LoginSign Up
← Back to all sparks
P

protr

ANALYTICS
Velocity0.0

Toolkit for generating various numerical features of protein sequences

github.com/nanxstats/protr ↗

protr's feature set is finished; the work now is surviving Bioconductor's churn.

proteomicssequence-descriptorsbioconductorfeature-paritymaintenance-mode
◆Current state
protr generates numerical descriptors from protein sequences for machine learning, plus alignment-based similarity between sequences. The descriptor functions have been stable for years. Recent releases divide cleanly into two kinds: extending the similarity computations to work under memory constraints, and absorbing the Bioconductor split that moved pairwise alignment out of Biostrings into pwalign.
◆Where it's heading
The similarity side is where the remaining engineering goes, and it follows a consistent pattern — whatever parSeqSim() gained, crossSetSim() eventually gets. Batching, verbose progress and a disk-backed variant all arrived for the single-set case first and were mirrored for the cross-set case in 1.7-1. That is a maintainer closing feature-parity gaps rather than opening new directions, and the two most recent releases contain no user-facing change at all.
◆Prediction
Expect the next release to react to another Bioconductor or R CMD check change, which accounts for three of the last four. The similarity functions now have parity, so there is no obvious internal backlog left.

◆Recent moves

  1. 11mo ago

    protr 1.7-5 silences a Debian r-devel check note

    A DESCRIPTION comment removed to clear an R CMD check note that appears only on Debian r-devel. No code, no behavior — the release exists to keep CRAN quiet.

    View source ↗
  2. 1y ago

    protr 1.7-4 checks alignment dependencies upfront

    Biostrings and pwalign availability is now checked before similarity calculations run, so a missing dependency raises a clear error instead of leaving error messages embedded in the results. Turning silently corrupted output into a loud failure is the right trade for a package feeding machine learning pipelines.

    View source ↗
  3. 1y ago

    protr 1.7-3 detects Biostrings version to find pwalign

    Bioconductor 3.19 moved pairwise alignment from Biostrings to pwalign, so protr now detects the installed Biostrings version at runtime and adapts. Handling the split dynamically rather than hard-pinning keeps the package working across the mixed R and Bioconductor versions real labs run.

    View source ↗
  4. 2y ago

    protr 1.7-2 fixes citation key and vignette accessibility

    A BibTeX citation key, typo fixes, and images moved to knitr::include_graphics() to resolve pkgdown alt-text hints. Documentation only, published within two minutes of 1.7-1 in this archive.

    View source ↗
  5. 2y ago

    protr 1.7-1 brings crossSetSim to parity with parSeqSim

    crossSetSim() gains the batches and verbose arguments parSeqSim() already had, and crossSetSimDisk() adds the disk-backed variant. Both address the same constraint: comparing large sequence sets on machines with more cores than RAM.

    View source ↗
  6. 2y ago

    protr 1.7-0 adds crossSetSim for two-set similarity

    A contributed function for alignment-based similarity between two distinct sets of sequences, rather than within one set. It opens the cross-set workflow that the next several releases spend their time making memory-efficient.

    View source ↗