The biobroom
package contains methods for converting standard objects in Bioconductor into a “tidy format”. It serves as a complement to the popular broom package, and follows the same division (tidy
/augment
/glance
) of tidying methods.
“Tidy data” is a data analysis paradigm that focuses on keeping data formatted as a single observation per row of a data table. For further information, please see Hadley Wickham’s seminal paper on the subject. “Tidy” is not a normative statement about the quality of an object’s structure. Rather, it is a technical specification about the choice of rows and columns. A tidied data frame is not “better” than an S4 object; it simply allows analysis with a different set of tools.
Tidying data makes it easy to recombine, reshape and visualize bioinformatics analyses. Objects that can be tidied include
ExpressionSet
object,GRanges
and GRangesList
objects,RangedSummarizedExperiment
object,MSnSet
object,limma
, edgeR
, and DESeq2
,qvalue
object for multiple hypothesis testing.We are currently working on adding more methods to existing Bioconductor objects. If any bugs are found please contact the authors or visit our github page. Otherwise, any questions can be answered on the Bioconductor support site.
The biobroom
package can be installed by typing in a R terminal:
source("http://bioconductor.org/biocLite.R")
biocLite("biobroom")
To find out more about the provided objects:
library(biobroom)
?edgeR_tidiers
?DESeq2_tidiers
?limma_tidiers
?ExpressionSet_tidiers
?MSnSet_tidiers
?qvalue_tidiers
The qvalue package is a popular package to estimate q-values and local false discovery rates. To get started, we can load the hedenfalk
dataset included in the qvalue
package:
library(qvalue)
data(hedenfalk)
qobj <- qvalue(hedenfalk$p)
names(qobj)
## [1] "call" "pi0" "qvalues" "pvalues" "lfdr"
## [6] "pi0.lambda" "lambda" "pi0.smooth"
qobj
is a qvalue
object, generated from the p-values contained in the hedenfalk
dataset. If we wanted to use a package such as dplyr
or ggplot
, we would need to convert the results into a data frame. The biobroom
package makes this conversion easy by using the tidy
, augment
and glance
functions:
tidy
returns one row for each choice of the tuning parameter lambda.augment
returns one row for each provided p-value, including the computed q-value and local false discovery rate.glance
returns a single row containing the estimated pi0
.Applying these functions to qobj
:
library(biobroom)
# use tidy/augment/glance to restructure the qvalue object
head(tidy(qobj))
## # A tibble: 6 × 3
## lambda smoothed pi0
## <dbl> <lgl> <dbl>
## 1 0.05 FALSE 0.8517350
## 2 0.10 FALSE 0.8068700
## 3 0.15 FALSE 0.7823344
## 4 0.20 FALSE 0.7563091
## 5 0.25 FALSE 0.7310200
## 6 0.30 FALSE 0.7138351
head(augment(qobj))
## # A tibble: 6 × 3
## p.value q.value lfdr
## <dbl> <dbl> <dbl>
## 1 0.01212618 0.08819163 0.1680943
## 2 0.07502524 0.20936729 0.4128630
## 3 0.99492114 0.66799864 1.0000000
## 4 0.04178549 0.16163643 0.3088248
## 5 0.84581388 0.63247386 1.0000000
## 6 0.25192429 0.37172329 0.6924674
head(glance(qobj))
## # A tibble: 1 × 2
## pi0 lambda
## <dbl> <dbl>
## 1 0.669926 0.95
The original data, or in this example the gene names, can be inputted into augment
using the data
argument:
# create sample names
df <- data.frame(gene = 1:length(hedenfalk$p))
head(augment(qobj, data = df))
## # A tibble: 6 × 4
## gene p.value q.value lfdr
## <int> <dbl> <dbl> <dbl>
## 1 1 0.01212618 0.08819163 0.1680943
## 2 2 0.07502524 0.20936729 0.4128630
## 3 3 0.99492114 0.66799864 1.0000000
## 4 4 0.04178549 0.16163643 0.3088248
## 5 5 0.84581388 0.63247386 1.0000000
## 6 6 0.25192429 0.37172329 0.6924674
The tidied data can be used to easily create plots:
library(ggplot2)
# use augmented data to compare p-values to q-values
ggplot(augment(qobj), aes(p.value, q.value)) + geom_point() +
ggtitle("Simulated P-values versus Computed Q-values") + theme_bw()
Additionally, we can extract out important information such as significant genes under a false discovery rate threshold:
