| Title: | Sample size for RNAseq studies |
|---|---|
| Description: | RNA-seq, sample size |
| Authors: | Terry M Therneau [aut, cre], Hart Stephen [ctb] |
| Maintainer: | Terry M Therneau <[email protected]> |
| License: | LGPL (>=2) |
| Version: | 1.47.0 |
| Built: | 2024-12-01 05:24:52 UTC |
| Source: | https://github.com/bioc/RNASeqPower |
Sample size and power computation for RNA-seq studies
rnapower(depth, n, n2 = n, cv, cv2 = cv, effect, alpha, power)rnapower(depth, n, n2 = n, cv, cv2 = cv, effect, alpha, power)
depth |
average depth of coverage for the transcript or gene of interest. Common values are 5-20, any numeric value >0 is valid. |
n |
sample size in group 1 (or both) |
n2 |
sample size in group 2 |
cv |
biological coefficient of variation in group 1 (or both). |
cv2 |
biological coefficient of variation in group 2 |
effect |
target effect size |
alpha |
size of the test statistic (two sided), i.e., the false positive rate, a number between 0 and 1 |
power |
power of the test, i.e., the fraction of true positives that will be detected, a number between 0 and 1 |
The error in an RNA-seq study arises from two causes: the technical variability of the sequencing itself, and the within group biological variability of the specimens being compared. For human samples this latter will often be .4-1, for inbred animal lines values are commonly .1 or less. That is, we might see a 2-fold difference in mean expression between treated and untreated samples, and at the same time a 50% variation in expression between samples within the control or treatment group. This would correspond to an effect size of 2 and a CV of 0.5. As a general rule, sequencing depths of more than 5/CV^2 will lead to only minor gains in study efficiency and/or power, whereas addition of further samples is always efficatious.
Depth is a required argument; any one of the others may be left missing and the function will solve for it. An argument may be a vector, in which case a vector of values is returned. If multiple arguments are vectors a matrix or array of results is returned.
Common values for alpha and power would be .05 and .9.
The effect parameter specifies the biological effect that, if
it were true, the experimenter would want to be able to detect; values
of 1.5 to 2 are commonly used. The statements that A has twice the
expression of B and that B has half that of A are symmetric,
likewise values of .5 and 2 for effect will yeild the same
result. The estimated CV of expression within group may be the
most difficult parameter to choose; see the vignette for an in depth
discussion of this along with recommended values for diffrent types
of data.
By default the samples sizes in the two groups are assumed to be
equal (n2=n) if a second sample size is not given. Likewise the
coefficients of variation in the two groups
are assumed to be equal if cv2 is not specified.
a vector, matrix, or array of values. If the function was used
with all the arguments except n supplied then the result will
be the needed samples size; if it were run with all arguments except
power the result will contain power estimates, etc.
Terry Therneau and Steven Hart
Steven Hart, Terry Therneau, Yuji Zhang, Gregory Poland, and Jean-Pierre Kocher. Calculating sample size estimates for RNA sequencing data. J. Comp. Biology, 2013, in press.
# What would the power be for a study with 12 per group, to detect a # 2-fold change, given deep (50x) coverage? rnapower(50, cv=.6, effect=2, n=12, alpha=.05) # Compute a sample size for several combinations of parameters temp <- rnapower(10, cv=.5, effect=c(1.5, 2), alpha=c(.05, .01, .001), power=c(.8, .9)) round(temp,1) # Result is an array with dimensions of effect size, alpha, and power # which contains the sample size per group for each combination.# What would the power be for a study with 12 per group, to detect a # 2-fold change, given deep (50x) coverage? rnapower(50, cv=.6, effect=2, n=12, alpha=.05) # Compute a sample size for several combinations of parameters temp <- rnapower(10, cv=.5, effect=c(1.5, 2), alpha=c(.05, .01, .001), power=c(.8, .9)) round(temp,1) # Result is an array with dimensions of effect size, alpha, and power # which contains the sample size per group for each combination.