1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
|
% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/lfcShrink.R
\name{lfcShrink}
\alias{lfcShrink}
\title{Shrink log2 fold changes}
\usage{
lfcShrink(
dds,
coef,
contrast,
res,
type = c("apeglm", "ashr", "normal"),
lfcThreshold = 0,
svalue = FALSE,
returnList = FALSE,
format = c("DataFrame", "GRanges", "GRangesList"),
apeAdapt = TRUE,
apeMethod = "nbinomCR",
parallel = FALSE,
BPPARAM = bpparam(),
quiet = FALSE,
...
)
}
\arguments{
\item{dds}{a DESeqDataSet object, after running \code{\link{DESeq}}}
\item{coef}{the name or number of the coefficient (LFC) to shrink,
consult \code{resultsNames(dds)} after running \code{DESeq(dds)}.
note: only \code{coef} or \code{contrast} can be specified, not both.
\code{apeglm} requires use of \code{coef}.
For \code{normal}, one of \code{coef} or \code{contrast} must be provided.}
\item{contrast}{see argument description in \code{\link{results}}.
only \code{coef} or \code{contrast} can be specified, not both.}
\item{res}{a DESeqResults object. Results table produced by the
default pipeline, i.e. \code{DESeq} followed by \code{results}.
If not provided, it will be generated internally using \code{coef} or \code{contrast}.
For \code{ashr}, if \code{res} is provided, then \code{coef} and \code{contrast} are ignored.}
\item{type}{\code{"apeglm"} is the adaptive Student's t prior shrinkage estimator from the 'apeglm' package;
\code{"ashr"} is the adaptive shrinkage estimator from the 'ashr' package,
using a fitted mixture of normals prior
- see the Stephens (2016) reference below for citation;
\code{"normal"} is the 2014 DESeq2 shrinkage estimator using a Normal prior;}
\item{lfcThreshold}{a non-negative value which specifies a log2 fold change
threshold (as in \code{\link{results}}). This can be used in conjunction with
\code{normal} and \code{apeglm}, where it will produce new p-values or
s-values testing whether the LFC is greater in absolute value than the threshold.
The s-values returned in combination with \code{apeglm} provide the probability
of FSOS events, "false sign or small", among the tests with equal or smaller s-value
than a given gene's s-value, where "small" is specified by \code{lfcThreshold}.}
\item{svalue}{logical, should p-values and adjusted p-values be replaced
with s-values when using \code{apeglm} or \code{ashr}. s-values provide the probability
of false signs among the tests with equal or smaller s-value than a given given's s-value.
See Stephens (2016) reference on s-values.}
\item{returnList}{logical, should \code{lfcShrink} return a list, where
the first element is the results table, and the second element is the
output of \code{apeglm} or \code{ashr}}
\item{format}{same as defined in \code{\link{results}},
either \code{"DataFrame"}, \code{"GRanges"}, or \code{"GRangesList"}}
\item{apeAdapt}{logical, should \code{apeglm} use the MLE estimates of
LFC to adapt the prior, or use default or specified \code{prior.control}}
\item{apeMethod}{what \code{method} to run \code{apeglm}, which can
differ in terms of speed}
\item{parallel}{if FALSE, no parallelization. if TRUE, parallel
execution using \code{BiocParallel}, see same argument of \code{\link{DESeq}}
parallelization only used with \code{normal} or \code{apeglm}}
\item{BPPARAM}{see same argument of \code{\link{DESeq}}}
\item{quiet}{whether to print messages}
\item{...}{arguments passed to \code{apeglm} and \code{ashr}}
}
\value{
a DESeqResults object with the \code{log2FoldChange} and \code{lfcSE}
columns replaced with shrunken LFC and SE.
For consistency with \code{results} (and similar to the output of \code{bayesglm})
the column name \code{lfcSE} is used here, although what is returned is a posterior SD.
For \code{normal} and for \code{apeglm} the estimate is the posterior mode,
for \code{ashr} it is the posterior mean.
\code{priorInfo(res)} contains information about the shrinkage procedure,
relevant to the various methods specified by \code{type}.
}
\description{
Adds shrunken log2 fold changes (LFC) and SE to a
results table from \code{DESeq} run without LFC shrinkage.
For consistency with \code{results}, the column name \code{lfcSE}
is used here although what is returned is a posterior SD.
Three shrinkage estimators for LFC are available via \code{type}
(see the vignette for more details on the estimators).
The apeglm publication demonstrates that 'apeglm' and 'ashr' outperform
the original 'normal' shrinkage estimator.
}
\details{
See vignette for a comparison of shrinkage estimators on an example dataset.
For all shrinkage methods, details on the prior is included in
\code{priorInfo(res)}, including the \code{fitted_g} mixture for ashr.
\strong{For type="apeglm":}
Specifying \code{apeglm} passes along DESeq2 MLE log2
fold changes and standard errors to the \code{apeglm} function
in the apeglm package, and re-estimates posterior LFCs for
the coefficient specified by \code{coef}.
\strong{For type="ashr":}
Specifying \code{ashr} passes along DESeq2 MLE log2
fold changes and standard errors to the \code{ash} function
in the ashr package,
with arguments \code{mixcompdist="normal"} and \code{method="shrink"}.
\strong{For type="normal":}
For design as a formula, shrinkage cannot be applied
to coefficients in a model with interaction terms.
For user-supplied model matrices, shrinkage is only
supported via \code{coef}. \code{coef} will make use
of standard model matrices, while \code{contrast}
will make use of expanded model matrices, and for the
latter, a results object should be provided to
\code{res}. With numeric- or list-style contrasts,
it is possible to use \code{lfcShrink}, but likely easier to use
\code{DESeq} with \code{betaPrior=TRUE} followed by \code{results},
because the numeric or list should reference the coefficients
from the expanded model matrix. These coefficients will be printed
to console if 'contrast' is used.
}
\examples{
set.seed(1)
dds <- makeExampleDESeqDataSet(n=500,betaSD=1)
dds <- DESeq(dds)
res <- results(dds)
# these are the coefficients from the model
# we can specify them using 'coef' by name or number below
resultsNames(dds)
res.ape <- lfcShrink(dds=dds, coef=2, type="apeglm")
res.ash <- lfcShrink(dds=dds, coef=2, type="ashr")
res.norm <- lfcShrink(dds=dds, coef=2, type="normal")
}
\references{
Publications for the following shrinkage estimators:
\code{type="apeglm"}:
Zhu, A., Ibrahim, J.G., Love, M.I. (2018) Heavy-tailed prior distributions for
sequence count data: removing the noise and preserving large differences.
Bioinformatics. \url{https://doi.org/10.1093/bioinformatics/bty895}
\code{type="ashr"}:
Stephens, M. (2016) False discovery rates: a new deal.
Biostatistics, 18:2. \url{https://doi.org/10.1093/biostatistics/kxw041}
\code{type="normal"}:
Love, M.I., Huber, W., Anders, S. (2014) Moderated estimation of fold change and
dispersion for RNA-seq data with DESeq2. Genome Biology, 15:550.
\url{https://doi.org/10.1186/s13059-014-0550-8}
Related work, the \code{bayesglm} function in the \code{arm} package:
Gelman, A., Jakulin, A., Pittau, M.G. and Su, Y.-S. (2009).
A Weakly Informative Default Prior Distribution For Logistic And Other Regression Models.
The Annals of Applied Statistics 2:4.
\url{http://www.stat.columbia.edu/~gelman/research/published/ priors11.pdf}
}
|
