Phase 5: enzyme usage by system

[edkerk]

Stacked on #6. Per enzyme and condition: how much of it the model uses
(mg/gDW) and what fraction of what was available that is. Then the annotated
systems, summarised and drawn.

python -m overflow.analyze_usage writes enzymeUsages.tsv, capUsage.tsv,
systemMedians.tsv and the two figures.

Exact parity on the part that can have it

The step from per-enzyme usage to capacity usage per system is pure arithmetic
on a table, so it is checked against the published output rather than described:
fed the committed enzymeUsages.txt, it reproduces the committed capUsage.txt
exactly — 201 rows, every value, every GO term, in the same row order.

That also pinned a detail worth having in a test: 30 of the 525 annotation rows
are proteins listed under two systems (ADE3 under both amino acid metabolism and
the THF cycle). R's match takes the first, so this does too.

Result, against the published one

Median capacity usage per system, GECKO 2 / this pipeline (rate-fitted
models, see #5):

system	CN4	CN22	CN38	CN75	hGR
Ribosome	28.8 / 26.3	33.9 / 28.2	33.7 / 47.3	33.6 / 57.3	66.4 / 74.0
ETC	100.0 / 40.6	96.0 / 66.8	100.0 / 100.0	70.6 / 100.0	99.5 / 100.0
PP shunt	5.6 / 11.0	11.7 / 1.9	20.5 / 0.0	20.6 / 0.0	41.7 / 6.1
Glycolysis	0.4 / 1.7	0.5 / 3.1	5.8 / 5.8	10.3 / 11.5	24.1 / 29.2
TCA cycle	2.1 / 3.7	2.1 / 7.1	2.2 / 16.1	0.0 / 43.9	9.2 / 0.1
Amino acid metabolism	1.7 / 1.3	4.6 / 1.9	2.1 / 2.8	3.2 / 3.4	5.6 / 7.8
Nitrogen metabolism	10.3 / 0.9	1.7 / 7.1	2.3 / 0.1	4.8 / 0.1	18.0 / 0.1
THF cycle	11.9 / 4.0	1.6 / 6.0	9.2 / 12.9	0.3 / 12.6	5.9 / 16.5

The ribosome trend survives and sharpens. Capacity usage rises monotonically
with dilution rate (26.3, 28.2, 47.3, 57.3, 74.0), where the GECKO 2 models were
flat across the four chemostats (28.8, 33.9, 33.7, 33.6) and only rose at hGR.

The ETC saturates at CN38, CN75 and hGR, as in the published models, and
rises with growth rate. CN4 (40.6) and CN22 (66.8) stay well below the published
100 and 96; I have not traced why.

The respiratory kcats in this model, against GECKO 2's (read from the prot_
coefficients in its S matrix) and the tutorial model the first version of this PR
used:

reaction		GECKO 2 kcat	tutorial model	this model
r_0438	cytochrome c oxidase (complex IV)	693.5 /s	6000 /s	1376 /s
r_0439	ubiquinol:cytochrome c reductase (complex III)	110 /s	240 /s	290 /s
r_1021	succinate dehydrogenase	167 /s	250–260 /s	250–260 /s
r_0226	ATP synthase	120 /s	120 /s	85 /s forward, 120 /s reverse

The tutorial model's 6000 /s for complex IV made the ETC look cheap and unsaturated
(median 10–17% at the low-uptake conditions). It comes from a bug in geckopy's
and GECKO's kcat matcher, which divides a substrate-specific BRENDA value by the
smallest coefficient in the reaction instead of the matched substrate's own
(BRENDA's 1500 /s divided by oxygen's 0.25); the fix is in
GECKO#484 and
geckopy#57. In this model the
value is 1376 /s.

Reading kcats out of GECKO 2 needs care: its coefficients are
stoichiometry / kcat, so the smallest implied value belongs to the
highest-stoichiometry subunit, not to the enzyme. Taken at face value it makes
ATP synthase look like 12 /s, which is really the c-ring at stoichiometry 10.

The PP shunt and TCA medians fall to about zero where the published ones do
not. The shunt is off at CN38 and CN75 because NADPH comes from IDP2 and
aldehyde dehydrogenase instead (see #5); the TCA median at hGR is 0.1% because
7 of the 21 annotated TCA enzymes are unused there, which puts the median on an
unused one. A median over a handful of enzymes moves a lot when a few switch off.

Figures

Ported from boxplotEnzymeUsage.R: one panel per pathway, one box per
condition, panels sharing a 0-100 axis. The publication's own four colours fail
a colourblind-safety check on two counts and put the beige at 1.83:1 against the
surface, so these use a validated four-slot palette instead; colour is redundant
with the panel title either way, and the numbers behind every panel are written
beside the figure as capUsage.tsv.

Tests

158 tests in this branch.