Riborex Manual

Wenzheng Li, Weili Wang, Philip J. Uren, Luiz OF Penalva, Andrew D. Smith
18 September, 2017

Introduction
Riborex is a R package for identifying differentially translated genes from Ribo-seq data. Riborex integrates
both RNA- and Ribo-seq read count data into a single generalized linear model (GLM) and generates a
modified design matrix reflecting the integration. At its core, Riborex applies existing RNA-seq analysis tools
such as edgeR, DESeq2 and Voom to this modified design matrix and identifies differential translation across
conditions.

Detailed example
First, we need to load Riborex library.
library(riborex)

The input for Riborex are two read count tables summarized from RNA-seq and Ribo-seq data respectively.
The read count table should be organized as a data frame with rows correspond to genes and columns
correspond to samples as shown below.
data(riborexdata)
RNACntTable <- rna
RiboCntTable <- ribo

We can check the first five lines of the table:

head(RNACntTable,5)

## BN_336 BN_337 BN_338 BN_339

## ENSRNOG00000000017 7 11 4 4
## ENSRNOG00000000024 2467 2478 3258 2316
## ENSRNOG00000000033 206 282 330 244
## ENSRNOG00000000034 758 672 1335 767
## ENSRNOG00000000036 237 163 211 189
head(RiboCntTable,5)

## BN_341 BN_342 BN_343 BN_344

## ENSRNOG00000000017 15 5 10 2
## ENSRNOG00000000024 5206 5921 2864 1985
## ENSRNOG00000000033 30 30 23 13
## ENSRNOG00000000034 943 775 842 311
## ENSRNOG00000000036 80 49 30 7
Then we need to prepare two vectors to indicate the treatments of samples in RNA- and Ribo-seq data. Both
RNA-seq and Ribo-seq can have different number of samples in control and treated condtions, and RNA-seq
and Ribo-seq data can have different number of samples.
rnaCond <- c("control", "control", "treated", "treated")
riboCond <- c("control", "control", "treated", "treated")

1
After the two read count table and two condition vectors are ready, we can use riborex (), and we can choose
which engine to use. By default, DESeq2 is used as the engine if you don’t specify the engine option. Use
help(riborex) in R to see more details about this function.
res.deseq2 <- riborex(RNACntTable, RiboCntTable, rnaCond, riboCond)

The format of the result is the same when DESeq2 is used in RNA-seq analysis.
res.deseq2

## log2 fold change (MLE): EXTRA1 treated vs control

## Wald test p-value: EXTRA1 treated vs control
## DataFrame with 13916 rows and 6 columns
## baseMean log2FoldChange lfcSE stat
## <numeric> <numeric> <numeric> <numeric>
## ENSRNOG00000000017 7.694934 1.53323724 1.5662846 0.9789008
## ENSRNOG00000000024 3544.238071 -0.10215095 0.2020511 -0.5055699
## ENSRNOG00000000033 111.439451 0.26525296 0.5033680 0.5269564
## ENSRNOG00000000034 783.974916 0.06513995 0.3506098 0.1857904
## ENSRNOG00000000036 99.151141 -0.63425877 0.5910827 -1.0730457
## ... ... ... ... ...
## ENSRNOG00000061895 105.24110 0.52804379 0.4846633 1.0895065
## ENSRNOG00000061899 40.04057 -0.49721003 0.7802896 -0.6372122
## ENSRNOG00000061910 4237.53481 0.30904740 0.2460783 1.2558905
## ENSRNOG00000061928 1651.61680 -0.08099135 0.2465675 -0.3284753
## ENSRNOG00000061989 107.36281 -0.10482910 0.4554566 -0.2301627
## pvalue padj
## <numeric> <numeric>
## ENSRNOG00000000017 0.3276290 0.7366558
## ENSRNOG00000000024 0.6131586 0.9019205
## ENSRNOG00000000033 0.5982239 0.8963865
## ENSRNOG00000000034 0.8526091 0.9748558
## ENSRNOG00000000036 0.2832506 0.6901202
## ... ... ...
## ENSRNOG00000061895 0.2759306 0.6824115
## ENSRNOG00000061899 0.5239866 0.8635147
## ENSRNOG00000061910 0.2091557 0.5984207
## ENSRNOG00000061928 0.7425523 0.9428604
## ENSRNOG00000061989 0.8179654 0.9676247
You can check the distribution p-values.
hist(res.deseq2$pvalue, main = 'DESeq2 unadjusted p-values', xlab='Unadjusted p-values')

