Identifier-bound pathway maps of the places where weak magnetic fields
could plausibly act in a cell — each node carrying its quantum chemistry, its
supporting citations, and an explicit statement of how much is actually known.
Companion software to the review “Plant Responses to Near-Null
Magnetic Fields: Geomagnetism, Bioenergetics, and Primary Metabolism in Arabidopsis
and Brassica” (D. M. Porterfield & R. Barker, Purdue University).
The ontology exists because these are different questions, and collapsing them is
how “contains a flavin” quietly becomes “is magnetically sensitive”:
Axis
Question
quantum_class
What chemistry does this entity carry?
Structural, and largely uncontroversial.
evidence_tier
How much is actually known about
its magnetic-field sensitivity? This is where the claims live.
A tier that asserts a demonstrated or inferred field effect cannot validate
without a resolved DOI. The tier is drawn as a border style on every map, so a
hypothesis looks provisional on the page:
Tier
Meaning
Nodes
T1
Demonstrated A field was applied and an effect measured on this entity.
4
T2
Inferred Shown for the containing complex or process, not the entity itself.
31
T3
Plausible A physical route exists; no field measurement has been made.
44
T4
Context Structural context; no quantum claim.
47
The maps are mostly dashed, and that is the finding.
44 of 126 nodes carry a plausible physical route
with no field measurement behind them. The Arabidopsis near-null-field literature is
developmental and nutritional; it has not yet measured a respiratory complex directly.
No base map encodes a quantitative value — numbers appear only when measured data is
projected onto one.
The maps
Each map compiles from a declarative source with no coordinates in
it — box geometry is derived from measured text, which is what stops a label
outgrowing its box. Every map is published as SBGN-ML (Process
Description) so it opens in Newt, VANTED, CySBGN and the
SBGN Pathway Visualizer, with the quantum annotation carried in
<extension> blocks that other renderers safely ignore.
QBM-01 · Mitochondrial electron transport and oxidative phosphorylation
The respiratory chain redrawn so that each node carries its quantum chemistry and the evidence behind it. Solid heavy borders mark entities where a magnetic-field effect has been demonstrated (T1); solid light, where it has been shown for the containing process (T2); dashed, where a physical route exists but no field measurement has been made (T3); hairline, structural context with no quantum claim (T4). The map is almost entirely dashed, and that is the finding: the Arabidopsis near-null-field literature is developmental and nutritional, and has not yet measured a respiratory complex directly. Alternative oxidase and the type II NAD(P)H dehydrogenases are drawn because they are the confounders — a plant can route electrons around a lesion in complex I or complex IV, so respiration rate alone cannot test the hypothesis and cytochrome-versus-alternative flux must be partitioned.
17 nodes · 23 edges · T2 1 · T3 10 · T4 6 ·
28 loci
QBM-02 · Photosynthetic electron transport and plastid redox
The photosynthetic chain redrawn with each node's quantum chemistry and the evidence behind it. Photosystem I is the strongest structural case in the atlas: charge separation at P700 makes a genuine spin-correlated radical pair, the species the radical-pair mechanism was written for. It is nevertheless only T2, because the one direct field measurement on photosynthesis in the review's bibliography was made in lima bean rather than Arabidopsis, and because that pair is extremely short-lived and its known field effects are seen well above geomagnetic strength. Cyclic electron flow is drawn for the same reason alternative oxidase appears on QBM-01 — it lets a chloroplast rebalance ATP against NADPH without changing linear flux, so measuring linear electron transport alone cannot separate "no effect" from "compensated effect". Four nodes are plastid-encoded (ATCG loci) and will receive no value from poly-A-selected RNA-seq, which is most of OSDR's plant studies.
