g**********y 发帖数: 423 | 1 I am doing regression on microarray data - X.
X is the gene expression level derived from rma normalization.
Suppose there are p independent variables (genes) and n samples (
observations)
and regression by Y = X'beta. (Y is response variables, or phenotype)
Can we state that the importance of each gene is determined by its
corresponding coeff beta?
should X be standardized to mean zero variance one for each gene across all
samples? Denote it as Z. Is it better to performan regression on Z,
Y=Z' | g**********y 发帖数: 423 | | g******7 发帖数: 19 | 3 standardization seems to be a solution but keep in mind that although each
IV now is in the scale of "standard deviation", the actual SD of each IV
could be different, if so, the rank that based upon standardized solution
may not make much sense. just a thought.
all
【在 g**********y 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I am doing regression on microarray data - X. : X is the gene expression level derived from rma normalization. : Suppose there are p independent variables (genes) and n samples ( : observations) : and regression by Y = X'beta. (Y is response variables, or phenotype) : Can we state that the importance of each gene is determined by its : corresponding coeff beta? : should X be standardized to mean zero variance one for each gene across all : samples? Denote it as Z. Is it better to performan regression on Z, : Y=Z'
| D******n 发帖数: 2836 | 4 this has been kinda discussed before, and no definite conlusion is found.
see 17623
all
【在 g**********y 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I am doing regression on microarray data - X. : X is the gene expression level derived from rma normalization. : Suppose there are p independent variables (genes) and n samples ( : observations) : and regression by Y = X'beta. (Y is response variables, or phenotype) : Can we state that the importance of each gene is determined by its : corresponding coeff beta? : should X be standardized to mean zero variance one for each gene across all : samples? Denote it as Z. Is it better to performan regression on Z, : Y=Z'
|
|