The goal of glystats is to perform statistical analysis on glycoproteomics and glycomics data. It works seamlessly with the glyexp package.
We recommend installing the meta-package glycoverse, which includes this package and other core glycoverse packages.
If you don’t want to install all glycoverse packages, you can only install glystats.
You can install the latest release of glystats from r-universe (recommended):
# install.packages("pak")
pak::repo_add(glycoverse = "https://glycoverse.r-universe.dev")
pak::pkg_install("glystats")Or from GitHub:
pak::pkg_install("glycoverse/glystats@*release")Or install the development version (NOT recommended):
pak::pkg_install("glycoverse/glystats")Note: Tips and troubleshooting for the meta-package glycoverse are also applicable here: Installation of glycoverse.
Analysis uses the object’s row/column axes, not ordinary
rowData$variable or colData$sample columns. Non-empty unique axis
names remain result IDs. Unnamed, duplicated, or empty-name axes use
unique input-position IDs without changing the experiment. Every SE
result includes identifiers, mapping those IDs to input positions and
the original labels (also with add_info = FALSE). Metadata are
attached by position; colliding annotation columns are preserved with an
_metadata suffix rather than overwriting result fields.
For position-keyed variables, pass the same unmodified input to
filter_sig_vars(). Re-run analysis after reordering or changing it;
position IDs are not persistent IDs across different inputs. See
?gly_identity for the full contract and how to recover original
labels.
All expression-analysis functions default to log_transform = NULL.
Record the supplied abundance scale in S4Vectors::metadata(exp)$scale:
"raw", "log" (natural logarithm), "log2", "log10", or "logx".
Known logged values are used unchanged. Raw values and old objects
without scale metadata retain each method’s previous default; Cox, ROC,
and fold change continue to use unlogged values by default. Use TRUE
to request the method’s log transformation for raw values, or FALSE to
use supplied values unchanged. If the recorded scale is already log,
log2, log10, or logx, TRUE warns and is ignored: no second
logarithm is applied. Setting metadata never transforms the data.
# Data already transformed using log10:
S4Vectors::metadata(exp)$scale <- "log10"
pca <- gly_pca(exp)
pca$preprocessing
# This warns and still uses the supplied log10 values unchanged:
cox <- gly_cox(exp, log_transform = TRUE)PCA, PLS-DA, OPLS-DA, t-SNE, and UMAP retain log(x + 1); other methods
use log2(x + 1e-6) when enabled. Set construction and inference share
the same choice. Results retain input metadata in meta_data and record
input/analysis scales and the actual transformation in preprocessing.
Model estimates, standard errors, intervals, and raw fits stay in
analysis units.
For raw inputs, differential-test fold changes remain log2 ratios of
arithmetic means. For ordinary logged inputs, they are differences of
group means converted to log2 units (geometric mean ratios). Natural-log
differences are divided by log(2); base-10 differences are multiplied
by log2(10). Limma contrasts follow the same change of base; raw-scale
contrasts are returned as estimate with NA log2fc.
filter_sig_vars() uses significance-only filtering for raw-scale limma
results and rejects explicit fold-change cutoffs. For raw input,
gly_fold_change(log_transform = TRUE) uses newly logged values,
including the original pseudo-count. See ?gly_scale for the full
contract.
"logx" identifies transformed coordinates such as CLR, ALR, or VSN
output. All analyses except gly_fold_change() can use these
coordinates unchanged, including negative and zero values.
log_transform = TRUE warns and is ignored for logx, just as for
ordinary logged scales. The method’s statistical assumptions still need
to be appropriate for the chosen coordinates; the label does not make
these transformations equivalent.
Original-abundance fold changes cannot be inferred from logx.
gly_fold_change() therefore rejects it. Differential tests retain
native estimates and effect sizes but return NA in log2fc; limma
reports its native contrast in estimate. filter_sig_vars() uses
significance-only filtering for logx, even for glycoproteomics; a
non-NULL fold-change cutoff warns and is ignored. Explicit
fc_cutoff = NULL disables fold-change filtering on every scale.
Omitting the argument retains the default cutoff of 2 for
glycoproteomics comparisons with available fold changes. Use a separate
experiment on a known abundance scale when fold changes are needed;
relabeling logx coordinates does not recover abundances.
Older development objects labeled "log" must be relabeled if their
actual base was 2 or 10. The package never guesses the base. Result
extraction and significance-filtering utilities do not transform
expression.
glystats is the downstream analysis package in the glycoverse
ecosystem. It provides statistical analysis functions for
glyexp::GlycomicSE and glyexp::GlycoproteomicSE objects. A common
workflow is to use glyread to
import data, glyclean to
preprocess data, and then glystats to perform statistical analysis.
Say we already have a preprocessed glycomics or glycoproteomics
container called exp:
# Two-sample t-test
ttest_res <- gly_ttest(exp)
# PCA analysis
pca_res <- gly_pca(exp)
# ROC analysis
roc_res <- gly_roc(exp)That’s it! These functions use glycoverse column conventions to load
needed data and perform analysis. All functions start with gly_ to
leverage the auto-completion in RStudio. They accept a
glyexp::GlycomicSE or glyexp::GlycoproteomicSE object, and return
analysis result as a tibble or a list of tibbles. See documentation for
each function for more details.
SE result identity columns are .variable and .sample, while ordinary
metadata variable and sample remain annotation columns. Check
result$identifiers$columns to distinguish this output contract from
older results. Correlation endpoints use .variable1/.variable2 or
.sample1/.sample2. This is an output API change, including for fully
named unique inputs; plotting consumers require adaptation.