Package 'aIc' reference manual

Title:	Testing for Compositional Pathologies in Datasets
Description:	A set of tests for compositional pathologies. Tests for coherence of correlations with aIc.coherent() as suggested by (Erb et al. (2020) <doi:10.1016/j.acags.2020.100026>), compositional dominance of distance with aIc.dominant(), compositional perturbation invariance with aIc.perturb() as suggested by (Aitchison (1992) <doi:10.1007/BF00891269>) and singularity of the covariation matrix with aIc.singular(). Currently tests five data transformations: prop, clr, TMM, TMMwsp, and RLE from the R packages 'ALDEx2', 'edgeR' and 'DESeq2' (Fernandes et al (2014) <doi:10.1186/2049-2618-2-15>, Anders et al. (2013)<doi:10.1038/nprot.2013.099>).
Authors:	Greg Gloor
Maintainer:	Greg Gloor <[email protected]>
License:	GPL (>= 3)
Version:	1.0
Built:	2024-11-13 05:00:27 UTC
Source:	https://github.com/ggloor/aic

Calculate the subcompositional coherence of samples in a dataset for a given correction.

Description

'aIc.coherent' compares the correlation coefficients of features in common of the full dataset and a subset of the dataset. This is expected to be false for all compositional datasets and transforms.

Usage

aIc.coherent(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  log = FALSE,
  group = NULL,
  cor.test = "spearman"
)
aIc.coherent(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  log = FALSE,
  group = NULL,
  cor.test = "spearman"
)

Arguments

`data`	can be any dataframe or matrix with samples by column
`norm.method`	can be prop, clr, RLE, TMM, TMMwsp, lvha, iqlr
`zero.remove`	is a value. Filter data to remove features that are 0 across at least that proportion of samples: default 0.95
`zero.method`	can be any of NULL, prior, GBM or CZM. NULL will not impute or change 0 values, GBM and CZM are from the zCompositions R package, and prior will simply add 0.5 to all counts.
`log`	is a logical. log transform the prop, RLE or TMM outputs, default=FALSE
`group`	is a vector containing group information. Required for clr, RLE,
`cor.test`	is either the pearson or spearman method (default)

Value

Returns a list with the correlation in cor, a yes/no binary decision in is.coherent, the x and y values for a scatterplot of the correlations in the full and subcompositions, and the plot and axis labels in main xlab and ylab.

Author(s)

Greg Gloor

Examples

data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.coherent(selex, group=group, norm.method='clr', zero.method='prior')
plot(x$plot[,1], x$plot[,2], main=x$main, ylab=x$ylab, xlab=x$xlab)

data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.coherent(selex, group=group, norm.method='clr', zero.method='prior')
plot(x$plot[,1], x$plot[,2], main=x$main, ylab=x$ylab, xlab=x$xlab)

`aIc.dominant` calculates the subcompositional dominance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a subset of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Description

aIc.dominant calculates the subcompositional dominance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a subset of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Usage

aIc.dominant(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  log = FALSE,
  distance = "euclidian",
  group = NULL
)
aIc.dominant(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  log = FALSE,
  distance = "euclidian",
  group = NULL
)

Arguments

`data`	can be any dataframe or matrix with samples by column
`norm.method`	can be prop, clr, RLE, TMM, TMMwsp
`zero.remove`	is a value. Filter data to remove features that are 0 across at least that proportion of samples: default 0.95
`zero.method`	can be any of NULL, prior, GBM or CZM. NULL will not impute or change 0 values, GBM (preferred) and CZM are from the zCompositions R package, and prior will simply add 0.5 to all counts.
`log`	is a logical. log transform the RLE or TMM outputs, default=FALSE
`distance`	can be euclidian, bray, or jaccard. euclidian on log-ratio transformed data is the same as the Aitchison distance. default=euclidian
`group`	is a vector containing group information. Required for clr, RLE, TMM, lvha, and iqlr based normalizations.

Value

Returns a list with the overlap between distances in the full and subcompositon in ol (expect 0), a yes/no binary decision in is.dominant and the table of distances for the whole and subcomposition in dist.all and dist.sub, a plot showing a histogram of the resulting overlap in distances in plot, and the plot and axis labels in main xlab and ylab

Author(s)

Greg Gloor

Examples

data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.dominant(selex, group=group, norm.method='clr', distance='euclidian', zero.method='prior')
plot(x$plot, main=x$main, ylab=x$ylab, xlab=x$xlab)
data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.dominant(selex, group=group, norm.method='clr', distance='euclidian', zero.method='prior')
plot(x$plot, main=x$main, ylab=x$ylab, xlab=x$xlab)

`aIc.perturb` calculates the perturbation invariance of distance for samples with a given correction. This compares the distances of samples of the full dataset and a the perturbed dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Description

aIc.perturb calculates the perturbation invariance of distance for samples with a given correction. This compares the distances of samples of the full dataset and a the perturbed dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Usage

aIc.perturb(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  distance = "euclidian",
  log = FALSE,
  group = NULL
)
aIc.perturb(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  distance = "euclidian",
  log = FALSE,
  group = NULL
)

