Friday, December 30, 2016

NOIseq RNAseq differential analysis


http://bioconductor.org/packages/release/bioc/html/NOISeq.html

M: fold-change differences
D: absolute expression differences

(M,D) pair for each gene is evaluated based on a null distribution estimated from technical or biological replicates or simulations in 2011GR.

In NIOSEQBIO, theta=(M+D)/2 seems to be the statistic used for null distribution based on my understanding of its manual.




Probability = 0.8 was the cutoff for differentially expressed genes in 2011GR.
Probability = 0.95 (FDR) is recommended for biologically replicated samples.

In its Tarazona2011GR, noiseq-real and noiseq-sim were used. These two versions have now evolved to noiseq and noiseqbio.  


NOISEQBIO is optimized for biological replicates.

When using noiseq and noiseqbio, normalization and filtering can be done through parameters, 'norm'.
Regarding the low-count filtering, it is not necessary to filter in NOISeq  method. In contrast, it is recommended to do it in NOISeqBIO , which by default fliters out low-count features with CPM method (filter=1 ).



# noiseq(input, k = 0.5, norm = c("rpkm","uqua","tmm","n"),  replicates = c("technical","biological","no"), factor=NULL, conditions=NULL, pnr = 0.2, nss = 5, v = 0.02, lc = 0)

mynoiseq = noiseq(mydata, k = 0.5, norm = "rpkm", factor="Tissue", pnr = 0.2, 
                  nss = 5, v = 0.02, lc = 1, replicates = "technical")
head(mynoiseq@results[[1]])

> myfactors
           Tissue TissueRun
R1L1Kidney Kidney  Kidney_1
R1L2Liver   Liver   Liver_1
R1L3Kidney Kidney  Kidney_1
R1L4Liver   Liver   Liver_1
R1L6Liver   Liver   Liver_1
R1L7Kidney Kidney  Kidney_1
R1L8Liver   Liver   Liver_1
R2L2Kidney Kidney  Kidney_2
R2L3Liver   Liver   Liver_2
R2L6Kidney Kidney  Kidney_2



mynoiseqbio = noiseqbio(mydata, k = 0.5, norm = "rpkm", factor="Tissue", lc = 1, r = 20, adj = 1.5, plot = FALSE,   a0per = 0.9, random.seed = 12345, filter = 2)
# "r=20" seems to indicate 20 bootstraps when biological replicate number <5. 





Authors stated that noiseq output prob are not equivalent to p-values? 




Q: what are "up" and "down" deg referenced to?









Output format


mynoiseq.deg1 = degenes(mynoiseq, q = 0.8, M = "up")










References:
 [1] S. Tarazona, F. Garca-Alcalde, J. Dopazo, A. Ferrer, and A. Conesa. Dierential expression in RNA-seq: A matter of depth. Genome Research , 21: 2213 - 2223, 2011.

[2] S. Tarazona, P. Furio-Tar, D. Turra, A. Di Pietro, M.J. Nueda, A. Ferrer, and A. Conesa. Data quality aware analysis of dierential expression in RNA-seq with NOISeq R/Bioc package. Nucleic Acids Research ,

43(21):e140, 2015.

[8] B. Efron, R. Tibshirani, J.D. Storey, V. Tusher. Empirical Bayes Analysis of a Microarray Experiment. Journal of the American Statistical Association , 2001.



No comments:

Post a Comment