GPU-Portable Pairwise and Codon-Aware Multiple Sequence Alignment • genoaligner

Pairwise and multiple sequence alignment for R, from one portable C++17 core.

genoaligner provides edit-distance (Levenshtein / WFA-equivalent) and Smith-Waterman local pairwise alignment with score and CIGAR reconstruction, plus progressive multiple sequence alignment in DNA, protein and codon-aware modes — all from a self-contained C++17 core that needs no GPU toolchain, so it builds and runs on CRAN and anywhere R runs.

# pairwise, vectorised over pair columns
align_edit(c("AAAACCC", "ACGT"), c("AAAATCC", "ACCT"), smax = 8)
align_sw(c("ACGTACGT", "GGGG"), "ACGTTCGT", scoring = c(2, -3, 5, 2))

# multiple sequence alignment: sequences in -> alignment + QC out
msa_align(c("ACGTACGTAA", "ACGTTCGT", "ACGTACG"))
msa_codon(lepus_cytb$sequence, names = lepus_cytb$accession, gc = 2)

Install

From CRAN:

install.packages("genoaligner")

From GitHub (development head):

remotes::install_github("alrobles/genoaligner-r")

What is this?

A pairwise aligner and a multiple aligner over the same core:

align_edit() — edit distance with CIGAR, bounded by smax. Pairs whose true distance exceeds smax are reported unresolved (never silently wrong).
align_sw() — local Smith-Waterman (affine gaps) with CIGAR and the aligned span’s coordinates.
msa_align() — progressive profile–profile MSA with a deterministic neighbor-joining guide tree, in dna or protein mode. Optional iter_refine rounds realign every guide-tree bipartition, and fft_band restricts each realignment to a band around Fourier-detected homology anchors (MAFFT FFT-NS-i class).
msa_codon() — codon-aware MSA (MACSE-class): whole-codon indels preserve the reading frame by construction; refine = TRUE runs a stage-2 frameshift pass for sequences that truly carry frameshifts. NCBI genetic codes 1 (standard) and 2 (vertebrate mitochondrial — CYTB/COI/ND).

msa_* return a genoaligner_msa object carrying the aligned matrix plus per-sequence QC (reading frame, stop codons, partial codons) — the QC the library measured travels inside the result. as_dnabin() coerces to ape::DNAbin.

v2 breaking change: align() was renamed align_edit() — the generic name collides with Biostrings/IRanges exports in mixed sessions. align() remains as a deprecated alias returning identical results.

The wavefront/GPU backend (ROCm/CUDA) that computes the same results with acceleration lives in the sibling C++ repository.

Documentation

See the site: function reference, the intro vignette, and the codon-aware MSA case study aligning 37 real hare cytochrome-b records.