14 nodes · 18 edges · T2 3 · T3 6 · T4 5 ·
17 loci
QBM-03 · Cryptochrome radical-pair photochemistry
Cryptochrome is the only plant entity in the review's bibliography where a magnetic field was applied and an effect measured on the protein's own chemistry rather than on a downstream phenotype, so it carries the atlas's only T1 annotations. The hyperfine partners that make this pair field-sensitive are the flavin's own nitrogens and protons — 14N at 99.6% natural abundance and 1H at essentially 100% — so unlike the 57Fe route argued for iron-sulfur clusters, this one is not abundance-limited. That asymmetry is much of why cryptochrome is the strongest case here and the Fe/S nodes elsewhere are not. Phytochrome B is drawn deliberately as context with no quantum claim: its photochemistry is a Z/E isomerisation offering no spin-based route, so reported field effects on phytochrome signalling are most simply a downstream consequence of cryptochrome. Note also that one blinded replication (Harris et al. 2009) did NOT reproduce cryptochrome-dependent field effects; that result is recorded in the ontology on the same footing as the positive reports.
QBM-04 · Iron-sulfur cluster biogenesis — ISC, SUF and CIA
Every Fe/S node on QBM-01, QBM-02 and QBM-05 depends on one of these three systems, so a defect here sits upstream of the whole atlas and would appear as a metabolic lesion well before any visible phenotype. Read the tiers carefully, because they mark a distinction that is easy to blur: the near-null-field evidence in this area (Islam 2020, Vigani 2021, Narayana 2018) measured iron UPTAKE and Fe-S interplay — nutritional physiology — and is attached to the Fe/S supply node, which is therefore T2. No experiment in the bibliography applied a magnetic field and measured cluster assembly itself, so every assembly node remains T3 no matter how attractive the mechanism looks. The single arrow from mitochondrial export to the CIA system is the reason this map reaches beyond bioenergetics: CIA clients include the DNA replication and repair enzymes, so a mitochondrial lesion propagates to genome maintenance.
11 nodes · 14 edges · T2 1 · T3 7 · T4 3 ·
23 loci
QBM-05 · TCA cycle and the Fe/S-dependent steps
The review's Figure 4 renders this as a linear chain from glucose to oxaloacetate with no OAA-to-citrate closure, which the draft's own margin note flags. The closure matters: it is what shows that the two Fe/S-dependent steps throttle flux around the whole loop instead of interrupting a line. Aconitase and succinate dehydrogenase are the review's sentinel nodes and are annotated once on QBM-01, referenced here rather than re-declared, so one entity cannot carry different evidence on two maps. The citrate-to-isocitrate ratio is the most direct available in-vivo proxy for Fe/S cluster integrity and is worth measuring in any near-null-field metabolomics experiment. Note that plant mitochondrial fumarase is a class-II enzyme with no Fe/S cluster, unlike the class-I bacterial form it is sometimes confused with.
13 nodes · 17 edges · T2 1 · T3 3 · T4 9 ·
15 loci
QBM-06 · Seed germination before photosynthesis
Before chloroplasts mature, an oilseed embryo is entirely heterotrophic: stored triacylglycerol must be mobilised, run through peroxisomal beta-oxidation and the glyoxylate cycle, and converted to ATP by mitochondrial respiration. That makes the first 0-72 hours after imbibition the window where a respiratory or Fe/S lesion would have most leverage, and the reason Arabidopsis and Brassica are comparable systems for the review's argument. The original Figure 6 carried a summary table of "normalized trend ranges" for germination rate, radicle length, ATP status and seedling vigour; those were qualitative cross-study impressions rather than measured values, so no number from them is reproduced here. The map shows the route and the endpoints; any quantity has to arrive from projected data. Every node on this map is T3 or T4, which is the honest state of the field: the review's central proposal is a geomagnetic-versus-near-null imbibition time course, and no such experiment exists in its bibliography.
