Coordinate synteny: definition, scoring, and validation
SynPAL tests whether the genes of a metabolic pathway remain physically clustered across 55 prokaryotic genomes and, where they do, derives a gene-order recommendation from the conserved arrangement. This page defines the metric, the null model, the design tiers, and the scope of the claims.
The transfer unit is a chromosomal layout
Pathway databases export reaction chemistry; chromosomal context must be recovered separately.
Metabolic engineering tools work from reaction networks: a pathway is a set of enzymes, and the database records which reactions occur, in which order of columns, in a pathway table. Heterologous reconstruction, however, requires a physical gene order — an arrangement on DNA — that the pathway table never contained.
The earlier SynGen study scored conservation using the column order of BioCyc pathways.col files and reported ~78.9% of pathways as conserved. That metric measures how curators arranged a table rather than how genomes arrange genes. A fair re-run of the same column-order metric on a clean panel still yields 79.4% MEDIUM+HIGH — while real coordinate synteny on the same panel gives 32.2%. The two numbers describe different objects.
SynPAL therefore starts from genomic coordinates: genes are mapped to real positions on real replicons, and conservation is defined over those positions. The transfer unit becomes the conserved cluster — an arrangement under evolutionary selection — rather than an arbitrary list of parts.
Coordinate synteny, co-directional
A cluster is a run of pathway genes on one replicon, one strand, within 2,000 bp of each other.
For each pathway and each organism, pathway genes are anchored to genomic coordinates by name. Genes are then grouped intoco-directional clusters: consecutive pathway genes qualify as one cluster when all of the following hold —
The gap is measured between the actual gene boundaries (next.lo − prev.hi), not between midpoints, and the default maximum is 2,000 bp. This is method v2(method v2 co-directional): the strand requirement makes every cluster a plausible transcriptional unit. The earlier v1 rule was proximity-only and mixed convergent and divergent runs.
Only organisms with at least two mapped genes contribute to scoring; a single hit carries no co-localization information.
Scores and conservation levels
Clustering, gene order, and strand consistency are scored separately, then combined.
Per pathway, over all qualifying organisms:
When order cannot be evaluated (too few shared genes), the synteny score falls back to the cluster score alone. The conservation level is thresholded on the cluster score:
| Level | cluster_score | Reading |
|---|---|---|
| HIGH | ≥ 0.7 | most organisms carry the pathway in tight clusters |
| MEDIUM | 0.5 – 0.7 | substantial partial clustering |
| LOW | 0.3 – 0.5 | clusters exist but are inconsistent |
| VERY_LOW | < 0.3 | effectively scattered |
A pathway is additionally flagged order_conservedwhen order_score ≥ 0.7 — clustering is commoner than conserved order, and the two should not be conflated.
Null model and known-operon recovery
The signal is checked against a randomized null model and against independently documented operons.
Null model
For each pathway, random gene sets matched for count are drawn from the same genomes and re-clustered with the identical rule (Monte-Carlo, 200 null sets per pathway). On the primary panel the observed mean cluster score is 0.436 against a null mean of 0.268, and 626 of 799 pathways (78.3%) exceed their null at p < 0.05; all 88 HIGH pathways are significant. Per-pathway detail ships with the atlas — for example PWY0-1324: observed 0.811 vs null mean 0.395 (p = 0.002, z = 104.2).
Known-operon recovery
Conservation should rediscover biology that was curated for other reasons. Of 12 HIGH ∩ order-conserved prime candidates, roughly 10 match documented operons or gene clusters in the literature (9 with matching gene order). Systematically, 43 of 47 dominant clusters from HIGH ∩ nOrg≥5 pathways are confirmed or partially confirmed against independent operon annotations, versus 8 of 15 VERY_LOW controls (Fisher p ≈ 0.0025). The metric is not recovering curator noise.
Column-order contrast
On the same clean panel, the legacy column-order metric classifies79.4% of the 63 name-matched pathways as MEDIUM+HIGH; coordinate synteny classifies32.2% of all 799 analyzable pathways. On the joined set the two scores barely correlate (Pearson ≈ 0.17). Column order is not synteny, and the older 78.9% headline should not be read as genomic conservation.
Design-readiness tiers T1–T4
Tiers are computed per pathway over organisms with ≥ 2 mapped genes, from the best cluster any single organism offers.
Let max_ratio be the best cluster ratio in any one organism and max_cluster_size the largest cluster found. Then:
| Tier | Rule | n | Interpretation |
|---|---|---|---|
| T1 | max_ratio ≥ 0.999 | 224 | at least one organism carries the full mapped pathway in one co-directional cluster — reconstructable as a single operon |
| T2 | cluster_score ≥ 0.5 | 83 | conserved partial clustering — several cassettes, one host |
| T3 | cluster_score < 0.5 | 409 | clusters exist but are weakly conserved — heterologous assembly across species |
| T4 | max_cluster_size < 2 | 171 | fully scattered — chemistry-driven design; no synteny evidence |
T1 + T2 = 307 of 887pathways are design-ready with synteny as the ordering evidence. T4 pathways are reported for completeness; the resource provides no layout evidence for them.
From conserved cluster to assembly-oriented construct
Three steps convert a top-scoring arrangement into downloadable sequence files.
Source selection
For each designable pathway the pipeline prefers asingle donor species: the organism whose cluster covers the most of the pathway in one co-directional run. Only when no single organism suffices (T2/T3) are modules combined from multiple species, and the number of source species is minimized. The chosen source is recorded in every construct.
Regulatory flanks, 40–800 bp
Each cassette extracts the native intergenic sequence on both sides: the real gap to the neighboring gene, clamped to[40, 800] bp per side. These flanks carry the donor's regulatory context; they are not predicted promoters, and constructs remain fully in silico.
Gibson primers — heuristic
Overlap primers for Gibson assembly are generated with a Wallace-rule heuristic — Tm = 2·(A+T) + 4·(G+C) on the short binding region, with placeholder vector-homology tails. There is no nearest-neighbor model and no hairpin, dimer, or misprime checking. The primers make the designs shaped for assembly, not ready to order: re-design with Primer3 (or equivalent) before ordering oligos.
Two panels, one method
The paper headline and this resource come from different organism panels. Both use method v2 with a 2,000 bp gap.
| Panel | Organisms | Pathways | M+H |
|---|---|---|---|
| Primary landscape (paper headline) | 45 | 799 | 32.2% |
| Resource DB (this site) | 55 | 887 | 32.7% |
The resource DB adds ten organisms to the locked 45-species panel; the row counts are therefore not identical, and the MEDIUM+HIGH fraction moves from 32.2% to 32.7%. Every number on this site is computed on the 55-organism resource DB (887 pathways); the 45-organism numbers appear only where the paper headline is discussed.
Per-pathway evidence and downloads
The atlas reports every pathway's scores, per-organism clusters, and null-model statistics; the design gallery carries the downloadable constructs.