Sequencing artifacts, dialed in

Real reads are noisy. 5'/3' loss eats segments. Indels shift frames. Base-callers fail. GenAIRR can dial each one independently — and the truth fields stay correct.

The intuition

A simulator that only models clean sequences is benchmarking your aligner's best case. The interesting case — the one your benchmark actually needs to cover — is when the molecule is half-eaten and the base-caller is guessing. Dial in each artifact and see how each one bites.

Live corruption

The top line is the clean post-mutation sequence. The bottom line is what your aligner would actually receive. Move the sliders.

Even with all artifacts cranked, the V/D/J coloring tells you exactly which base belongs to which segment. That mapping is the truth GenAIRR preserves through every transformation.

The corresponding code

Each artifact is its own pass on the pipeline. length= ranges feed 5'/3' loss; count= ranges feed indels and N-bases. Drop a pass to skip that artifact entirely.

import GenAIRR as ga

result = (
    ga.Experiment.on("human_igh")
       .recombine()
       .mutate(model="s5f", count=(5, 25))
       .corrupt_5prime_loss(length=(0, 20))
       .corrupt_3prime_loss(length=(0, 20))
       .corrupt_indels(count=(0, 2), insertion_prob=0.5)
       .corrupt_ns(count=(0, 5))
       .run_records(n=1000, seed=42, expose_provenance=True)
)