Skip to content

Single-Pass Bacterial & Prophage ORF Benchmark

This presentation evaluates OrfPredictor on the complete Escherichia coli str. K-12 substr. MG1655 gold-standard reference chromosome (GenBank: CP014225.1, 4,656,178 bp) and its 9 structurally embedded cryptic and lambdoid prophage islands (CP4-44, CPS-53, CPZ-55, CP4-57, CP4-6, DLP12, e14, Rac, and Qin).

Horizontally acquired prophages frequently encode seroconverting polysaccharide-modification cassettes (such as gtrABC glucosyltransferases and oac O-acetyltransferases) characterized by sharp local AT-enrichment (34%–44% GC vs. 50.8% host GC), short (90–120 bp) accessory ORFs, and tightly coupled 1-bp (TGATG/TAATG) or 4-bp (ATGA) stop-start operon overlaps that standard single-genome bacterial ORF callers miss.

Key Takeaways

  • 29-State Frameshift & Operon-Overlap Profile-HMM: OrfPredictor decodes forward and reverse coding frames, Shine-Dalgarno RBS motifs, 1-bp/2-bp sequencing or pseudogene frameshift indels (IS_FRAMESHIFT), and direct 1-bp (TGATG/TAATG) and 4-bp (ATGA) stop-start operon overlaps (E → M1 / Er → M1r) in a single Numba-compiled Viterbi pass.
  • Sliding 4-kb Prefix-Sum Local GC Adaptation: Rather than relying on a single chromosome-wide hexamer model, OrfPredictor computes a prefix-sum 4-kb local GC bin (0..43) at every nucleotide so codon emission log-odds adapt seamlessly across AT-rich prophage islands and polysaccharide synthesis operons.
  • Prophage Recovery With Zero Two-Pass Overhead: Enabling OrfPredictor(include_phage=True) boosts prophage 3'-stop recall to 95.6% (+7.4% over bacterial-only baseline, rescuing +17 prophage ORFs and +25 exact 5'+3' boundaries) while simultaneously improving host chromosome recall (98.7% → 99.1%) at >60 Mb/s (~71 ms for the 4.66 Mb chromosome).

Run Locally

# Present as interactive slideshow
uv run --with marimo --with altair marimo run scripts/benchmark_prophage_orfs.py

# Edit notebook cells
uv run --with marimo --with altair marimo edit scripts/benchmark_prophage_orfs.py

Comments