Accessing metadata

Metadata in an RSet is stored in the same slots as in a PharmacoSet, and can be accessed using the same generic accessor functions as in PharmacoGx. A unique slot, radiation has additional accessor methods to retrieve the radiation types used in a given sensitivity experiment.

# Get the radiation info data.frame from an RSet
radInf <- radiationInfo(clevelandSmall)

knitr::kable(radInf)

Currently, only one type of radiation has been used in an RSet. However, we hope to add new RSets covering a wider range of radiation sensitivity and perturbation experiments in the near future. The following method is also available to retrieve the radiation types as a character vector instead of a data.frame.

radTypes <- radiationTypes(clevelandSmall)
radTypes

## [1] "radiation"

Accessing molecular data

Molecular data in an RSet is contained in the molecularProfiles slot and can be accessed the same way it is for a PSet.

# Get the list (equivalent to @molecularProfiles, except that it is robust to changes in RSet structure
str(molecularProfilesSlot(clevelandSmall), max.level=2)

## List of 4
##  $ rna     :Formal class 'SummarizedExperiment' [package "SummarizedExperiment"] with 5 slots
##  $ rnaseq  :Formal class 'SummarizedExperiment' [package "SummarizedExperiment"] with 5 slots
##  $ mutation:Formal class 'SummarizedExperiment' [package "SummarizedExperiment"] with 5 slots
##  $ cnv     :Formal class 'SummarizedExperiment' [package "SummarizedExperiment"] with 5 slots

# Get the names from the list
mDataNames(clevelandSmall)

## [1] "rna"      "rnaseq"   "mutation" "cnv"

All molecular data in an RSet (any class inheriting from CoreSet, actually) is contained in a SummarizedExperiment object. While SummarizedExperiment comes with it’s own set of accessors, we recommend using available RadioGx accessor methods as it allows your scripts to be robust to future changes in the structure of a RadioSet object.

# Get sample metadata
phenoInf <- phenoInfo(clevelandSmall, 'cnv')

To keep the document formatted nicely, the following tables have been subset to the first three rows and columns.

# Get feature metadata
featInfo <- featureInfo(clevelandSmall, 'rna')

# Access the moleclar feature data
mProf <- molecularProfiles(clevelandSmall, 'rnaseq')

Accessing response data

Data from radiation sensitivity and/or perturbation experiments is also retrieved the same way it is for a PSet. Currently, only sensitivity experiments have been included in a RadioSet.

# Get sensitivity slot
sens <- treatmentResponse(clevelandSmall)

# Get sensitivity raw data
sensRaw <- sensitivityRaw(clevelandSmall)

# Get sensitivity profiles
sensProf <- sensitivityProfiles(clevelandSmall)

# Get sensitivity info
sensInfo <- sensitivityInfo(clevelandSmall)

Dose-Response Curves

The doseResponseCurve function can be used to generate plots of surviving fraction vs dose for radiation sensitivity experiments. In this example we provide raw data values to create the plot. When the plot.type is set to “Both”, a linear-quadratic model will also be fit to the supplied dose-response values.

doseResponseCurve(
  Ds=list("Experiment 1" = c(0, 2, 4, 6)),
  SFs=list("Experiment 1" = c(1,.6,.4,.2)),
  plot.type="Both"
)

Additionally, doseResponseCurve can be used to create dose response curves directly from a curated RadioSet object. When utilizing this feature, a cell-line must be selected from the RadioSet. This can be done by name if you know which cell-line you are looking for. If you don’t know which cell-line you want to visualize, the available cell-lines can be explored using the cellInfo function.

doseResponseCurve(
  rSets=list(clevelandSmall),
  cellline=cellInfo(clevelandSmall)$sampleid[5]
)

Summarizing Sensitivity

To retrieve a radiation type by cell-line summary of a sensitivity experiment, we use the summarizeSensitivityProfiles function. This will return a matrix where rows are radiation type³, columns are cell-line and values are viability measurements summarized using summary.stat⁴. The sensitivity measure to summarize can be specified using sensitivity.measure⁵.

