Pathway satisfiability — background: pathway hole filling

IN_PROGRESS PIPELINE

Background: the pathway-hole-filling landscape

Companion to Pathway satisfiability. This page situates the
project against ~20 years of prior art on pathway holes — the question "a pathway looks like
it should run, but a step has no gene assigned; what is going on, and which gene fills it?"
The short version: the field is mature for microbial genomes, where the discriminating
variable is gene presence/absence, and essentially silent on the metazoan version of the
question, where the genome is constant and the discriminating variable is which isozyme is
expressed in which context. That gap is what this project targets.

The term "pathway hole" itself comes from SRI's Pathway Tools (Karp lab): a hole is a
reaction in an inferred pathway for which no enzyme in the organism has been assigned.

Three families of prior work

The literature clusters into three problems that are often conflated but are actually
distinct.

1. Step-finding: does the genome encode this pathway? (presence/absence)

Given a genome and a pathway written as a set of required steps (often with alternative
routes), decide whether every step has a plausible gene. This is the family closest to what
this project does — the difference is the oracle.

2. Hole-filling proper: find the gene for a known-but-unassigned step

Here the pathway is believed to run (or the reaction is known to occur), one step lacks an
assigned enzyme, and the task is to nominate the gene. This is where the phrase "pathway
hole filler" originates.

3. Network gap-filling: add reactions so the model balances

A different sense of "gap": in a genome-scale metabolic model, a blocked metabolite (a
dead-end that cannot be produced/consumed at steady state) is a gap, and the fix is to add
reactions — not necessarily assign genes — so the model can produce biomass on a defined
medium.

This family is flux-driven and gene-agnostic: it can restore a viable model by adding a
reaction nobody has a gene for. Useful, but it answers "what reaction must exist," not "which
gene, in which context."

Where this project sits

granularity of the "hole" discriminating variable candidate output
GapMind / KEGG modules / MinPath pathway step gene presence in a genome which step is missing
Pathway Hole Filler / IMG reaction with no gene genome context (operon, profiles) the gene that fills it
GapFind / ModelSEED / gapseq blocked metabolite steady-state flux the reaction to add
this project pathway step expression/context (tissue, cell zone) the context-localised gene hypothesis

Every tool above answers a genome-level question: given the gene content of one genome,
is the pathway present, and if a step is empty, which gene fills it? That is exactly the right
question for a microbe, where a genome roughly is an organism.

It is the wrong question for a metazoan. Every human cell carries the whole genome, so
"is gluconeogenesis present?" is trivially yes everywhere — yet glucose output is restricted to
a few tissues and, within the liver, to a few cell layers. The discriminating variable is not
presence but which isozyme is expressed where. None of the microbial hole-fillers can ask
this, because their oracle is gene-content, which does not vary across a metazoan's cells.

This project keeps the GapMind-style logic (a module as steps with OR-branches; find the gate;
report the gap) but swaps the oracle from genome presence to context expression (GTEx
tissue, liver zonation, substrate-entry route). A "hole" becomes not "this genome lacks the
gene" but "this pathway cannot be wired up in this context — and here is the gene and place
where it fails." Crossing that with an independent activity claim (a known tissue function, a
growth phenotype) is the abduction step, which turns a context-specific hole into a
reviewable, gene- and location-localised hypothesis — the metazoan analogue of the Bayesian
hole filler's candidate list.

References