LoginSign Up
← Back to all sparks
G

glyclean

ANALYTICS
Velocity0.0

Perform Preprocessing on Glycomics and Glycoproteomics Data

github.com/glycoverse/glyclean ↗

glyclean stopped trusting QC samples to choose its preprocessing strategy.

glycomicspreprocessingimputationnormalizationbioconductor
◆Current state
glyclean handles preprocessing and QC for glycomics and glycoproteomics data: filtering, imputation, normalization, batch correction, and compositional transforms. The defining change in this window is 0.14.0, which abandoned QC coefficient-of-variation heuristics for choosing imputation and normalization methods in favor of rules keyed to sample size. The 0.15.x releases then finished removing the deprecated QC arguments and moved the whole package onto glyexp's SummarizedExperiment containers.
◆Where it's heading
Two commitments are visible. First, defaults should be defensible rather than adaptive: the maintainer explicitly judged CV-in-QC-samples not robust and replaced it with sample-size thresholds. Second, the package wants containers, not matrices, and 0.15.0 makes bare matrix inputs an error. Dependency pruning runs alongside both, with imputeLCMD reimplemented away so auto_clean() works out of the box.
◆Prediction
The compositional data thread is the least finished part of the package, so further CoDA methods or a broader auto_coda() are the likeliest next additions.

◆Recent moves

  1. 1mo ago

    Docs recommend the new SE containers

    Documentation and vignettes move to GlycomicSE and GlycoproteomicSE with SummarizedExperiment accessors, closing glyclean's part of the Stage II migration.

    View source ↗
  2. 1mo ago

    Preprocessing behaves the same across both containers

    Preprocessing and QC functions produce consistent behavior on legacy experiment() and current SummarizedExperiment inputs, so a half-migrated pipeline does not silently preprocess differently at different stages.

    View source ↗
  3. 1mo ago

    Matrix inputs rejected; containers now required

    The migration release, and a firm one: normalization, imputation, filtering, CoDA, and batch correction now require a glyexp container or plain SummarizedExperiment, and matrix inputs error outright. The long-deprecated qc_name and to_try arguments are removed, and aggregate() gets substantially faster on large glycoproteomics experiments.

    View source ↗
  4. 3mo ago

    auto_clean() works without extra package installs

    impute_min_prob() is reimplemented to drop the imputeLCMD dependency, making auto_clean() usable out of the box. Small in code, but it removes the install friction from the function most new users reach for first.

    View source ↗
  5. 3mo ago

    Imputation strategy now keyed to sample size, not QC

    auto_impute() picks its method from sample size and data type rather than QC sample CVs, and auto_normalize() and auto_remove() stop consulting QC samples entirely. The maintainer states plainly that the old CV-based heuristic was not robust, which makes this the reasoning that the next three releases carry out.

    View source ↗
  6. 4mo ago

    CoDA transforms aligned with published methods

    transform_clr() and transform_alr() are brought in line with a published method, auto_coda() adds the glycowork compositional strategy for glycomics, and auto_normalize() narrows to total abundance normalization for glycomics data. Compositional handling is the newest analytical thread in the package.

    View source ↗