Chapter 5 Normalization

The main function is r_norm_dataset, and it normalizes a dataset based on customizable templates. Depending on the template chosen, some parameters are used and others not.

5.1 Parameters

Here are the parameters:

param	short desc	class	required	default	description
datasetName	object name	character	required
template	template to apply	character	optional	none	one of: “logarithm”, “scaling”, “norm_quantiles”, “combat”, “rnaseqcount”, “unlog_rnaseqcount”, “unlog_metagenomics_2”, “metagenomics_1”, “metagenomics_2”, “microarray”, “compositional”
log_base	base of the logarithm	character	optional	“2”	can be: ‘2’, ‘e’, ‘10’
prior.count	offset of the logarithm	numeric	optional	0.5
offset	simple offset	numeric	optional	0.5
filter_type	type of filter on variables	character	optional	‘absolute’	can be: ‘max’, ‘absolute’, ‘relative’
filter_threshold	threshold for the filters	numeric	optional	5
normalization	type of normalization	character	optional	CLR	can be: ‘CLR’, ‘ILR’, ‘CSS’
scale_reduce	when scaling, should the data be reduced?	boolean	optional	FALSE
unlog_base	base of the logarithm when unlogging	character	optional	same as log_base
unlog_prior.count	offset when unlogging	numeric	optional	0.5
datasetName_batches	dataset name in which to find the batch variable	character	optional	NULL
varName_batches	name of the batch variable	character	optional	NULL
has_log	should cpm+log be applied?	boolean	optional	TRUE
has_filter	should a filter be applied	boolean	optional	TRUE
has_TMM	should TMM factors be computed?	boolean	optional	TRUE
has_TMMwsp	should TMMwsp factors be computed?	boolean	optional	FALSE
datasetName_colours	dataset name in which to find the variable to colour plots	character	optional	NULL
varName_colours	name of the variable to colour plots	character	optional	NULL

return a list of two objects (the dataset and its analysis) and three tables (two data views and one string: the name of the data object). When template is ‘none’, objects are not returned.

5.2 Outputs of the function

This data can only be processed with templates: logarithm, combat, rnaseqcount

res <- r_norm_dataset(datasetName = "proteins", template = "logarithm", prior.count = 10)
names(res)

## [1] "Table"     "Graphical" "Object"

Name of the produced object:

5.2.1 Table

The Table component only contains the name of the produced object.

res$Table

## $ObjectName
## [1] "normalized_1"

5.2.2 Graphs

2 graphs are produced (here boxplots before and after the normalization).

For all templates, always 2 graphs are produced.

names(res$Graphical)

## [1] "PlotBefore" "PlotAfter"

res$Graphical$PlotBefore

res$Graphical$PlotAfter

5.3 Different templates and options

Depending on the nature of the selected dataset, as well as on its attributes logt, normalized and norm_factors, different templates are proposed, following the logic in the table below. The column norm_factors denotes the presence / absence of normalization factors (that can be saved as attributes in the dataset object). Note that the natures rna-count and metabolite-compo are treated in the same way as metagenomics-count and metagenomics-compo, respectively. The nature general stands for any nature not mentioned here.

nature	logt	normalized	norm_factors	templates
general	no	no	FALSE	logarithm, scaling, norm_quantiles, combat
general	no	yes	FALSE	logarithm, scaling
general	yes	no	FALSE	scaling, norm_quantiles, combat
general	yes	yes	FALSE	scaling
metagenomics-compo	no	no	FALSE	logarithm, compositional
metagenomics-compo	no	yes	FALSE	logarithm
metagenomics-compo	yes	no	FALSE	scaling
metagenomics-compo	yes	yes	FALSE	scaling
metagenomics-count	no	yes	TRUE	metagenomics_2
metagenomics-count	yes	yes	TRUE	scaling
metagenomics-count	no	no	FALSE	logarithm, combat, metagenomics_1, metagenomics_2
metagenomics-count	no	yes	FALSE	logarithm, metagenomics_2
metagenomics-count	yes	no	FALSE	scaling, combat, unlog_metagenomics_2
metagenomics-count	yes	yes	FALSE	scaling
rna-count	no	yes	TRUE	rnaseqcount
rna-count	yes	yes	TRUE	scaling
rna-count	no	no	FALSE	logarithm, combat, rnaseqcount
rna-count	no	yes	FALSE	logarithm, rnaseqcount
rna-count	yes	no	FALSE	scaling, combat, unlog_rnaseqcount
rna-count	yes	yes	FALSE	scaling
microarray	no	no	FALSE	logarithm, scaling, combat, microarray
microarray	no	yes	FALSE	logarithm, scaling
microarray	yes	no	FALSE	scaling, combat
microarray	yes	yes	FALSE	scaling
metabolite-compo	no	no	FALSE	logarithm, compositional
metabolite-compo	no	yes	FALSE	logarithm
metabolite-compo	yes	no	FALSE	scaling
metabolite-compo	yes	yes	FALSE	scaling

