What is the meaning of zscore in the results of differentially expressed analysis?

[AIBio]

Hi, I want to cite the package SCDE in my paper. But I do not know how to discribe the rationale of this packages. Most importantly, it is difficlult for me to understand the parameters, including lb, mle, ub, ce, Z and cZ. I have attempted to search the relevant information in your paper publised on journal-Nature Method(Bayesian approach to single-cell differential expression analysis). But, my effort ended in failure. Could you explain the meaning behind these parameters or provide me some learning materials to read. Thank you~~~

[pkharchenko]

lb - lower bound of the 95% confidence interval for the log2 fold change mle - maximum likelihood estimate of the fold change (log2 scale) ub - upper bound of the 95% confidence interval Z - raw Z score (signed P value, converted to Z score) cZ - Z score, corrected for the multiple hypothesis testing Best, -peter.

…

On Aug 19, 2019, at 23:08, Hanwen_Yu ***@***.***> wrote: Hi, I want to cite the package SCDE in my paper. But I do not know how to discribe the rationale of this packages. Most importantly, it is difficlult for me to understand the parameters, including lb, mle, ub, ce, Z and cZ. I have attempted to search the relevant information in your paper publised on journal-Nature Method(Bayesian approach to single-cell differential expression analysis). But, my effort ended in failure. Could you explain the meaning behind these parameters or provide me some learning materials to read. Thank you~~~ — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[AIBio]

@pkharchenko
Hi, peter, thank you for your reply. But I still cannot understand these parameters. How can I differentiate the up- or down-regulated genes from these parameters. In DESeq2, the analytical results will show the log2foldofchange and the qvalue that can be used to choose the targeted gene list, such as differentially up-regulated genes by setting up threshold log2foldofchange > 1 & qvalue < 0.05. In results of scde, I do not know how to choose the genes I want. Could you teach me how to use these parameters to achieve this goal? Thank you~~~
Best wish,
Hanwen

[pkharchenko]

You can use them much in the same way. The mle provides the log2 fold change estimate. cZ encodes the same information as Q value, but on a different scale (i.e. pnorm(cZ,lower.tail=F) will give you the corrected p value of gene being up-regulated, pnorm(cZ, lower.tail=T) of the gene being down-regulated). Z scores are also easy to work with directly … a Z score of 3 corresponds to the P value around 1e-3, and Z score of 2 corresponds to P value around 025. The sign tells you wither the gene is up (positive) or down (negative). Also, ‘ce’ column contains the conservative log2 fold change estimate (i.e. the 95% CI bound that’s closest to 0, or 0 if the CI intersects 0), which makes it easy to select genes that are differentially expressed beyond a certain fold change threshold with 95% confidence.

…

On Aug 20, 2019, at 03:53, Hanwen_Yu ***@***.***> wrote: @pkharchenko Hi, peter, thank you for your reply. But I still cannot understand these parameters. How can I differentiate the up- or down-regulated genes from these parameters. In DESeq2, the analytical results will show the log2foldofchange and the qvalue that can be used to choose the targeted gene list, such as differentially up-regulated genes by setting up threshold log2foldofchange > 1 & qvalue < 0.05. In results of scde, I do not know how to choose the genes I want. Could you teach me how to use these parameters to achieve this goal? Thank you~~~ Best wish, Hanwen — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[AIBio]

You can use them much in the same way. The mle provides the log2 fold change estimate. cZ encodes the same information as Q value, but on a different scale (i.e. pnorm(cZ,lower.tail=F) will give you the corrected p value of gene being up-regulated, pnorm(cZ, lower.tail=T) of the gene being down-regulated). Z scores are also easy to work with directly … a Z score of 3 corresponds to the P value around 1e-3, and Z score of 2 corresponds to P value around 025. The sign tells you wither the gene is up (positive) or down (negative). Also, ‘ce’ column contains the conservative log2 fold change estimate (i.e. the 95% CI bound that’s closest to 0, or 0 if the CI intersects 0), which makes it easy to select genes that are differentially expressed beyond a certain fold change threshold with 95% confidence.
…
On Aug 20, 2019, at 03:53, Hanwen_Yu @.***> wrote: @pkharchenko Hi, peter, thank you for your reply. But I still cannot understand these parameters. How can I differentiate the up- or down-regulated genes from these parameters. In DESeq2, the analytical results will show the log2foldofchange and the qvalue that can be used to choose the targeted gene list, such as differentially up-regulated genes by setting up threshold log2foldofchange > 1 & qvalue < 0.05. In results of scde, I do not know how to choose the genes I want. Could you teach me how to use these parameters to achieve this goal? Thank you~~~ Best wish, Hanwen — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

@pkharchenko
Your answer is really helpful! It is easy to understand that the cZ value can be tranformed into Q-value by function pnorm in R language. But a little problem still confuse me. The Z score can tell me whether the expression level of certain gene is up-regulated or down. As I understrand, the mle that provides the log2 fold change estimate is the real parameter to judge the expression level change. For example, the genes with positive value represents that their expression levels is up-regulated. Do I misunderstand your explanation?
best wish,
Hanwen