11 nodes · 15 edges · T2 1 · T3 5 · T4 5 ·
12 loci
QBM-07 · ROS production, scavenging and redox buffering
A weak-field effect on an electron-transfer branching ratio may be far too small to measure as a rate change, but ROS signalling is threshold-like, so the same small change can register as a transcriptional response. That is why the ROS-responsive programme is the best-evidenced node in the whole atlas, and equally why it cannot identify a mechanism: an amplifier reports that something moved upstream, never what. The map deliberately includes enzymes with no spin route at all — the peroxiredoxin/thioredoxin system is pure two-electron thiol chemistry with no radical intermediate and no metal — so that "redox" is not read as a synonym for "magnetically addressable". The heme peroxidases do cycle through a genuine porphyrin radical, but it is exchange-coupled to the ferryl iron and extremely short-lived, and that caveat travels with the annotation.
12 nodes · 14 edges · T2 3 · T3 5 · T4 4 ·
17 loci
QBM-08 · From magnetic environment to plant phenotype
The flagship map, and the one that has to be read most carefully. The review states that its Figure 7 is designed for hypothesis generation and experimental planning rather than as a proven causal network, and the tiers here preserve that distinction: the input perturbation and the developmental outputs carry real evidence, while the molecular route between them does not. Read the border styles from left to right and the shape of the field's problem is visible — the evidence is concentrated at the two ends, on the perturbation and on the phenotypes, while the middle, where the mechanism would have to live, is almost entirely dashed. Primary metabolism, the review's central thesis, is drawn as T4 on purpose: the metabolic route is argued rather than demonstrated, and the map should not let it borrow credibility from the developmental evidence beside it. The model is explicitly modular — cryptochrome may dominate in blue-light-grown seedlings, mitochondria in dark-imbibed seeds, plastids in greening tissue, and nutrient uptake in roots — so a single phenotype may be produced by different weightings of the same components depending on tissue, stage and light regime.
19 nodes · 27 edges · T1 1 · T2 8 · T3 4 · T4 6 ·
34 loci
QBM-09 · Iron uptake and sulfate assimilation
One of two maps the review motivates but does not draw. That is an odd gap, because mineral nutrition is where the direct Arabidopsis near-null-field evidence is strongest: three separate studies report effects on root mineral nutrition, iron uptake efficiency and Fe-S interplay. The map makes the loop visible. Sulfate assimilation supplies the sulfide that every cluster on QBM-04 is built from, and sulfite reductase is itself a ferredoxin-dependent siroheme-[4Fe-4S] enzyme drawing on plastid reducing power from QBM-02 — so iron and sulfur status are not two inputs but one feedback system, which is exactly why Vigani et al. found the effect most pronounced under combined Fe and S deficiency. Note the tier pattern: uptake and assimilation are T2 on real evidence, but they are classed as structural context because no quantum route is proposed for them. A demonstrated effect with no mechanism is recorded here as precisely that.
10 nodes · 13 edges · T2 4 · T3 3 · T4 3 ·
34 loci
QBM-10 · Hormone and circadian integration of the flowering response
The second map the review motivates but does not draw, covering the other area where direct Arabidopsis evidence is strongest. Delayed flowering is the most reproducible near-null-field phenotype in the whole literature, and separate studies have implicated cryptochrome, gibberellin and auxin in producing it. Drawn together, the route from a photoreceptor to a phenotype becomes concrete and testable with existing mutants — which is what the review's call for cry1 cry2, phytochrome and hormone genetics amounts to. The clock is the map's most speculative element and is marked T3: no field measurement on clock components exists in the bibliography. It is drawn because a clock is a gate, so the same perturbation applied at different times can produce opposite outcomes — which could explain some of the inconsistency across near-null-field studies, and is testable rather than merely plausible. That is offered as an experimental design consideration, not a finding.
11 nodes · 15 edges · T1 1 · T2 6 · T3 1 · T4 3 ·
24 loci
The ontology is anchored on Arabidopsis, because that is where the
near-null-field literature is. Projecting onto another organism runs
two independent orthology backbones and reports where they disagree,
rather than silently preferring one:
Ensembl Compara pan_homology — the only division that
crosses kingdoms. The plants division does not reach animals at all.
OrthoDB v12 — the committed human-anchored matrix from
OSDR_X-species_V2.
