Divide a BRENDA kcat by the substrate it was measured on

[edkerk]

Problem

A BRENDA kcat is a turnover per molecule of one named substrate. To use it as a
reaction turnover it has to be divided by that substrate's stoichiometric
coefficient. match_kcat divided by the smallest coefficient in the reaction
instead.

Those are the same number only when the reaction happens to be written per the
measured substrate. When another substrate has a coefficient below 1, the
division inflates the kcat rather than reducing it.

Effect

yeast-GEM writes cytochrome c oxidase (r_0438) per one cytochrome c, so its
substrates are ferrocytochrome c at 1.0, H+ at 1.266 and oxygen at 0.25.
BRENDA's S. cerevisiae value under EC 7.1.1.9 is 1500/s, and 1500 / 0.25 gives
the 6000/s carried by the full_ecModel tutorial model, at wildcard level 0 and
origin 1.

Measured turnover for cytochrome c oxidase is in the low hundreds per second.
The inflated value makes respiration look cheap enough that the electron
transport chain no longer saturates in proteomics-constrained models built on
this ecModel.

Fix

Divide by the coefficient of the substrate the BRENDA row matched on. The
matching loop already knows which substrate that is.

In the full_ecModel tutorial this changes 26 of 4671 matched reactions, all of
them previously inflated, by factors of 2, 3 and 4. No reaction changes which
BRENDA row or match level it draws from.

Two regression tests are included: one for the cytochrome c oxidase case, one
for the case the division is meant to handle (two molecules of the measured
substrate per turnover).

The same line is in GECKO MATLAB, src/gather_kcats/fuzzyKcatMatching.m line
401; a matching PR is open there.