2
 DESeq2 unadjusted p−values
2500
Frequency

1500
500
0

0.0 0.2 0.4 0.6 0.8 1.0

Unadjusted p−values
We can see for this dataset, the p-value distribution is as expected based on DESeq2 manual which is uniformly
distribution with differrentially expressed genes enriched with small p-values. We will show another dataset
later for which the p-value distribution is skew to the right and how it can be fixed with fdrtool.
Also, you can use summary () for your results.
summary(res.deseq2)

##
## out of 13916 with nonzero total read count
## adjusted p-value < 0.1
## LFC > 0 (up) : 1107, 8%
## LFC < 0 (down) : 1217, 8.7%
## outliers [1] : 0, 0%
## low counts [2] : 540, 3.9%
## (mean count < 6)
## [1] see 'cooksCutoff' argument of ?results
## [2] see 'independentFiltering' argument of ?results
And results can be saved by:
write.table(res.deseq2, "riborex_res_deseq2.txt", quote=FALSE)

If you want to use edgeR as your engine, you can use riborex () as:
res.edgeR <- riborex(RNACntTable, RiboCntTable, rnaCond, riboCond, "edgeR")

The format of the result is the same when edgeR is used in RNA-seq analysis.
head(res.edgeR$table)

## logFC logCPM LR PValue FDR

## ENSRNOG00000000017 1.36290149 -2.456968 1.80649827 0.1789289 0.4627345

3
## ENSRNOG00000000024 -0.30127172 6.212404 2.13670654 0.1438103 0.4037250
## ENSRNOG00000000033 0.07178854 1.313235 0.02631769 0.8711269 0.9636097
## ENSRNOG00000000034 -0.13430329 4.029136 0.12541035 0.7232390 0.9111612
## ENSRNOG00000000036 -0.82540899 1.132478 2.15554356 0.1420562 0.4008219
## ENSRNOG00000000040 -0.19057283 -1.555003 0.07537378 0.7836675 0.9338658
For edgeR engine, you can also choose to estimate dispersion of RNA-seq and Ribo-seq data separately by
specifying engine as “edgeRD”.
res.edgeRD <- riborex(RNACntTable, RiboCntTable, rnaCond, riboCond, "edgeRD")

If you want to use Voom as the engine, you can run riborex () as:
res.voom <- riborex(RNACntTable, RiboCntTable, rnaCond, riboCond, "Voom")

The format of the result is the same when Voom is used in RNA-seq analysis.
head(res.voom)

## logFC AveExpr t P.Value adj.P.Val

## ENSRNOG00000000017 1.39674189 -2.8437969 1.2847048 0.2268559 0.5243467
## ENSRNOG00000000024 -0.30047073 6.0075571 -1.8737990 0.0893910 0.2991861
## ENSRNOG00000000033 0.07661457 0.7070877 0.1687028 0.8692764 0.9540355
## ENSRNOG00000000034 -0.13777833 3.9701893 -0.3933633 0.7020190 0.8871606
## ENSRNOG00000000036 -0.89165405 0.7106061 -1.3890222 0.1939374 0.4826701
## ENSRNOG00000000040 -0.20967655 -1.7042630 -0.2991046 0.7707701 0.9157225
## B
## ENSRNOG00000000017 -4.928058
## ENSRNOG00000000024 -5.682138
## ENSRNOG00000000033 -6.320171
## ENSRNOG00000000034 -7.130133
## ENSRNOG00000000036 -5.425369
## ENSRNOG00000000040 -5.854482

Case-study with “incorrect” p-value distribution

RNACntTable.corrected <- rna.null

RiboCntTable.corrected <- ribo.null

We can check the first five lines of the table:

head(RNACntTable.corrected)

