BABAPPAlign v1.2.0

Major Update: Native Codon Alignment Mode

BABAPPAlign v1.2.0 introduces a fully integrated codon alignment mode.

The alignment core (affine-gap Gotoh DP with explicit traceback matrices) remains unchanged. This release extends functionality without modifying the alignment engine.

New Features

Codon Mode

Run:

babappalign cds.fasta --model babappascore --mode codon

The program will:

1. Validate CDS sequences

   • Length divisible by 3
   • No internal stop codons
   • Valid nucleotide alphabet

2. Translate CDS → protein

3. Perform embedding-based progressive protein alignment

4. Back-map to codon alignment (PAL2NAL-style logic)

Outputs:

cds.protein.aln.fasta   (intermediate protein alignment)
cds.codon.aln.fasta     (final codon alignment)

No external PAL2NAL dependency is required.

Gap penalties are automatically scaled in codon mode to maintain biological consistency.

CLI Improvements

The -o/--output option has been removed.

Output filenames are automatically generated:

Protein mode: input.protein.aln.fasta

Codon mode: input.protein.aln.fasta input.codon.aln.fasta

This simplifies usage and prevents accidental overwriting.

Alignment Core Integrity

The alignment engine is unchanged:

• Three-state affine-gap DP (M, Ix, Iy)
• Explicit traceback matrices
• Exact dynamic programming
• Neural scoring performed outside DP recursion

Scientific reproducibility from previous versions is preserved.

Model Loading Improvements

• HuggingFace warnings suppressed
• No pooler initialization warnings
• Explicit model resolution maintained
• SHA-256 checksum printed at runtime

Validation

Codon mode validated against PAL2NAL for:

• Frame preservation
• Triplet gap integrity
• Alignment length consistency
• Per-site identity comparison

Installation

pip install --upgrade babappalign

Requirements

• External trained scoring model (BABAPPAScore)
• Model distributed via Zenodo

Concept DOI: https://doi.org/10.5281/zenodo.18053200

Summary

v1.2.0 adds native codon alignment while preserving the exact alignment core. The tool now supports both protein and CDS workflows in a single, reproducible framework.