audit-synteny-claim

Referee a synteny / rearrangement conclusion. Input: a synteny result (collinear blocks, syntelogs, an SV inventory INV/TRANS/DUP, a “synteny lost between X and Y” claim, a riparian/dotplot conclusion) plus the biology drawn from it. Output: a verdict — is the conclusion a synteny/rearrangement artifact rather than biology?

Cardinal sin: an artifact read as synteny biology. Every gate below screens one sub-form. A conclusion that survives an unmasked/fragmented/circular pipeline, or that flips with tool stringency, is not certified — it is flagged.

Gate 0 — provenance + claim restatement

Do not evaluate biology until the pipeline is on the table. Capture, or mark UNKNOWN:

1. Tool + version that produced the result (MCScanX / MCScan-2008 / GENESPACE / SyRI / AnchorWave / chain-net / Cactus / other).
2. Anchor level — gene-order (BLASTP/DIAMOND anchors + GFF) vs sequence-level WGA.
3. Masking status of both inputs — were transposons + simple/tandem repeats masked BEFORE anchoring/alignment? (RepeatMasker + Tandem Repeat Finder period <=12; Cactus soft-mask precondition.) UNKNOWN = treat as unmasked until shown otherwise.
4. Assembly contiguity of both genomes — chromosome-level? contig N50? scaffolded, and if so, against what reference?
5. Aligner + preset (for WGA substrates) — minimap2 asm5/asm10/asm20, nucmer mode.
6. WGD/ploidy history of the pair, and the depth setting used (AnchorWave -R/-Q).
7. Stringency knobs actually set (block-size, cscore, onlyOgAnchors, E-value).
8. Restate the claim as one testable proposition (“synteny between X and Y is lost in region R” / “R is an inversion in Y relative to X” / “this block is collinear”). If nothing was actually claimed, STOP — nothing to referee.

If tool, masking, contiguity, or (for WGA) preset is UNKNOWN and load-bearing for the claim → the conclusion is not certifiable; report the missing provenance as the blocker.

Gate 1 — route by synteny method

• Gene-order anchor tools (MCScanX, MCScan-2008, GENESPACE) → Gates 2, 3, 7, 8, 9.
• Sequence-level / rearrangement callers (SyRI, chain-net, minimap2/nucmer/LASTZ/ Cactus substrates) → Gates 2, 4, 5, 6, 9.
• WGD/polyploid comparisons (any tool, pair with unequal WGD history) → Gate 7 always.
• Terminology/citation gate (10) applies to every route.

Gate 2 — TE / repeat-driven false anchors

Unmasked dispersed repeats/TEs produce many spurious homology hits that chain into spurious collinear blocks (gene-order tools) or spurious chains (WGA). Guard = masking before anchoring.

• If Gate 0.3 shows inputs were not masked (or UNKNOWN) and the claim asserts a present block/synteny → FLAG as possible repeat-driven false anchor. The remedy is RepeatMasker (transposons) + Tandem Repeat Finder (simple repeats, period <=12) before alignment (chain-net); Cactus makes soft-masking a hard precondition.
• MCScanX’s duplicate gene classifier separates dispersed vs tandem vs WGD/segmental anchors — a block built mostly from dispersed-repeat anchors is suspect.
• Do NOT assert a numeric inflation fold. The notes describe the mechanism and the masking remedy but quantify NO TE false-anchor fold-change. Any “~Nx inflation” figure is unsupported → state the direction (inflation) without a number.

Gate 3 — tandem-duplicate block inflation

Tandem arrays inflate anchor counts and inflate block “confidence.”

• MCScanX collapses consecutive shared-gene matches whose partners are within 5 genes to one representative pair (smallest E-value); tandem = gene-rank-difference 1, proximal = rank-difference <20. MCScan-2008 collapses tandems <50 kb apart. GENESPACE collapses tandem arrays to array-representative genes before rank-ordering.
• If a block’s apparent strength comes from a tandem array that was not collapsed (collapse disabled, or a long-distance tandem that survives the 50-kb rule — Tang notes residual long-distance tandems) → FLAG; the block count is inflated, not biology.

Gate 4 — fragmented-assembly false “synteny loss”

Short contigs truncate blocks and manufacture artifactual “loss.”

• SyRI requires chromosome-level assemblies; its syntenic-path spine runs end-to-end along a chromosome and breaks on contig input. It stays “nearly complete” only down to N50 > 500 kb; below that, calling degrades. Remedy: build pseudo-chromosomes first.
• If a “synteny lost” claim is drawn from fragmented / sub-chromosomal input → FLAG as possible fragmentation artifact, never pass.
• Aligner false-absence check: for WGA substrates, verify the aligner/preset could have detected the sequence at all. minimap2 asm5 “will not extend to regions with 5% or higher sequence divergence” (asm10 ~1%, asm20 ~several %); nucmer seeds on long exact matches and by default only unique-in-reference, so divergent/repetitive regions yield few anchors. A “loss” that is really sensitivity below threshold is false absence. There is NO ~70% identity floor in the minimap2 or MUMmer4 notes — do not invoke one.
• [GAP: no ingested primary states a hard chromosome-level N50 cutoff (e.g. >=1 Mb). SyRI gives N50>500 kb as a graceful-degradation point, not a validity floor. Do not assert a

  =1 Mb threshold as cited.]

Gate 5 — reference-guided-scaffold circular reasoning

If the query assembly was ordered/scaffolded using reference X (or pseudo-chromosomes were built by homology to X), then computing “synteny to X” re-derives the scaffolding input — circular, not independent evidence.