Arguments

`data`	can be any dataframe or matrix with samples by column
`norm.method`	can be prop, clr, RLE, TMM, TMMwsp
`zero.remove`	is a value. Filter data to remove features that are 0 across at least that proportion of samples: default 0.95
`zero.method`	can be any of NULL, prior, GBM or CZM. NULL will not impute or change 0 values, GBM (preferred) and CZM are from the zCompositions R package, and prior will simply add 0.5 to all counts.
`distance`	can be euclidian, bray, or jaccard. euclidian on log-ratio transformed data is the same as the Aitchison distance. default=euclidian
`log`	is a logical. log transform the RLE or TMM outputs, default=FALSE
`group`	is a vector containing group information. Required for clr, RLE, TMM, lvha, and iqlr based normalizations.

Value

Returns a list with the maximum proportional perturbation in ol (expect 0, but values up to 1 is.perturb, the table of distances for the whole and perturbaton in dist.all and dist.perturb, the histogram of the perturbations in plot, and the plot and axis labels in main xlab and ylab. .

Author(s)

Greg Gloor

Examples

data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.perturb(selex, group=group, norm.method='clr', distance='euclidian', zero.method='prior')
plot(x$plot, main=x$main, ylab=x$ylab, xlab=x$xlab)
data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.perturb(selex, group=group, norm.method='clr', distance='euclidian', zero.method='prior')
plot(x$plot, main=x$main, ylab=x$ylab, xlab=x$xlab)

`aIc.plot` plots the result of the distance tests.

Description

aIc.plot plots the result of the distance tests.

Usage

aIc.plot(test.out)
aIc.plot(test.out)

Arguments

test.out

is the output from either aIc.dominant, aIc.scale, aIc.perturb

Value

returns a plot of the density of the distance test results. test result.

Author(s)

Greg Gloor

Examples

data(selex)
group = c(rep('N', 7), rep('S', 7))
test.out <- aIc.dominant(selex, norm.method='prop', group=group)
aIc.plot(test.out)
data(selex)
group = c(rep('N', 7), rep('S', 7))
test.out <- aIc.dominant(selex, norm.method='prop', group=group)
aIc.plot(test.out)

`aIc.runExample` loads the associated shiny app This will load the selex example dataset with the default group sizes, the user can upload their own local dataset and adjust groups accordingly.

Description

aIc.runExample loads the associated shiny app This will load the selex example dataset with the default group sizes, the user can upload their own local dataset and adjust groups accordingly.

Usage

aIc.runExample()
aIc.runExample()

Value

No return value, but instead opens a shiny connection to your default web browser with the selex dataset as an example.

Author(s)

Greg Gloor

Examples


library(aIc)
aIc.runExample()

library(aIc)
aIc.runExample()

`aIc.scale` calculates the scaling invariance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a scaled version of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Description

aIc.scale calculates the scaling invariance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a scaled version of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Usage

aIc.scale(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  distance = "euclidian",
  log = FALSE,
  group = NULL
)
aIc.scale(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  distance = "euclidian",
  log = FALSE,
  group = NULL
)

Arguments

`data`	can be any dataframe or matrix with samples by column
`norm.method`	can be prop, clr, iqlr, lvha, RLE, TMM, TMMwsp
`zero.remove`	is a value. Filter data to remove features that are 0 across at least that proportion of samples: default 0.95
`zero.method`	can be any of NULL, prior, GBM or CZM. NULL will not impute or change 0 values, GBM (preferred) and CZM are from the zCompositions R package, and prior will simply add 0.5 to all counts.
`distance`	can be euclidian, bray, or jaccard. euclidian on log-ratio transformed data is the same as the Aitchison distance. default=euclidian
`log`	is a logical. log transform the RLE or TMM outputs, default=FALSE
`group`	is a vector containing group information. Required for clr, RLE, TMM, lvha, and iqlr based normalizations.

Value

Returns a list with the overlap between distances in the full and scaled composition in ol (expect 0), a yes/no binary decision in is.scale and the table of distances for the whole and scaled composition in dist.all and dist.scale, a plot showing a histogram of the resulting overlap in distances in plot, and the plot and axis labels in main xlab and ylab

Author(s)

Greg Gloor

Examples

data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.scale(selex, group=group, norm.method='clr', zero.method='prior')
plot(x$plot, main=x$main, ylab=x$ylab, xlab=x$xlab)
data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.scale(selex, group=group, norm.method='clr', zero.method='prior')
plot(x$plot, main=x$main, ylab=x$ylab, xlab=x$xlab)

`aIc.singular` tests for singular data. This is expected to be true if the transform is behaving rationally in compositional datasets and also true in the case of datasets with more features than samples.

Description

aIc.singular tests for singular data. This is expected to be true if the transform is behaving rationally in compositional datasets and also true in the case of datasets with more features than samples.

