Let D represent any number of features measured across N biological replicates exposed to a binary or continuous event E. For example, E could indicate case-control status, treatment status, treatment dose, or time. This function converts a "count matrix" with N rows and D columns into a proportionality matrix of D rows and D columns.
For phi, the result of phit, one can think of the resultant matrix as
analogous to a distance matrix, except that it has no symmetry unless forced.
For rho, the result of perb, one can think of the resultant matrix as
analogous to a correlation matrix.
For phs, the result of phis, one can think of the resultant matrix as
either a naturally symmetric variant of phi or a monotonic variant of rho.
Another function, corr, calculates Pearsons' correlation using
log-ratio transformed data.
These methods all use the centered log-ratio transformation by default,
but will use an additive log-ratio transformation instead if a scalar
ivar is provided. When using an additive log-ratio transformation,
this function will return rho = 0 for the column and row in the
@matrix slot that would contain the reference feature.
Setting ivar to a numeric or character vector will transform
data using the geometric mean of only the indexed features.
Alternatively, setting ivar to "iqlr" will transform data using
the geometric mean of only the features with variances that fall in
the inter-quartile range of all per-feature variances. We base this
"iqlr" transformation on the ALDEx2 package.
Log-ratio transformation, by its nature, fails if the input data contain any zero values. To avoid an error in this case, these methods automatically replace all zero values with 1. However, the topic of zero replacement is controversial. Proceed carefully when analyzing data that contain any zero values.
The select argument subsets the feature matrix
after log-ratio transformation but before calculating
proportionality. This reduces the run-time and RAM
overhead without impacting the final result. Removing
lowly abundant features prior to log-ratio transformation
could otherwise change the proportionality measure.
# S4 method for propr
initialize(.Object, counts, ivar, select)phit(counts, ivar = 0, select, symmetrize = TRUE)
perb(counts, ivar = 0, select)
phis(counts, ivar = 0, select)
corr(counts, ivar = 0, select)
Missing. Ignore. Leftover from the generic method definition.
A data.frame or matrix. A "count matrix" with subjects as rows and features as columns.
A numeric scalar. Specifies reference feature(s) for additive log-ratio transformation. The argument will also accept feature name(s) instead of the index position(s). Set to "iqlr" to use inter-quartile log-ratio transformation. Ignore to use centered log-ratio transformation.
Subsets via object@counts[, select].
Optional. Use this argument to subset the proportionality
matrix before building without altering the final result.
A logical. If TRUE, forces symmetry
by reflecting the "lower left triangle".
Returns a propr object.
# NOT RUN {
library(propr)
data(mail)
phi <- phit(mail)
rho <- perb(mail)
phs <- phis(mail)
# }
Run the code above in your browser using DataLab