Twintron Candidate Report

Each candidate receives a Twintron Evidence Score (0-100) from 7 independently-capped components. Differential case-vs-control usage is deliberately excluded from this score -- a real twintron can be used equally in both cell lines, so differential usage is reported as a separate column, not a requirement. Annotation status (Annotated/Novel/Unknown) is also excluded from this score -- the goal here is finding ALL twintrons between the two cell lines, de novo and already-annotated alike, so whether a candidate's inner junction happens to already exist as an ordinary intron elsewhere in GENCODE doesn't make it less real; that status is reported as its own column for characterizing what you found, not used to penalize or reward.

Tiers: ≥90 High-confidence 70-89 Probable 50-69 Possible <50 Likely false positive

Note: before any candidate is scored, its parent outer U12 junction must have at least min_outer_reads (see header above) pooled reads in at least one condition -- this is a hard filter, separate from the "Outer junction validated" scoring component above (which further rewards especially well-supported outer junctions beyond that minimum).

Caveat: this is a heuristic evidence aggregator, not a statistical test, and the splice-site PWMs are illustrative consensus approximations (not a literal MaxEntScan reproduction). Sensitivity is kept deliberately high -- treat this as a shortlist for manual UCSC/IGV inspection and RT-PCR validation, not a confirmed call set.

Passing candidates within the same U12 intron are further classified into one of three structures (evidence score is unaffected by class -- this is about mechanism, not confidence). Classification priority when a candidate could match more than one pattern: Class II > Class III > Class I (a pair that also individually looks like Class III is reported as Class II -- this is the normal case, not an edge case, since a genuine flanking pair's two junctions typically each touch the outer intron's own boundary too).

Structural class is independent of the Annotation column -- a candidate can be, e.g., "Class II, Annotated" or "Class II, Novel"; read them together, not as substitutes for one another.

Terminology note: Classes I-III are this project's operational structural categories, not an established taxonomy from the twintron literature -- described here as "nested events were operationally classified into three structural categories for the purposes of this study."

A 4th class (large-scale skipping across multiple already-annotated exons, spanning beyond a single U12 intron) is intentionally not classified here -- out of scope for this pass.

Important interpretation caveat: structural class describes geometry only. A nested junction is a computational candidate, not an experimentally proven twintron. Class II in particular can also arise from conventional alternative splicing involving a cassette micro-exon (the ZDHHC16 case study, documented in docs/interpretation.md); transcript-level/manual inspection (UCSC/IGV) may still be necessary to determine whether a candidate is a genuine twintron versus conventional alternative splicing.

• Computational candidates, not experimental proof. Everything in this report is a computationally detected candidate nested splicing event. Detection here means geometric evidence (a splice junction inside an annotated U12 intron) supported by split reads; it is not experimental validation of a twintron mechanism.
• Nested geometry can arise from conventional alternative splicing. The ZDHHC16 example demonstrated that a geometrically nested junction can be produced by a conventional cassette-microexon alternative-splicing event whose flanking introns are both GENCODE-annotated. Structural Class II in particular is flagged for manual review for exactly this reason.
• Manual inspection may still be necessary. For candidates you plan to report, verify transcript-level structure in UCSC/IGV (and consider RT-PCR) before calling them twintrons.
• Thresholds are empirical. min_inner_reads / min_outer_reads / the differential threshold (delta inner fraction ≥ 0.15) and the PWM splice-site weights are heuristics, not statistically calibrated cutoffs.
• Mappability is optional. If no --mappability-bigwig is supplied, a soft-mask FASTA heuristic is used; on a non-soft-masked genome this is a no-op that awards full credit (see docs/methodology.md).
• Annotation status is metadata, not evidence. Annotated/Novel describes whether GENCODE already documents the inner junction. It neither proves nor disproves a twintron and does not affect the score.