Usage

aIc.singular(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  log = FALSE,
  group = NULL
)
aIc.singular(
  data,
  norm.method = "prop",
  zero.remove = 0.95,
  zero.method = "prior",
  log = FALSE,
  group = NULL
)

Arguments

`data`	can be any dataframe or matrix with samples by column
`norm.method`	can be prop, clr, RLE, TMM, TMMwsp
`zero.remove`	is a value. Filter data to remove features that are 0 across at least that proportion of samples: default 0.95
`zero.method`	can be any of NULL, prior, GBM or CZM. NULL will not impute or change 0 values, GBM (preferred) and CZM are from the zCompositions R package, and prior will simply add 0.5 to all counts.
`log`	is a logical. log transform the RLE or TMM outputs, default=FALSE
`group`	is a vector containing group information. Required for clr, RLE, TMM, lvha, and iqlr based normalizations.

Value

Returns a list with a yes/no binary decision in is.singular and the covariance matrix in cov.matrix

Author(s)

Greg Gloor

Examples

data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.singular(selex, group=group, norm.method='clr', zero.method='prior')
data(selex)
group = c(rep('N', 7), rep('S', 7))
x <- aIc.singular(selex, group=group, norm.method='clr', zero.method='prior')

16S rRNA tag-sequencing data

Description

A count table of a 16S rRNA amplicon data Two groups, pupils and centenarians are represented with 198 and 161 samples per group respectively. samples are by column and OTU ids are by row.

Usage

data(meta16S)
data(meta16S)

Format

A data frame with 359 columns and 860 rows

Source

doi: 10.1128/mSphere.00327-17

meta-transcriptome data

Description

A count table of a mixed population or metatranscriptome experiment. Two groups, H and BV are represented with 7 and 10 samples per group respectively. samples are by column and functions are by row.

Usage

data(metaTscome)
data(metaTscome)

Format

A data frame with 17 columns and 3647 rows

Source

doi:10.1007/978-3-030-71175-7_17 and doi:10.1007/978-1-4939-8728-3_13

Selection-based differential sequence variant abundance dataset

Description

This data set gives the differential abundance of 1600 enzyme variants grown under selective (NS) and selective (S) conditions

Usage

data(selex)
data(selex)

Format

A data frame with 14 columns and 1600 rows

Source

DOI:10.1073/pnas.1322352111

single cell transcriptome data

Description

A count table of a single cell transcriptome data subset from the count table from doi:10.1038/s41592-019-0372-4. Two groups memory T cells, and cytotoxic T cells, 1000 cells per group. samples are by column and genes are by row.

Usage

data(singleCell)
data(singleCell)

Format

A data frame with 2000 columns and 1508 rows

Source

https://www.nature.com/articles/s41592-019-0372-4

Saccharomyces cerevisiae transcriptome

Description

A count table of a highly replicated RNA-seq experiment with samples by column and genes by row. Two groups composed of SNF2 knockout and WT, 48 samples in each.

Usage

data(transcriptome)
data(transcriptome)

Format

A data frame with 96 columns and 5892 rows

Source

DOI: 10.1261/rna.053959.115 and PRJEB5348

Package 'aIc'

Help Index

Calculate the subcompositional coherence of samples in a dataset for a given correction.

Description

Usage

Arguments

Value

Author(s)

Examples

aIc.dominant calculates the subcompositional dominance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a subset of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Description

Usage

Arguments

Value

Author(s)

Examples

aIc.perturb calculates the perturbation invariance of distance for samples with a given correction. This compares the distances of samples of the full dataset and a the perturbed dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Description

Usage

Arguments

Value

Author(s)

Examples

aIc.plot plots the result of the distance tests.

Description

Usage

Arguments

Value

Author(s)

Examples

aIc.runExample loads the associated shiny app This will load the selex example dataset with the default group sizes, the user can upload their own local dataset and adjust groups accordingly.

Description

Usage

Value

Author(s)

Examples

aIc.scale calculates the scaling invariance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a scaled version of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

Description

Usage

Arguments

Value

Author(s)

Examples

aIc.singular tests for singular data. This is expected to be true if the transform is behaving rationally in compositional datasets and also true in the case of datasets with more features than samples.

Description

Usage

Arguments

Value

Author(s)

Examples

16S rRNA tag-sequencing data

Description

Usage

Format

Source

meta-transcriptome data

Description

Usage

Format

Source

Selection-based differential sequence variant abundance dataset

Description

Usage

Format

Source

single cell transcriptome data

Description

Usage

Format

Source

Saccharomyces cerevisiae transcriptome

Description

Usage

Format

Source

`aIc.dominant` calculates the subcompositional dominance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a subset of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

`aIc.perturb` calculates the perturbation invariance of distance for samples with a given correction. This compares the distances of samples of the full dataset and a the perturbed dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

`aIc.plot` plots the result of the distance tests.

`aIc.runExample` loads the associated shiny app This will load the selex example dataset with the default group sizes, the user can upload their own local dataset and adjust groups accordingly.

`aIc.scale` calculates the scaling invariance of a sample in a dataset for a given correction. This compares the distances of samples of the full dataset and a scaled version of the dataset. This is expected to be true if the transform is behaving rationally in compositional datasets.

`aIc.singular` tests for singular data. This is expected to be true if the transform is behaving rationally in compositional datasets and also true in the case of datasets with more features than samples.