library(dplyr)
# Find significant genes under 0.05 threshold
sig.genes <- augment(qobj) %>% filter(q.value < 0.05)
head(sig.genes)
## # A tibble: 6 × 3
## p.value q.value lfdr
## <dbl> <dbl> <dbl>
## 1 0.0007129338 0.02315186 0.03652658
## 2 0.0017507886 0.03645186 0.05965671
## 3 0.0026498423 0.04292489 0.07511774
## 4 0.0011640379 0.03014667 0.04757577
## 5 0.0029842271 0.04479572 0.08021691
## 6 0.0023533123 0.03966387 0.07033209
To demonstrate tidying on DESeq2
objects we have used the published airway
RNA-Seq experiment, available as a package from Bioconductor:
source("https://bioconductor.org/biocLite.R")
biocLite("airway")
Import the airway
dataset:
library(DESeq2)
library(airway)
data(airway)
airway_se <- airway
airway_se
is a SummarizedExperiment
object, which is a type of object used by the DESeq2
package. Next, we create a DESeqDataSet
object and show the output of tidying this object:
airway_dds <- DESeqDataSet(airway_se, design = ~cell + dex)
head(tidy(airway_dds))
## # A tibble: 6 × 3
## gene sample count
## <chr> <chr> <int>
## 1 ENSG00000000003 SRR1039508 679
## 2 ENSG00000000005 SRR1039508 0
## 3 ENSG00000000419 SRR1039508 467
## 4 ENSG00000000457 SRR1039508 260
## 5 ENSG00000000460 SRR1039508 60
## 6 ENSG00000000938 SRR1039508 0
Only the gene counts are outputted since there has been no analysis performed. We perform an analysis on the data and then tidy
the resulting object:
# differential expression analysis
deseq <- DESeq(airway_dds)
results <- results(deseq)
# tidy results
tidy_results <- tidy(results)
head(tidy_results)
## # A tibble: 6 × 7
## gene baseMean estimate stderror statistic
## <chr> <dbl> <dbl> <dbl> <dbl>
## 1 ENSG00000000003 708.6021697 0.37415246 0.09884435 3.7852692
## 2 ENSG00000000005 0.0000000 NA NA NA
## 3 ENSG00000000419 520.2979006 -0.20206175 0.10974241 -1.8412367
## 4 ENSG00000000457 237.1630368 -0.03616686 0.13834540 -0.2614244
## 5 ENSG00000000460 57.9326331 0.08445399 0.24990709 0.3379415
## 6 ENSG00000000938 0.3180984 0.08413901 0.15133424 0.5559813
## # ... with 2 more variables: p.value <dbl>, p.adjusted <dbl>
As an example to show how easy it is to manipulate the resulting object, tidy_results
, we can use ggplot2
to create a volcano plot of the p-values:
ggplot(tidy_results, aes(x=estimate, y=log(p.value),
color=log(baseMean))) + geom_point(alpha=0.5) +
ggtitle("Volcano Plot For Airway Data via DESeq2") + theme_bw()
Here we use the hammer
dataset included in biobroom
package. edgeR
can be used to perform differential expression analysis as follows:
library(edgeR)
data(hammer)
hammer.counts <- Biobase::exprs(hammer)[, 1:4]
hammer.treatment <- Biobase::phenoData(hammer)$protocol[1:4]
y <- DGEList(counts=hammer.counts,group=hammer.treatment)
y <- calcNormFactors(y)
y <- estimateCommonDisp(y)
y <- estimateTagwiseDisp(y)
et <- exactTest(y)
The results of the analysis are stored in et
, which is an DGEExact
object. We can tidy
this object using biobroom
:
head(tidy(et))
## # A tibble: 6 × 4
## gene estimate logCPM p.value
## <chr> <dbl> <dbl> <dbl>
## 1 ENSRNOG00000000001 2.64635814 1.49216267 1.309933e-06
## 2 ENSRNOG00000000007 -0.40869816 -0.22616605 1.000000e+00
## 3 ENSRNOG00000000008 2.22296029 -0.40665547 1.288756e-01
## 4 ENSRNOG00000000009 0.00000000 -1.31347471 1.000000e+00
## 5 ENSRNOG00000000010 0.03307909 1.79448965 1.000000e+00
## 6 ENSRNOG00000000012 -3.39210151 0.07939132 3.745676e-03
glance
shows a summary of the experiment: the number of genes found significant (at a specified alpha
), and which contrasts were compared to get the results.
glance(et, alpha = 0.05)
## significant comparison
## 1 6341 control/L5 SNL
Additionally, we can can easily manipulate the resulting object and create a volcano plot of the p-values using ggplot2
:
ggplot(tidy(et), aes(x=estimate, y=log(p.value), color=logCPM)) +
geom_point(alpha=0.5) + ggtitle("Volcano Plot for Hammer Data via EdgeR") +
theme_bw()
To demonstrate how biobroom
works with limma
objects, we generate some simulated data to test the tidier for limma
objects.