Projecting the atlas's 125 loci to human maps 68, with both backbones agreeing on 22 and disagreeing on 0. The other species run on Ensembl alone (mouse 64/125, fly 63/125, worm 61/125, yeast 68/125) because the committed OrthoDB matrix is human-anchored — stated rather than left to look like two-method corroboration. The unmapped set is the validation: the alternative oxidase and type II NAD(P)H dehydrogenase loci fail to map, which is correct, because humans have neither. A projection that matches nothing raises rather than returning an empty frame that would render as a blank map.
Demonstrations
Each page projects real measurements onto the maps and states its provenance class,
because the three available kinds of data do not carry equal weight:
genelab_processed (NASA's own pipeline), depositor_normalised
(the depositors' ratios, no model fitted) and published_results (numbers
read out of a paper's supplementary tables, because nothing was deposited).
26 loci, 6 timepoints across
2 tissues, 196 values —
7 of them on atlas nodes. The first overlay in this atlas
to carry time rather than a single snapshot.
Provenance:published_results. No raw data
was deposited; the Data Availability Statement reads
“no Data Availability Statement is present in the article; the quantitative data appear only as main-text tables”.
21 loci, 1 timepoints across
1 tissues, 21 values —
4 of them on atlas nodes. The first overlay in this atlas
to carry time rather than a single snapshot.
Provenance:published_results. No raw data
was deposited; the Data Availability Statement reads
“Data will be made available on request”.
194 loci, 7 timepoints across
2 tissues, 2,716 values —
12 of them on atlas nodes. The first overlay in this atlas
to carry time rather than a single snapshot.
Provenance:published_results. No raw data
was deposited; the Data Availability Statement reads
“Data are available as supplementary tables and further data can be provided upon request”.
Drop a differential-expression table and see it projected onto any map, with the
orthology run for you if it is not Arabidopsis. Nothing is uploaded
— the file is read in your browser and joined against the catalogue that ships with this
site.
It carries the same refusals as the generated figures: an empty join stops with an
explanation rather than drawing a blank map, coverage below 25% warns and states the
number, and a namespace that cannot reach your chosen species is caught before the join
rather than after. It also names the nodes that collapse onto a shared gene.
A dose × time radiation course in Arabidopsis (2 doses,
4 timepoints, 123 atlas loci), asked
against the magnetic-field studies. Radiation is not a field regime and
the page does not treat it as one — the question is overlap.
2 of 11 shared loci respond to both
radiation and a near-null field in the same direction. Reported as a lead, not
a result: eleven loci is too few to test, and the page says so rather than computing a
p-value that would look like evidence.
It also forced a check the atlas lacked: under radiation almost every multi-locus node
has loci that disagree in sign, so a node's single trace can be the aggregator
choosing between genes doing different things. The AOX family is the worked case.
The cleanest field-only contrast in the atlas: 1g inside the magnet against 1g
outside it, in Arabidopsis, at 100% node coverage on every
map. The study also carries gravity controls that use no magnet at all.
And the result that constrains every other page: the atlas's
125 loci move no more than 29,202 random
loci from the same array (p = 0.71). Selecting for quantum
chemistry did not select for field responsiveness here, and the page says so.
Provenance:depositor_normalised — the
depositors' normalised ratios, no model fitted, so no significance is shown.
The cross-species bridge doing real work: 63 of 125 atlas
loci reach the fly, and Arabidopsis CRY1 lands on Drosophilacry,
the canonical animal magnetoreceptor. Of 420 contrasts, 5 isolate
the field from the gravity the levitation also changes.
The result is null and reported as one: none of the 15 nodes
with data moves the same way in all 5 contrasts. The page also
states what the projection destroys — on the cryptochrome map three distinct nodes
collapse onto one fly gene.
Provenance:genelab_processed, and a
strong-field study — the opposite end of the axis from the review.