## rna_T0_r1 rna_T0_r2 rna_T0_r3 rna_T24_r1 rna_T24_r2

## ENSG00000116032.5 0 0 0 0 2
## ENSG00000188026.12 803 691 763 1118 973
## ENSG00000171174.13 198 149 236 227 193
## ENSG00000166257.8 0 0 0 0 0
## ENSG00000149136.8 9301 7705 12490 4795 4842
## ENSG00000136938.8 6325 5433 8880 3163 4074
## rna_T24_r3
## ENSG00000116032.5 0
## ENSG00000188026.12 656
## ENSG00000171174.13 198
## ENSG00000166257.8 0
## ENSG00000149136.8 4151

4
## ENSG00000136938.8 2996
head(RiboCntTable.corrected)

## ribo_T0_r1 ribo_T0_r2 ribo_T0_r3 ribo_T24_r1

## ENSG00000116032.5 2 0 0 2
## ENSG00000188026.12 382 432 360 784
## ENSG00000171174.13 75 113 77 161
## ENSG00000166257.8 0 0 0 0
## ENSG00000149136.8 2536 3546 2702 2743
## ENSG00000136938.8 892 1473 1060 979
## ribo_T24_r2 ribo_T24_r3
## ENSG00000116032.5 2 5
## ENSG00000188026.12 880 890
## ENSG00000171174.13 220 164
## ENSG00000166257.8 3 2
## ENSG00000149136.8 3678 2765
## ENSG00000136938.8 1473 947
The condition vectors can be created as:
rnaCond.corrected <- c(rep('T0', 3), rep('T24',3))
riboCond.corrected <- rnaCond.corrected
rnaCond.corrected

## [1] "T0" "T0" "T0" "T24" "T24" "T24"

riboCond.corrected

## [1] "T0" "T0" "T0" "T24" "T24" "T24"

The results from DESeq2 can be obtained as:
res.deseq2.corrected <- riborex(RNACntTable.corrected, RiboCntTable.corrected, rnaCond.corrected, riboCo

## DESeq2 mode selected

## combining design matrix
## applying DESeq2 to modified design matrix
## estimating size factors
## estimating dispersions
## gene-wise dispersion estimates
## mean-dispersion relationship
## final dispersion estimates
## fitting model and testing
We can check the p-value distribution as:
hist(res.deseq2.corrected$pvalue, main = 'DESeq2 unadjusted p-values', xlab='Unadjusted p-values')

5
 DESeq2 unadjusted p−values
800
600
Frequency

400
200
0

0.0 0.2 0.4 0.6 0.8 1.0

Unadjusted p−values We can

see from the histogram that the distribution of p-values is skew to the right, that means the null dis-
tribution is not “correct”, we can fix it by reestimating the p-values using fdrtool:
results.corrected <- correctNullDistribution(res.deseq2.corrected)

## correcting null distribution by reestimating pvalues

## Step 1... determine cutoff point
## Step 2... estimate parameters of null distribution and eta0
## Step 3... compute p-values and estimate empirical PDF/CDF
## Step 4... compute q-values and local fdr
We can see the p-value distribution after correction:
hist(results.corrected$pvalue, main = 'DESeq2 unadjusted p-values after correction',
xlab='Corrected unadjusted p-values')

6
 DESeq2 unadjusted p−values after correction
800
Frequency

600
400
200
0

0.0 0.2 0.4 0.6 0.8 1.0

Corrected unadjusted p−values We can

see after the correction, the distribution of p-values is as expected. And the adjusted pvalues are cor-
rected also.

Multi-factor experiment
Since we don’t find any available ribosome profiling data generated in a multi-factor experiement, here we
generate a pseudo dataset to demonstrate the usage of riborex in a multi-factor experiment. The pseudo
dataset have 8 samples in RNA-seq and Ribo-seq, and two factors are included.
rna <- RNACntTable[,c(1,2,3,4,1,2,3,4)]
ribo <- RiboCntTable[,c(1,2,3,4,1,2,3,4)]

