2018-9-30

edgeR exactTest VS glmLRT VS QL

design <- model.matrix(~Subject+Treat)#设计一个实验设计矩阵
fit <- glmFit(y, design)#构建一个基因表达值得广义线性模型
lrt <- glmLRT(fit)#通过似然比检验得到最后的结果

2. GLM model
   quasi-likelihood (QL)
   The NB model can be extended with quasi-likelihood (QL) methods

F-test glmQLFit() and glmQLFTest()

likelihood ratio test

Dispersion

reference

1. https://en.wikipedia.org/wiki/Statistical_dispersion wiki 统计上的离差
2. https://support.bioconductor.org/p/75970/ edgeR 的离差
3. https://support.bioconductor.org/p/50447/

Common examples of measures of statistical dispersion are the variance, standard deviation, and interquartile range

edgeR fold changes

glmtreat

1. https://support.bioconductor.org/p/70361/#70368
2. https://support.bioconductor.org/p/65635/
3. https://support.bioconductor.org/p/79444/

intercept and edgeR

https://support.bioconductor.org/p/78693/
https://support.bioconductor.org/p/78693/

Design matrix not of full rank

https://support.bioconductor.org/p/80408/

Reference

1. https://support.bioconductor.org/p/86375/ QL method avoid type I error OR low counts and/or large dispersions,
2. https://blog.csdn.net/hzau_yang/article/details/78118257 edgeR 的使用
3. https://support.bioconductor.org/p/84291/ exact test or QLF test
4. https://blog.csdn.net/chenzeyu110/article/details/53265315 edgeR 的使用
5. https://www.jianshu.com/p/a9d5065f82a6 normalization

https://support.bioconductor.org/p/84338/
https://support.bioconductor.org/p/84291/#84292

RUVseq

1. using normalized counts or raw counts for ruvg

How to show both edgeR and deseq2 results in a single volcano plot; highliting overlaps
https://www.biostars.org/p/314123/