A pre-registered test, and why it could not settle anything
The hypothesis, null, data sources and confounds were committed before the
script was run. The question: are human orthologs carrying spin-bearing cofactors
over-represented among genes conserved-suppressed across six species in spaceflight?
carries a spin cofactor
does not
conserved down
13
330
conserved up
6
241
Odds ratio 1.582 (95% CI 0.587–3.925),
Fisher exact p = 0.48 — fail to reject the null. Iron-sulfur
chemistry specifically, the mechanism the review argues hardest for, gives an odds
ratio of 0.90.
That negative result settles nothing, and saying so
is the point. Simulation on the realised marginals puts the minimum detectable
odds ratio at 4.0 for 80% power; a genuine doubling of odds would have been
missed three times in four.
Three structural reasons this dataset cannot answer the question:
only 19 of 590 directional genes carry a spin-bearing cofactor; the mitochondrial
stratum has no contrast at all, because every conserved mitochondrial gene in the
table is suppressed and none is up-regulated; and spaceflight bundles microgravity,
ionising radiation and a hypomagnetic environment, so attribution to the magnetic
component was never available. The result strengthens the review's own call for direct
bioenergetic measurements under controlled conditions.
A curated study database whose DOIs were mostly wrong
Every DOI in the companion repository's nnmf_study_database.csv was
resolved against CrossRef: only 7 of 49 title-match. Its Belyavskaya
row cites a paper on calcium gradients in the fish inner ear. The manuscript's own
bibliography, by contrast, is clean — 42 of 42 resolve and title-match —
so the evidence base here is built from the bibliography, re-verified on every run,
and the build refuses to write its output if any reference fails.
Gene symbols are not safe to type
Resolving every symbol against Ensembl during authoring caught five collisions that
would each have put the wrong gene on a map: ACO2 returns ACC oxidase 2
rather than aconitase 2, LIP1 a lipase rather than lipoyl synthase,
CAT2 and CAT3 cationic amino acid transporters rather than
catalases, and KAT2 a potassium channel rather than the thiolase. The
ontology stores loci, never symbols.
A cofactor table with no iron-sulfur entries
The KEGG-derived cofactor mapping in OSDR_X-species_V2 contains zero
iron-sulfur entries despite its generating script listing Fe-S among the cofactors to
map, and only 5 heme and 10 FMN genes genome-wide. The script skips any cofactor whose
cache file is missing without reporting the skip. UniProt/Swiss-Prot was used instead,
which curates 71 human Fe-S proteins.
Figure legibility as a test, not a review step
The review draft carries the margin note “the text spills out of the box's… and
the TCA cycle needs some work”. Both are fixed structurally: boxes are sized around
measured text so a label cannot overflow by construction, and the TCA cycle is redrawn
as an actual closed cycle. A browser-side probe then measures what was really painted —
using the browser's own text layout rather than re-running the compiler's arithmetic —
and asserts no overflow, no overlapping boxes or compartment bands, nothing clipped,
and a caption that states provenance. All ten maps pass across 1,447 elements.
Using the atlas
The published catalogue is catalog/manifest.json:
a flat, self-describing list of every map with its SBGN URL and a per-node annotation
sidecar carrying evidence tier, quantum class and citations — so a consumer can draw
the tier channel without parsing SBGN extensions.
pip install pyyaml matplotlib scipy
PYTHONPATH=src python3 -c "
from qbio import ontology, maps
onto = ontology.load()
for r in maps.compile_all(onto):
print(r['id'], r['nodes'], 'nodes', r['tiers'])"
Project a NASA OSDR contrast onto a map:
PYTHONPATH=src python3 -c "
from qbio import ontology, maps, osdr, project
onto = ontology.load()
spec = maps.load_spec('maps/src/QBM-01_mitochondrial_etc.yaml')
nodes = [(n, onto.get(spec.nodes.get(n, {}).get('qbo',''))) for n in spec.node_ids]
table = osdr.load_contrast(38, contrast='(FLT)v(GC)')
proj = project.project_expression(map_id=spec.id, nodes=nodes, table=table)
print(proj.summary())"
