Introduction
The NanostringNCTools package enables independent analysis of NanoString nCounter data (e.g. RCC/RLF data) in R.
In this vignette we will:
Create a Nanostring data object (e.g. NanoStringRCCSet) integrating RCC + RLF + Annotation data Learn to summarize, subset, transform the data object Perform and visualize quality control assessments of the data Perform data normalization for downstream analyses
The NanoStringRCCSet was inherited from Biobase’s ExpressionSet class. The NanoStringRCCSet class was designed to encapsulate data and corresponding methods for Nanostring RCC files generated from the NanoString nCounter platform.
Apply Functions Across Assay Data
Similar to the ExpressionSet’s esApply function, an equivalent method is available with NanoStringRCCSet objects. Functions can be applied to assay data feature- or sample-wise.
Add the demoElem data which is computed as the logarithm of the count matrix (exprs) into the demoData by using assayDataApply function. The accessor function assayDataElement from eSet returns matrix element from assayData slot of object.
- MARGIN: 1 (loop over the feature) or 2 (loop over the sample).
- FUN: the function you want to apply on the data, e.g. log, mean, etc.
- elt: refers to the element you want to apply on in the demoData.
# calculate the log counts of gene expressions for each sample
assayDataElement(demoData, "demoElem") <-
assayDataApply(demoData, MARGIN=2, FUN=log, base=10, elt="exprs")
assayDataElement(demoData, "demoElem")[1:3, 1:2]
#> SKMEL2-DMSO-8h-R1_04.RCC
#> Endogenous_TP53_NM_000546.2 2.540329
#> Endogenous_IL22RA2_NM_181310.1 0.698970
#> Endogenous_IL2_NM_000586.2 0.301030
#> SKMEL2-DMSO-8h-R2_04.RCC
#> Endogenous_TP53_NM_000546.2 2.7788745
#> Endogenous_IL22RA2_NM_181310.1 0.9542425
#> Endogenous_IL2_NM_000586.2 0.9030900
# calculate the mean of log counts for each feature
assayDataApply(demoData, MARGIN=1, FUN=mean, elt="demoElem")[1:5]
#> Endogenous_TP53_NM_000546.2 Endogenous_IL22RA2_NM_181310.1
#> 2.7380941 0.6850787
#> Endogenous_IL2_NM_000586.2 Endogenous_CCR5_NM_000579.1
#> 0.5595554 -Inf
#> Endogenous_PRLR_NM_001204318.1
#> 0.9411847
# split data by Treatment and calculate the mean of log counts for each feature
head(esBy(demoData,
GROUP = "Treatment",
FUN = function(x){
assayDataApply(x, MARGIN = 1, FUN=mean, elt="demoElem")
}))
#> DMSO VEM
#> Endogenous_TP53_NM_000546.2 2.6962710 2.7799171
#> Endogenous_IL22RA2_NM_181310.1 0.5927171 0.7774403
#> Endogenous_IL2_NM_000586.2 0.4808935 0.6382173
#> Endogenous_CCR5_NM_000579.1 -Inf 0.7821118
#> Endogenous_PRLR_NM_001204318.1 1.0082754 0.8740940
#> Endogenous_LIF_NM_002309.3 1.8067671 1.4661876
There is also a preloaded nCounter normalization method that comes with the NanoStringRCCSet class. This includes the default normalization performed in nSolver.
- type: data type of the object. Values maybe RNA, protein.
- fromELT: name of the assayDataElement to normalize.
- toELT: name of the assayDataElement to store normalized values.
Below is a heatmap of log normalized data generated by NanoStringRCCSet autoplot method with unsupervised clustering dendrograms. Each row of the heatmap represents a gene and each column represents a sample. Users can custom the heatmap by setting different parameters.
- type: reference the type of plot to generate, in this case, “heatmap-genes”.
- elt: the name of the expression matrix, i.e. the data you used to generate the plot.
- heatmapGroup: referencing pData columns to color samples by in heatmap.
- show_colnames_gene_limit: determine the number of features to display column-wise. In this case, the number of samples is larger than the limit, no column names will be displayed.
- show_rownames_gene_limit: determine the number of features to display row-wise. In this case, the number of genes is larger than the limit, no row names will be displayed.
- log2scale: indicating expression data is on log2 scale
Built-in Quality Control Assessment
Quality control metrics are used to assess the technical performance of the nCounter profiling assay in a study. Users can flag samples that fail QC thresholds or have borderline results based on housekeeper and ERCC expression, imaging quality, binding density. First, housekeeper genes assess sample integrity by comparing the observed value versus a predetermined threshold for suitability for data analysis. The machine performance is assessed using percentage of fields of view that were attempted versus those successfully analyzed. The binding density of the probes within the imaging area, ERCC linearity, and limit of detection are used as readouts of the efficiency and specificity of the chemistry of the assay. Any sample deemed as failing any one of these QC checkpoints will be removed from the analysis.
