Distributed Consequence Finding: Partition-Based and Cooperative Approaches - Agents and Artificial Intelligence - page 435

l-2-aminoadipate
isocitrate

l -lysine

2-oxe-glutarate

itra

trans- aconitate

citrate beta- alanine ala

beta ala

taurine taurine

succinate Fumarate fumarate fumarate succinate cin

l-as citrulline citrulline

nmnd nmn nmna

las l-as

nd

hippurate pp hippurate

nmna

format formate te arginine g arginine ornithine ornithine ithi thi

urea

urea urea rea

formaldehyde hyde y sarcosine creatine

acetylcoa acetate

formalde hyde sarcosine osine atine ati ea rea crreatin

tylcoa acetate tylc y

methylamine methylamine y

creatinine ccr ne eeatinine eatinine

tmao o

mao

pyruvate uva

acryloyl-coa

lactate

lactat

glucose

Fig. 5. Decomposed metabolic pathway problem

source metabolite (here glucose, in red), some target metabolites (here, tmao, acetate
and transaconitate, in blue) and some possible additional metabolites (here arginine and
orthinine, in white), the problem is to determine whether the target metabolites can
be produced by the source metabolite (with eventual addition of the specified additional
metabolites). The production field is then [tmao, acetate, transaconitate,
-arginine, -orthinine] , and the rules about the networks are divided among 6
agents as shown on Fig. 5. The problem is logically formalized with each reaction gives
one activation rule [-reactant, reaction] , one production rule [-reaction,
product] , plus some composition or decomposition rules if respectively the reactant
or the product is a compound (indicated by triangles on our depiction). Each agent's
theory contains all the rules that concerns of one of its metabolites (so communication
languages initially contains only reaction symbols). A last clause indicates that glucose
is initially present ( [glucose] . This clause can be used as a top-clause for computing
Newcarc , otherwise it is considered as a part of the initial theory. In this example, both
Carc and Newcarc contains 4 consequences: [acetate], [transconitate],
[-arginine, tmao], [-ornithine, tmao] . We then compare our two ap-
proaches (partition-base for computing Carc , and cooperative one for Newcarc ) with
the centralized case (computing Carc ).
Table 1 shows the number of resolution steps in each method. As partition-based
approach is affected by the choice of the root, and cooperative one by the order in
which agents are considered for sending of a clause 1 , we give here 6 sets of results,
considering in turn each agents as possible root (or first element in the agent ordering).
Comparing two distributed methods with the centralized one, the total number of
resolution steps becomes fewer in both methods. This is because (i) the partition-based

1

To limit variability, we consider that when sending a clause, an agent always check potentially
relevant neighbour in the same order.