LoginSign Up
← Back to all sparks
G

glysmith

ANALYTICS
Velocity0.0

Full-Featured Analysis Pipeline for Glycomics and Glycoproteomics

github.com/glycoverse/glysmith ↗

glysmith turns an LLM interview into a reproducible glycomics analysis pipeline.

glycomicspipeline orchestrationllm interfacesbioconductorr packages
◆Current state
glysmith is the orchestration layer of glycoverse: inquire_blueprint() interviews the user, and forge_analysis() runs the resulting step blueprint end to end. The recent 0.12.x releases wired that pipeline onto glyexp's new SummarizedExperiment containers, and 0.12.0 pulled QC sample handling out of step_preprocess() entirely. The larger feature release was 0.11.0, which added structure inference and three more enrichment steps to the step catalogue.
◆Where it's heading
The package is growing along two axes: more steps in the catalogue, and less ceremony around getting to them. QC handling and preprocessing arguments keep being removed rather than added, while the AI layer widens to more providers instead of getting more elaborate. glysmith is also the package that absorbs the rest of the stack's reorganizations, routing enrichment to glyfun after glystats deprecated it and pulling structure inference from glyanno.
◆Prediction
Expect the step catalogue to keep tracking sibling packages, with new steps appearing shortly after the packages behind them ship the underlying capability.

◆Recent moves

  1. 1mo ago

    Blueprint functions accept the new SE containers

    inquire_blueprint() and modify_blueprint() accept GlycomicSE and GlycoproteomicSE, and the documentation now recommends them. This closes glysmith's part of the Stage II migration while keeping legacy inputs working.

    View source ↗
  2. 1mo ago

    forge_analysis() runs natively on SummarizedExperiment

    The pipeline runner and its individual steps gain native support for GlycomicSE and GlycoproteomicSE across supported glyexp versions. Since forge_analysis() executes every step in a blueprint, this is the change that actually makes migrated data usable end to end.

    View source ↗
  3. 2mo ago

    QC sample handling removed from preprocessing

    step_preprocess() drops the qc_name argument and no longer removes QC samples after preprocessing, and the blueprint interview stops asking about QC names. It mirrors the same retreat from QC-driven heuristics that glyclean made in 0.14.0.

    View source ↗
  4. 3mo ago

    Structure inference and three new enrichment steps

    step_infer_structure() brings glyanno's composition-to-structure inference into the pipeline, and step_sig_enrich_ncg(), step_sig_enrich_wp(), and step_sig_enrich_do() add three more enrichment analyses. The release also reroutes enrichment to glyfun and widens AI report generation past a single provider.

    View source ↗
  5. 5mo ago

    step_preprocess() matches its documented QC behavior

    A one-line fix making step_preprocess() remove QC samples as documented. Superseded three months later by 0.12.0, which removed the behavior instead.

    View source ↗
  6. 5mo ago

    Motif quantification splits into dynamic and branch steps

    step_quantify_motifs() is removed and replaced by step_quantify_dynamic_motifs() and step_quantify_branch_motifs(), with distinct downstream result types and separate differential analysis outputs. The split follows glymotif's own separation of the two motif families.

    View source ↗