The table below shows how logt and normalized are changed by applying the templates above, along with some parameter combinations. Note that rnaseqcount has the same effects as metagenomics_2, while unlog_rnaseqcount has the same effects as unlog_metagenomics_2. On top of modifying these attributes, the template metagenomics_1 also modifies the data nature: it turns from metagenomics-count to metagenomics-compo.

template	has_TMM + has_TMMwsp	has_log	normalization	normalized	logt
logarithm					yes
scaling
norm_quantiles				yes
combat				yes
metagenomics_1			CLR	yes	yes
metagenomics_1			CSS	yes
metagenomics_2	TRUE+FALSE	TRUE		yes	yes
metagenomics_2	FALSE+TRUE	TRUE		yes	yes
metagenomics_2	FALSE+FALSE	TRUE			yes
metagenomics_2	TRUE+FALSE	FALSE		yes
metagenomics_2	FALSE+TRUE	FALSE		yes
metagenomics_2	FALSE+FALSE	FALSE
unlog_metagenomics_2	TRUE+FALSE	TRUE		yes
unlog_metagenomics_2	FALSE+TRUE	TRUE		yes
unlog_metagenomics_2	FALSE+FALSE	TRUE
unlog_metagenomics_2	TRUE+FALSE	FALSE		yes	no
unlog_metagenomics_2	FALSE+TRUE	FALSE		yes	no
unlog_metagenomics_2	FALSE+FALSE	FALSE			no
rnaseqcount	TRUE+FALSE	TRUE		yes	yes
rnaseqcount	FALSE+TRUE	TRUE		yes	yes
rnaseqcount	FALSE+FALSE	TRUE			yes
rnaseqcount	TRUE+FALSE	FALSE		yes
rnaseqcount	FALSE+TRUE	FALSE		yes
rnaseqcount	FALSE+FALSE	FALSE
unlog_rnaseqcount	TRUE+FALSE	TRUE		yes
unlog_rnaseqcount	FALSE+TRUE	TRUE		yes
unlog_rnaseqcount	FALSE+FALSE	TRUE
unlog_rnaseqcount	TRUE+FALSE	FALSE		yes	no
unlog_rnaseqcount	FALSE+TRUE	FALSE		yes	no
unlog_rnaseqcount	FALSE+FALSE	FALSE			no
compositional				yes	yes
microarray				yes	yes

5.3.1 Necessary parameters for each template

“logarithm”: datasetName, log_base and prior.count
“scaling”: datasetName and scale_reduce
“norm_quantiles”: datasetName
“combat”: datasetName, datasetName_batches, varName_batches
“rnaseqcount”: datasetName, filter_type, log_base, filter_threshold, prior.count, has_filter, has_TMM, has_TMMwsp, has_log
“unlog_rnaseqcount”: datasetName, filter_type, log_base, unlog_base, filter_threshold, prior.count, unlig_prior.count, has_filter, has_TMM, has_TMMwsp, has_log
“unlog_metagenomics_2”: datasetName, filter_type, log_base, unlog_base, filter_threshold, prior.count, unlig_prior.count, has_filter, has_TMM, has_TMMwsp, has_log
“metagenomics_1”: datasetName, filter_type, normalization, offset, filter_threshold, has_filter
“metagenomics_2”: datasetName, filter_type, log_base filter_threshold, prior.count, has_filter, has_TMM, has_TMMwsp, has_log
“microarray”: datasetName, filter_threshold
“compositional”: datasetName, normalization

datasetName_colours and varName_colours are always optional (never required) and not used on the interface (maybe useful for future developments or to remove).

5.3.2 Templates and parameters

Here is another view on the templates and their parameters, giving parameter options and default values on top. Indications in purple refer to template names or parameter names.

Note that the template rnaseqcount is identical to template metagenomics_2.

Note that the template unlog_rnaseqcount is identical to template unlog_metagenomics_2.