sensSummary <- summarizeSensitivityProfiles(clevelandSmall)

sensSummary[, 1:3]

##   CHP-212    IMR-32 KP-N-S19s 
## 1.3440925 0.5490508 2.3251731

Summarizing Molecular Data

mprofSummary <- summarizeMolecularProfiles(clevelandSmall, mDataType='rna', summary.stat='median', fill.missing=FALSE)

mprofSummary

## class: SummarizedExperiment 
## dim: 20049 5 
## metadata(3): experimentData annotation protocolData
## assays(2): exprs se.exprs
## rownames(20049): ENSG00000000003 ENSG00000000005 ... ENSG00000280439
##   ENSG00000280448
## rowData names(7): Probe EnsemblGeneId ... BEST rownames
## colnames(5): CHP-212 IMR-32 KP-N-S19s MHH-NB-11 NB1
## colData names(25): samplename filename ... rownames tissueid

Due to a lack of replicates in the clevelandSmall RSet, the returned SummarizedExperiment object contains the same information as the original. For other experiments with replicates, however, the result should contain one column per unique cell-line id. For ease of interoperability with the response data contained in an RSet, if fill.missing is FALSE empty columns for the cell-lines in the sensitivity experiment, but not in the molecular profile will be added to the SummarizedExperiment such that the dimensions are equal.

Molecular Signatures for Biomarker Discovery

The true usefulness of the RadioGx packages comes from the ability to determine gene signatures for a cell-lines from a sensitivity experiment. Cell-lines of interest to a given researcher can be selected and a molecular signature computed which correlates specific molecular features with a given sensitivity profile. Using this method one could identify signatures associated with either radio-sensitivity or radio-resistance. Combining this signature with drug response gene signatures from PharmacoGx, as will be done in the subsequent section, one can identify drugs which could augment the effectiveness of a given radiation signature. Perhaps more powerfully, drugs which target features associated with radio-resistance can be found, potentially synergistically increasing the overall effectiveness of the combined treatment.

radSensSig <- radSensitivitySig(clevelandSmall, mDataType='rna', features=fNames(clevelandSmall, 'rna')[2:5], nthread=1)

## Computing radiation sensitivity signatures...

radSensSig@.Data

## , , estimate
## 
##                  radiation
## ENSG00000000005 -0.3794080
## ENSG00000000419  0.7569346
## ENSG00000000457  0.9303156
## ENSG00000000460  0.4064764
## 
## , , se
## 
##                 radiation
## ENSG00000000005 0.5341815
## ENSG00000000419 0.3772930
## ENSG00000000457 0.2117490
## ENSG00000000460 0.5275026
## 
## , , n
## 
##                 radiation
## ENSG00000000005         5
## ENSG00000000419         5
## ENSG00000000457         5
## ENSG00000000460         5
## 
## , , tstat
## 
##                  radiation
## ENSG00000000005 -0.7102606
## ENSG00000000419  2.0062252
## ENSG00000000457  4.3934837
## ENSG00000000460  0.7705675
## 
## , , fstat
## 
##                  radiation
## ENSG00000000005  0.5044701
## ENSG00000000419  4.0249395
## ENSG00000000457 19.3026988
## ENSG00000000460  0.5937743
## 
## , , pvalue
## 
##                  radiation
## ENSG00000000005 0.52877664
## ENSG00000000419 0.13848843
## ENSG00000000457 0.02184966
## ENSG00000000460 0.49708576
## 
## , , df
## 
##                 radiation
## ENSG00000000005         3
## ENSG00000000419         3
## ENSG00000000457         3
## ENSG00000000460         3
## 
## , , fdr
## 
##                  radiation
## ENSG00000000005 0.52877664
## ENSG00000000419 0.27697686
## ENSG00000000457 0.08739866
## ENSG00000000460 0.52877664