• Check Gate 0.4: was the query scaffolded against, or pseudo-chromosome-built from, the same genome the synteny is measured against? (SyRI explicitly allows pseudo-chromosome construction “using homology to a chromosome-level reference” — that reference must not then be the synteny partner the conclusion leans on.)
• If yes → FLAG as circular; the synteny signal is partly an artifact of assembly construction.
• [GAP: no ingested primary states this circularity criterion. It is a real validity axis; encode the check but label the criterion unsourced / convention, not cited.]

Gate 6 — micro/macro-synteny conflation

A “synteny loss” (or “synteny conserved”) claim that ignores scale.

• Block reporting is gated by size: MCScanX reports chains scoring >250 = >=5 collinear gene pairs; GENESPACE blkSize default = 5 unique hits within a block; MCScan-2008 gates on block score >300 (a score, not an anchor count). Micro-synteny (local gene order) and macro-synteny (chromosome-scale collinearity) are separated by these thresholds — a conclusion must state which scale it is about.
• If the claim conflates scales (e.g. “synteny lost” from a handful of below-threshold micro-rearrangements, or “synteny conserved” macro-claim from a single small block) → FLAG.
• [GAP: no in-corpus primary supplies a synteny decay rate / half-life. Do not assert a decay/half-life number to argue expected micro-synteny erosion.]

Gate 7 — WGD / polyploid 1:many collapse

Polyploid paralogs collapse many-to-one into blocks, faking loss or faking simple 1:1 orthology.

• Correct control is alignment depth: AnchorWave sets max depth per genome via -R (reference) / -Q (query) — e.g. maize<->sorghum -R 1 -Q 2. MCScan-2008 reads depth/multiplicity against the pair’s per-lineage WGD history (expected depth 3 for gamma-only Carica/Vitis, 6 for Populus, 12 for Arabidopsis).
• If the pair has unequal WGD history and the analysis used depth 1:1 (or ignored ploidy), a “1:1 synteny” or “synteny lost” read is suspect → FLAG.
• Confabulation flag: any citation of proali --ploidy (or any --ploidy flag) is wrong — the AnchorWave note documents only -R / -Q; --ploidy is not in the source. Flag it as confabulated and correct to -R/-Q.

Gate 8 — method / stringency divergence

The same input yields different block sets under different tools/stringency.

• MCScanX (E-value + score/gap thresholds) vs JCVI-MCScan cscore (a reciprocal-best-hit-ratio control) vs GENESPACE orthogroup-constrained anchors (onlyOgAnchors) are different stringency regimes. cscore is a JCVI-software feature — ABSENT from the Tang-2008 paper; MCScan-2008 controls stringency via E-value (1e-5), best-5-hits, DP scoring, and the score-300 / E<1e-10 cutoffs.
• If the conclusion would flip under a defensible alternative tool/stringency (or the authors tuned to the setting that produces their claim), → FLAG as stringency-dependent, not robust. A single-setting synteny claim with no sensitivity check is not certified.

Gate 9 — SyRI small-INV / strand-artifact noise

SV callers emit many tiny “inversions” that are strand/alignment noise, not biology.

• Goel/SyRI states NO minimum inversion or SV calling size; local variants overlapping syntenic/inverted regions are filtered, but small calls are otherwise reported. WGA strand artifacts also arise at block ends (AnchorWave: opposite-strand anchors at block ends drop/skip sequence).
• If a biological claim rests on many small INV calls with no size/strand filtering described → FLAG as possible small-INV noise.
• [GAP: the common >=5 kb inversion filter is convention, not cited — Goel states no minimum size. Do not assert a >=5 kb (or any) INV floor as sourced; label it convention.]

Gate 10 — terminology / citation precision (anti-invention)

• If the claim leans on a synteny-vs-collinearity definitional citation, do not accept an unverified one. [GAP: there is no clean in-corpus primary for the distinction; the commonly-cited “Fitch 1976, J Mol Evol 7:271” does not resolve (that page is Zuckerkandl, not Fitch). Synteny is commonly credited to Renwick 1971; Fitch’s actual terminology paper is 1970 (Syst Zool 19:99, ortholog/paralog).] Flag any specific definitional citation as unverified unless the author supplies a resolvable primary.
• More generally: any threshold or version invoked “per ” that the references do not support (a >=5 kb INV floor, a >=70% identity floor, a >=1 Mb N50 floor, a --ploidy flag, a cscore attributed to Tang-2008, blkSize as “orthogroups”) must be labeled convention / unsourced / [GAP], never asserted as cited.

Verdict synthesis

Collect the flags. For each fired gate, state: the sub-form, the evidence (or missing provenance), and whether it is artifact-suspected or unsourced-claim.

• FLAGGED — likely artifact: >=1 gate fires on a mechanism (unmasked repeats, tandem inflation, fragmentation, circular scaffold, WGD collapse, stringency flip, small-INV noise) that would produce the observed result absent the claimed biology. Do not certify; name the remedy (mask, use chromosome-level input, set -R/-Q, run a stringency sensitivity check, filter small INVs).
• FLAGGED — unsourced claim: a threshold/citation is asserted without support (Gate 10). Require it be labeled convention or replaced with a resolvable primary.
• PASS: no gate fires — inputs masked, chromosome-level, non-circular, ploidy-aware, robust across stringency, scale stated, citations resolvable.
• [GAP: no ingested source supplies a rubric for combining multiple partial flags into a single confidence score. Report each flag explicitly; do not invent a weighted combined-confidence number.]