to use the RPKM but i can`t run my data, i think because my reference genome need to be in other extension is that right?.


thanx for your help.
have a good day

I am also at home now. :-)

To answer your question, I need to know which tool you used to calculate the RPKM values.

Thanks!

[Gamaliel]

i think is a UCSC refFlat format. how can i created that archive

[Gamaliel]

Thanks Gamaliel.

So your difficulty is to estimate the gene expression levels from ununiformly distributed reads, right? First, RNA-seq experiments following the random priming protocol are supposed to generate uniformly distributed reads from transcripts' 5'ends to 3'ends. However, I still saw some ununiformity on our data, but RPKM still worked. I think if the reads in your data is not extremely ununiformly distributed, RPKM still works. Second, if you still concern the ununiformity, you can give up the concept of ORFs, but use sliding windows to scan the genome, to see the read enrichment. Maybe the sliding window size is a key parameter.

Hi Xi Wang, sorry to answer leate back. im already at home.
thanx for your helpe. i intersting to use the RPKM but i can`t run my data, i think because my reference genome need to be in other extension is that right?.


thanx for your help.
have a good day

[Xi Wang]

Thanks Gamaliel.

So your difficulty is to estimate the gene expression levels from ununiformly distributed reads, right? First, RNA-seq experiments following the random priming protocol are supposed to generate uniformly distributed reads from transcripts' 5'ends to 3'ends. However, I still saw some ununiformity on our data, but RPKM still worked. I think if the reads in your data is not extremely ununiformly distributed, RPKM still works. Second, if you still concern the ununiformity, you can give up the concept of ORFs, but use sliding windows to scan the genome, to see the read enrichment. Maybe the sliding window size is a key parameter.

[Gamaliel]

If the reads along the transcripts follow the uniform distribution assumption, you can the RPKM concept to calculate the proportion of trancribed copies for different genes. RPKM means reads per kilo-base of transcript per million reads. Taking your example, suppose the total reads of the experiment is 10 million, the RPKM for gene "a" is 50/(1.2k)/(10M)=4.17, which RPKN for gene "b" 48/(0.8k)/(10M)=6. So, gene "b" has a higher expression level than gene "a".

Hi Xi Wang i understand you. but the reads along the transcripts are not uniform it mean the covert is difrent a long the ORF (transcrpts). thanx for your help this is my email [email protected] if you need some help too.

[Xi Wang]

If the reads along the transcripts follow the uniform distribution assumption, you can the RPKM concept to calculate the proportion of trancribed copies for different genes. RPKM means reads per kilo-base of transcript per million reads. Taking your example, suppose the total reads of the experiment is 10 million, the RPKM for gene "a" is 50/(1.2k)/(10M)=4.17, which RPKN for gene "b" 48/(0.8k)/(10M)=6. So, gene "b" has a higher expression level than gene "a".

[Gamaliel]

Can I understand your question in this way: as there are some background noise of the RNA-seq data, some regions may have reads, but the DNA don't transcribe. I think there are lots of researcher are doing this project or related, and I did not find any publication yet. A naive way could be filtering out the regions with RPKM value less than a given cutoff (say, 1 RPKM). And then you can compare the remaining between conditions.

hi xi wang thanx again, yes i mean it the background of my RNAseq data, my other cuestion is: in the genome there are genes of diferents legths for example:

if i get 2 genes one gene size 1.2kb gen "a" and the other size 800pb gen "b"

gen "a" __________________________ 1.2kb
_________
_______ ________

imagine that i have for this gen "a" 50 reads


gen "b" ________________ 800pb
_____ ______ ____
_____ _____ ____

and for the gen "b" i have 48 reads

my question is if the size of the orf import, because the gen "a" got more reads than gen "b" just for the size and not for is transcribed more than the gen "b"

well thanx for all your suport and help, have a good day

[Xi Wang]

Hi Xi wang, the problem is that i have to normalize one condition firts because i have estaocastic reads and i need to know in just one condition what genes are trasncribed (determinat transcription) and what are trasncribed product of the estocastic level (inespesific trascription). after compare to other condition.


thanx for your time and help

Can I understand your question in this way: as there are some background noise of the RNA-seq data, some regions may have reads, but the DNA don't transcribe. I think there are lots of researcher are doing this project or related, and I did not find any publication yet. A naive way could be filtering out the regions with RPKM value less than a given cutoff (say, 1 RPKM). And then you can compare the remaining between conditions.

[Gamaliel]

I can guess what you meant to type. RPKM is recommended if you want to do the comparison between samples. A tool named "cufflink" can be used to carry out your job.

Hi Xi wang, the problem is that i have to normalize one condition firts because i have estaocastic reads and i need to know in just one condition what genes are trasncribed (determinat transcription) and what are trasncribed product of the estocastic level (inespesific trascription). after compare to other condition.


thanx for your time and help