# create random data and design
dat <- matrix(rnorm(1000), ncol=4)
dat[, 1:2] <- dat[, 1:2] + .5 # add an effect
rownames(dat) <- paste0("g", 1:nrow(dat))
des <- data.frame(treatment = c("a", "a", "b", "b"),
confounding = rnorm(4))
We then use lmFit
and eBayes
(functions included in limma
) to fit a linear model and use tidy
to convert the resulting object into tidy format:
lfit <- lmFit(dat, model.matrix(~ treatment + confounding, des))
eb <- eBayes(lfit)
head(tidy(lfit))
## # A tibble: 6 × 3
## gene term estimate
## <chr> <chr> <dbl>
## 1 g1 treatmentb 0.08332257
## 2 g2 treatmentb -0.04953880
## 3 g3 treatmentb -0.01453496
## 4 g4 treatmentb 0.41682061
## 5 g5 treatmentb -0.67130920
## 6 g6 treatmentb -0.70373680
head(tidy(eb))
## # A tibble: 6 × 6
## gene term estimate statistic p.value lod
## <chr> <chr> <dbl> <dbl> <dbl> <dbl>
## 1 g1 treatmentb 0.08332257 0.10584241 0.9191545 -4.602258
## 2 g2 treatmentb -0.04953880 -0.06378033 0.9512153 -4.602317
## 3 g3 treatmentb -0.01453496 -0.01630035 0.9875229 -4.602349
## 4 g4 treatmentb 0.41682061 0.54381547 0.6061535 -4.599990
## 5 g5 treatmentb -0.67130920 -0.76232807 0.4747325 -4.597911
## 6 g6 treatmentb -0.70373680 -0.78576895 0.4618724 -4.597659
Analysis can easily be performed from the tidied data. The package ggplot2
can be used to make a volcano plot of the p-values:
ggplot(tidy(eb), aes(x=estimate, y=log(p.value), color=statistic)) +
geom_point() + ggtitle("Nested Volcano Plots for Simulated Data Processed with limma") +
theme_bw()
tidy
can also be run directly on ExpressionSet
objects, as described in another popular Bioconductor
package Biobase.
The hammer
dataset we used above (which is included in the biobroom
package) is an ExpressionSet
object, so we’ll use that to demonstrate.
library(Biobase)
head(tidy(hammer))
## # A tibble: 6 × 3
## gene sample value
## <chr> <chr> <int>
## 1 ENSRNOG00000000001 SRX020102 2
## 2 ENSRNOG00000000007 SRX020102 4
## 3 ENSRNOG00000000008 SRX020102 0
## 4 ENSRNOG00000000009 SRX020102 0
## 5 ENSRNOG00000000010 SRX020102 19
## 6 ENSRNOG00000000012 SRX020102 7
We can add the phenotype information by using the argument addPheno
:
head(tidy(hammer, addPheno = TRUE))
## # A tibble: 6 × 8
## gene sample sample.id num.tech.reps protocol
## <chr> <chr> <fctr> <dbl> <fctr>
## 1 ENSRNOG00000000001 SRX020102 SRX020102 1 control
## 2 ENSRNOG00000000007 SRX020102 SRX020102 1 control
## 3 ENSRNOG00000000008 SRX020102 SRX020102 1 control
## 4 ENSRNOG00000000009 SRX020102 SRX020102 1 control
## 5 ENSRNOG00000000010 SRX020102 SRX020102 1 control
## 6 ENSRNOG00000000012 SRX020102 SRX020102 1 control
## # ... with 3 more variables: strain <fctr>, Time <fctr>, value <int>
Now we can easily visualize the distribution of counts for each protocol by using ggplot2
:
ggplot(tidy(hammer, addPheno=TRUE), aes(x=protocol, y=log(value))) +
geom_boxplot() + ggtitle("Boxplot Showing Effect of Protocol on Expression")
tidy
can also be run directly on MSnSet
objects from MSnbase
, which as specialised containers for quantitative proteomics data.
library(MSnbase)
data(msnset)
head(tidy(msnset))
## # A tibble: 6 × 3
## protein sample.id value
## <chr> <chr> <dbl>
## 1 X1 iTRAQ4.114 1347.6158
## 2 X10 iTRAQ4.114 739.9861
## 3 X11 iTRAQ4.114 27638.3582
## 4 X12 iTRAQ4.114 31892.8928
## 5 X13 iTRAQ4.114 26143.7542
## 6 X14 iTRAQ4.114 6448.0829
head(tidy(msnset, addPheno = TRUE))
## # A tibble: 6 × 5
## protein sample mz reporters value
## <fctr> <fctr> <dbl> <fctr> <dbl>
## 1 X1 iTRAQ4.114 114.1 iTRAQ4 1347.6158
## 2 X10 iTRAQ4.114 114.1 iTRAQ4 739.9861
## 3 X11 iTRAQ4.114 114.1 iTRAQ4 27638.3582
## 4 X12 iTRAQ4.114 114.1 iTRAQ4 31892.8928
## 5 X13 iTRAQ4.114 114.1 iTRAQ4 26143.7542
## 6 X14 iTRAQ4.114 114.1 iTRAQ4 6448.0829
All biobroom tidy
and augment
methods return a tbl_df
by default (this prevents them from printing many rows at once, while still acting like a traditional data.frame
). To change this to a data.frame
or data.table
, you can set the biobroom.return
option:
options(biobroom.return = "data.frame")
options(biobroom.return = "data.table")