For multi-factor experiment, we prepare two data frames to indicate the treatment under each factor. Here
for the 8 samples in both RNA- and Ribo-seq experiement, the 3rd and 4th samples are treated with drug1
and the 7th and 8th samples are treated with drug2.
rnaCond <- data.frame(factor1=(c("control1", "control1", "treated1", "treated1",
"control1", "control1", "control1", "control1")),
factor2=(c("control2", "control2", "control2", "control2",
"control2", "control2", "treated2", "treated2")))

riboCond <- data.frame(factor1=(c("control1", "control1", "treated1", "treated1",

"control1", "control1", "control1", "control1")),
factor2=(c("control2", "control2", "control2", "control2",
"control2", "control2", "treated2", "treated2")))

Also we need to prepare a contrast to specify the comparison we want to perform, for example, if we want to
compare the influence of the usage of drug2. The contrast can be constructed as:

7
contrast = c("factor2", "control2", "treated2")

Then riborex () is used with contrast specified.

res.deseq2 <- riborex(rna, ribo, rnaCond, riboCond, "DESeq2", contrast = contrast)

We can see the summary of the result:

summary(res.deseq2)

##
## out of 13916 with nonzero total read count
## adjusted p-value < 0.1
## LFC > 0 (up) : 1887, 14%
## LFC < 0 (down) : 1987, 14%
## outliers [1] : 0, 0%
## low counts [2] : 270, 1.9%
## (mean count < 3)
## [1] see 'cooksCutoff' argument of ?results
## [2] see 'independentFiltering' argument of ?results
edgeR and edgeRD can be used in a similar way.
res.edgeR <- riborex(rna, ribo, rnaCond, riboCond, "edgeR", contrast = contrast)

res.edgeRD <- riborex(rna, ribo, rnaCond, riboCond, "edgeRD", contrast = contrast)

Currently, you can’t choose Voom as the engine in a multi-factor experiment yet.

Setup
This analysis was conducted on
sessionInfo()

## R version 3.3.2 (2016-10-31)

## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 17.04
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] parallel stats4 stats graphics grDevices utils datasets
## [8] methods base
##
## other attached packages:
## [1] riborex_2.3.4 fdrtool_1.2.15
## [3] edgeR_3.16.5 limma_3.30.13
## [5] DESeq2_1.16.1 SummarizedExperiment_1.4.0
## [7] Biobase_2.34.0 GenomicRanges_1.26.4

8
## [9] GenomeInfoDb_1.10.3 IRanges_2.8.2
## [11] S4Vectors_0.12.2 BiocGenerics_0.22.0
##
## loaded via a namespace (and not attached):
## [1] genefilter_1.58.1 locfit_1.5-9.1 splines_3.3.2
## [4] lattice_0.20-35 colorspace_1.3-2 htmltools_0.3.6
## [7] yaml_2.1.14 base64enc_0.1-3 blob_1.1.0
## [10] survival_2.41-3 XML_3.98-1.6 rlang_0.1.2
## [13] foreign_0.8-69 DBI_0.7 BiocParallel_1.8.2
## [16] bit64_0.9-7 RColorBrewer_1.1-2 plyr_1.8.4
## [19] stringr_1.2.0 zlibbioc_1.20.0 munsell_0.4.3
## [22] gtable_0.2.0 htmlwidgets_0.9 memoise_1.1.0
## [25] evaluate_0.10.1 latticeExtra_0.6-28 knitr_1.17
## [28] geneplotter_1.52.0 AnnotationDbi_1.38.0 htmlTable_1.9
## [31] Rcpp_0.12.12 acepack_1.4.1 xtable_1.8-2
## [34] scales_0.5.0 backports_1.1.0 checkmate_1.8.3
## [37] Hmisc_4.0-3 annotate_1.52.1 XVector_0.14.1
## [40] bit_1.1-12 gridExtra_2.3 ggplot2_2.2.1
## [43] digest_0.6.12 stringi_1.1.5 grid_3.3.2
## [46] rprojroot_1.2 tools_3.3.2 bitops_1.0-6
## [49] magrittr_1.5 lazyeval_0.2.0 RCurl_1.95-4.8
## [52] tibble_1.3.4 RSQLite_2.0 Formula_1.2-2
## [55] cluster_2.0.6 Matrix_1.2-11 data.table_1.10.4
## [58] rmarkdown_1.6 rpart_4.1-11 nnet_7.3-12