OligoMiner2

Genome-scale oligonucleotide probe design for DNA and RNA FISH.

OligoMiner2 is a Python package for designing oligonucleotide probes used in fluorescence in situ hybridization (FISH) experiments. It takes a target genome or transcriptome as input and returns candidate probe sequences that are thermodynamically optimized and filtered for specificity.

The pipeline

1. Mine candidate probes from FASTA sequences, filtering by melting temperature, GC content, length, homopolymer runs, sequence entropy, soft-masked repeats, and prohibited subsequences.
2. Align candidates against a reference genome with Bowtie2 and reconstruct each alignment's duplex from the reference.
3. Predict a binding probability for every reconstructed duplex, screening with a model and verifying the ones that could change a decision with exact NUPACK pDup. K-mer frequency analysis runs as a parallel track.

Targets

Probes can be mined against a genome, a transcriptome built from a genome and its annotation, or individual transcripts. Transcript-aware modes cover exon-exon junctions, regions that discriminate between isoforms, and introns.

Probe architectures

Beyond conventional single-oligo probes, the package designs padlock probes with split homology arms, and split architectures where one functional element spans two oligos binding adjacent sites -- HCR 3.0 split-initiator pairs and split-FISH bridge pairs. Both members of a pair are tracked as one targeting unit, so filtering cannot leave a half-probe in an order.

Synthetic domains -- barcodes, primer sites, amplifier initiators -- are appended through a domain layout, which places a sequence at a named slot rather than at an end.

Models

Duplex stability is predicted by one of four registered models, selected by name: a condition-aware physics-feature gradient-boosted model, a duplex BiLSTM, the PaintSHOP model, and the OligoMiner v1 classifier. The first two can be retrained on new data through the package.

Scale

K-mer counting dispatches between jellyfish and a pure-numpy backend. Alignment and duplex construction stream rather than loading a genome's alignments into memory, and every entry point resolves its core count from the scheduler's grant rather than from the machine size.

Quick start

from oligominer import mine_fasta

# mine candidate probes with default parameters
df = mine_fasta('genome.fa')

Documentation

Full documentation including API reference and example notebooks is available at the OligoMiner2 docs site.

License

OligoMiner2 is open-source software. See LICENSE for details.