Additionally, QC results can be visualized using the NanoStringRCCSet autoplot method.
- type: reference the type of plot to generate.
- qcCutoffs: you can also set qcCutoffs in the autoplot function. Since it has been set in the setQCFlags function, you don’t need to set this parameter again in autoplot.
# Use the setSeqQCFlags function to set Sequencing QC Flags to your dataset. The default cutoff are displayed in the function.
demoData <- setQCFlags(demoData,
qcCutoffs = list(Housekeeper = c(failingCutoff = 32, passingCutoff = 100),
Imaging = c(fovCutoff = 0.75),
BindingDensity = c(minimumBD = 0.1, maximumBD = 2.25, maximumBDSprint = 1.8),
ERCCLinearity = c(correlationValue = 0.95),
ERCCLoD = c(standardDeviations = 2)))
# show the last 6 column names in the data
tail(svarLabels(demoData))
#> [1] "BindingDensity" "CartridgeID" "CartridgeBarcode"
#> [4] "Sample ID" "QCFlags" "QCBorderlineFlags"
# show the first 2 rows of the QC Flags results
head(protocolData(demoData)[["QCFlags"]], 2)
#> Imaging Binding Linearity LoD Housekeeping
#> SKMEL2-DMSO-8h-R1_04.RCC FALSE FALSE FALSE FALSE FALSE
#> SKMEL2-DMSO-8h-R2_04.RCC FALSE FALSE FALSE FALSE FALSE
# show the first 2 rows of the QC Borderline Flags results
head(protocolData(demoData)[["QCBorderlineFlags"]], 2)
#> Imaging Binding Linearity LoD Housekeeping
#> SKMEL2-DMSO-8h-R1_04.RCC FALSE FALSE FALSE FALSE FALSE
#> SKMEL2-DMSO-8h-R2_04.RCC FALSE FALSE FALSE FALSE FALSE
Housekeeping Genes QC
The HK Genes QC plot shows the geometric mean of housekeeper genes in each sample. Each dot represents a sample in this plot. The sample IDs are labeled at x-axis. The corresponding geometric mean of housekeeper genes are at y-axis. If you hover mouse over a point, you can find the sample name and its geometric mean. Samples with low housekeeper signal suffer from either low sample input or low reaction efficiency. Ideally the geometric mean of counts will be above 100 for all samples, and a minimum geometric mean of 32 counts is required for analysis. Samples in-between these two thresholds are considered in the analysis, but results from these “borderline” samples should be treated with caution. In this case, all samples are above 100, so they all pass Housekeeping Genes QC.
You can generate QC plots with a subset of data using the subset function. The HK Genes QC plot below only contains samples with Treatment DMSO.
Binding Density QC
The binding density represents the concentration of barcodes measured by the instrument in barcodes per square micron. Each dot in this QC plot represents a sample. The lane ID of samples are labeled at x-axis and the binding density is at y-axis. If you hover mouse over a dot, it will display its Sample ID, lane ID and binding density. The Digital Analyzer may not be able to distinguish each probe from the others if too many are present. The ideal range for assays run on an nCounter MAX or FLEX system is 0.1 - 2.25 spots per square micron and assays run on the nCounter SPRINT system should have a range of 0.1 - 1.8 spots per square micron. In this case, all samples are in the ideal range, so they all pass Binding Density QC.
Imaging QC
The Imaging QC metric reports the percentage of fields of view (FOVs) the Digital Analyzer or SPRINT was able to capture. Each dot represents a sample. The lane IDs are labeled at x-axis and the counted FOV is at y-axis. If you hover mouse over a point, it will display its sample ID, lane ID and counted FOV. At least 75% of FOVs should be successful to obtain robust data. In this case, all samples passed the 75% threshold, so they all pass Imaging QC.
ERCC Linearity QC
The ERCC Linearity QC metric performs a correlation analysis after Log(2) transformation of the expression values. Each line in this plot represents a sample. The concentration is labeled at x-axis and gene expressions are displayed at y-axis. If you hover mouse over a line, it will show the sample ID, concentration, gene expresson and the correlation. The correlation is tested between the known concentrations of positive control target molecules added by NanoString and the resulting Log(2) counts. Correlation values lower than 0.95 may indicate an issue with the hybridization reaction and/or assay performance. In this case, all samples have correlation values above or equal to 0.95, so they all pass ERCC linearity QC.
ERCC LOD QC
The ERCC limit of detection of the assay compares the positive control probes and the negative control probes. Specifically, it is expected that the 0.5 fM positive control probe (Pos_E) will produce raw counts that are at least two standard deviations higher than the mean of the negative control probes (represented by the boxplot). Each dot in this plot represents a sample. The sample IDs are displayed at x-axis and the raw counts of Pos_E are at y-axis. If you hover mouse over a point, it will show the sample ID and its Pos_E counts. The critical value for each sample is drawn as a red horizontal line for each sample. In this case, all samples pass the LOD QC.
