In-Situ Probes

1. 01Tile any open sequence into candidate oligos under a length, melting-temperature, G+C and composition specification — OligoMiner's own crawler, including its incremental thermodynamics
2. 02Screen those candidates against an alignment for off-target binding, using the linear discriminant OligoMiner trained at six hybridization temperatures
3. 03Three screening decisions, not one: the trained model, unique-alignment-only, and aligns-nowhere for a sequence that should be absent
4. 04Gather the survivors into the probe SET that images best — centred on the molecule, compact, and evenly spread (iFISH's three features)
5. 05Every stage reports what it removed and why: how many windows failed on melting temperature, on G+C, on a prohibited run, and how many oligos the specificity screen dropped
6. 06Formamide and sodium are controls, not constants, and they move every melting temperature in the design
7. 07A coverage track along the target showing each oligo and every stretch left uncovered, with the gaps measured rather than eyeballed
8. 08Melting-temperature and G+C distributions drawn against the window you asked for, so a set clustered at one end reads as clustered
9. 09Click any oligo on the track to select those bases on the sequence map
10. 10Export the surviving oligos as FASTA or BED
11. 11Prohibited subsequences are editable; homopolymer runs are banned by default because they hybridize promiscuously

packages/ports/oligo/probe merges three published tools, each ported faithfully and each graded against its own upstream: OligoMiner blockParse for tiling and outputClean for the trained specificity model (both MIT, Beliveau lab), and iFISH ifpd.query for probe-set selection (MIT, Bienko-Crosetto lab). The goldens regenerate upstream's own committed example output byte-for-byte before grading anything, and cover all 94 RefSeq CDS of yeast chromosome I as well as upstream's human example window. 256 tests. The panel runs on @ports/probe, @ports/workflow.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

Why is tiling not enough on its own?

Because an oligo that satisfies every thermodynamic criterion can still bind somewhere else in the genome, and then it images the wrong thing. In OligoMiner's own published example 13 windows pass the specification and only 12 survive specificity screening. Tiling and specificity are two different questions, and a probe graded only by the first looks finished and is not.

What does the specificity model actually use?

Three numbers per oligo: its length, the alignment score of its best off-target hit, and its percent G+C. A linear discriminant trained at your hybridization temperature turns those into the probability that the off-target site is thermodynamically real. It is not a rule of thumb — it is the model published with the method.

Why only six hybridization temperatures?

Because those are the six the model was trained at, and its coefficients change sign across the range: the length term is negative at 32 °C and positive at 57 °C. A temperature between two of them is a different model rather than a blend of two, so an untrained value is refused instead of interpolated.

Do I need an alignment?

Not to tile, but without one the oligos have not been checked for off-target binding, and the reading says so in those words rather than presenting them as finished probes. Align the candidates against your genome and paste the SAM.

What makes one probe set better than another?

Three things, and you choose which one defines the band and which one ranks inside it. Centrality is how close the set sits to the middle of the region you are imaging. Size is how much of the target it spans — smaller is one spot rather than a smear. Homogeneity is how evenly the oligos are spread, because clumped oligos leave dark stretches.

Why does formamide matter so much?

It lowers every melting temperature by roughly 0.65 °C per percent, so a 50% formamide buffer moves a 78 °C oligo to about 46 °C. The whole specification is stated at the bench condition rather than in water, which is why the buffer is a control here and not a constant.

Is anything uploaded?

No. The design runs in the browser on the sequence already in your project, and the alignment you paste is never sent anywhere.