Skip to contents

Changelog

Source: NEWS.md

bulkdeconv 0.1.0

First version.

  • bulkdeconv(data, expr, species, method, ...) runs one method on a gene-by-sample table (first column gene symbols, other columns samples), the layout rsemverse::rsem_read() returns.
  • 18 methods that need no registration: human (quantiseq, mcp_counter, xcell, estimate, consensus_tme, immucellai2, imsig, spatialdecon_safetme, epic, abis, timer), mouse (mmcp_counter, dcq, immucellai_mouse, ssmd, seqimmucc, base), both (xcell2). seqimmucc takes counts (LLSR only); the other 17 take TPM.
  • 15 methods are implemented in the package (verified against the output of the original code): quantiseq, mcp_counter, xcell, estimate, consensus_tme, immucellai2, imsig, spatialdecon_safetme, abis, timer, mmcp_counter, immucellai_mouse, seqimmucc, base and xcell2. xCell2 inference includes the nine pre-trained presets. EPIC, dcq and ssmd use their original packages.
  • immucellai2 takes 1.7 s instead of 18 s (the original reads a 41 MB text file on every call). estimate, immucellai2, spatialdecon_safetme and immucellai_mouse no longer need ggplot2, dplyr, GeomxTools, SeuratObject, GSVA or logNormReg.
  • base is about 3x faster than the original code (51 s vs 175 s for 6 samples at 100 permutations).
  • Licence: GPL-3.
  • Each method runs in a separate R process from a private library at a pinned version (bulkdeconv_install()).
  • Input validation, a registry (bulkdeconv_methods()), single-sample handling, and a standard result object.
  • Installing bulkdeconv is enough for the 15 methods implemented in the package: their reference data ship with it and the packages they use (quadprog, MASS, sva) are installed as dependencies. Only epic, dcq and ssmd need bulkdeconv_install().