Stage 2: Sequencing Error Modeling

Context-aware Noise Introduction

After generating a clean read and label vector, the simulator introduces base-level errors to approximate ONT/PacBio characteristics. Error operators include:

• mismatches (--mismatch-rate)
• insertions (--insertion-rate, bounded by --max-insertions)
• deletions (--deletion-rate)

Error rates are applied in a label-dependent manner. In the default implementation, polyA/polyT segments use a distinct error rate (--polyt-error-rate) that reflects homopolymer instability, while other segments use the provided global rates. This allows protocol-induced homopolymers to exhibit distinct noise properties without corrupting structured motifs beyond what is intended by the simulation.

Orientation Augmentation

Optionally, the reverse complement of each read can be generated (--rc), with labels reversed accordingly. This augmentation improves robustness to strand orientation and is particularly useful for datasets with mixed orientation or uncertain strand assignment.

Artifact Simulation

To increase robustness to real-world failure modes, a user-controlled fraction of reads (--invalid-fraction) are simulated as structurally invalid or artifactual. Supported artifact modes include read concatenation, repeated adapter blocks at the 5′ (or 3′) end, and truncations from either end. Optionally, strand inversions or reverse complements may be included during concatenation. Invalid reads in the simulated dataset reduces overconfident labeling of malformed inputs and improves downstream filtering and diagnostics.