TotemBioNet Enrichment Methodology: Application to the

Qualitative Regulatory Network of the Cell Metabolism

Laetitia Gibart, Gilles Bernot, H

ene Collavizza and Jean-Paul Comet

University C

ote d’Azur, I3S Laboratory, UMR CNRS 7271, CS 40121, 06903 Sophia Antipolis Cedex, France

Keywords:

Formal Methods, Regulatory Networks, R. Thomas’ Modelling, Parameter Identiﬁcation, Model Enrichment.

Abstract:

When designing a biological regulatory network, new information or wet experiments can require adding

variables or interactions, inside a previously validated model. They can result in complete reconsiderations

of established behaviours. Fortunately, formal methods allow for fully automated veriﬁcation of properties,

and TotemBioNet is an efﬁcient software integrating a collection of formal approaches for regulatory net-

works. It allowed us to develop a multidisciplinary methodology for designing large dynamical models in an

incremental way, including non regression proofs (preservation of important biological properties).

1 INTRODUCTION

In the ’70s, qualitative modelling of biological regula-

tory networks (Thomas, 1973; Glass, 1975) has led to

signiﬁcant advances in the understanding of the main

causalities of some observed cell behaviours. Models

were handmade, and parameter identiﬁcation mainly

used simulations (Gonzalez et al., 2006). In the early

2000s, formal methods automated the identiﬁcation

of parameters, managing as a whole the exhaustive

sets of suited parameter settings (Bernot et al., 2004).

Software platforms handling formal methods (Khalis

et al., 2009; Batt et al., 2004; Paulev

e, 2017) have

made it possible to design regulatory graphs, where

numerous parameter values are compatible with the

biological knowledge, where simulations are help-

less, and where the slightest modiﬁcation can have

huge consequences (Gebser et al., 2010; Khoodeeram

et al., 2017; Naldi et al., 2018).

Enrichments of models do not preserve previ-

ously established properties (Bernot and Tahi, 2009;

Siebert, 2009; Mabrouki et al., 2011). Nonethe-

less, in practice, the only way to design big regula-

tory models is to enrich previously validated mod-

els. So, we are moving into an era where intensive

veriﬁcations of formal properties play the role that

simulations played for handmade models: Model re-

vision becomes the current practice (Gouveia et al.,

2018). We describe here a methodology to efﬁ-

ciently design large validated models by successive

small enrichments of previously validated ones. It

uses TotemBioNet, a software that offers opti-

mized management of the exhaustive set of parameter

values compatible with biological knowledge (Boyen-

val et al., 2020). Here, we describe our methodology

and we give a (very small but representative) example

of enrichment.It starts from one of the largest regu-

latory network model using the multivalued Thomas

framework (Snoussi, 1989), namely the generic reg-

ulation of the cell metabolism (Khoodeeram et al.,

2017). It abstracts the relationships between the main

actors of the metabolism in such a way that, for ex-

ample, the Crabtree/Warburg effects can be studied,

and their main causalities understood. The model

considers biomass as a whole, without distinguishing

between nucleic acids, proteins or lipids. Here, we

simply show how to distinguish between lipidic and

non lipidic biomass, so that the enrichment example is

small enough to focus on the TotemBioNet method

rather than describing detailed biological regulations.

Section 2 reminds the Thomas approach and

the formal methods implemented in TotemBioNet.

Section 3 presents our methodology, including the

role of different kinds of behaviour “property matri-

ces” that greatly aid decision making for model de-

sign. Section 4 shows the illustrating example.

2 TotemBioNet RATIONALS

TotemBioNet assists the design of R. Thomas’

qualitative regulatory network models. It handles all

the possible parameterizations, which is the key point

of the modelling process.

Gibart, L., Bernot, G., Collavizza, H. and Comet, J.

TotemBioNet Enrichment Methodology: Application to the Qualitative Regulatory Network of the Cell Metabolism.

DOI: 10.5220/0010186200850092

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 3: BIOINFORMATICS, pages 85-92

ISBN: 978-989-758-490-9

2.1 Thomas’ Modelling Framework

A regulatory network is a labelled directed graph in

which vertices are variables (represented by circles)

or multiplexes (rectangles). Variables abstract bio-

logical entities (gene products, biological functions

or metabolic pathways) and multiplexes contain for-

mulas that encode situations where a group of vari-

ables (inputs of multiplexes) inﬂuence the evolution

of some variables (outputs of multiplexes), the sim-

plest multiplexes being the formations of complexes

(the formula contains a simple conjunction). In Fig. 1

the multiplex PBM abstracts the biomass production,

it targets variables BM (BioMass) and ATP. It is ac-

tivated when at least one of the multiplexes AAS, LS

or PPP is activated. Among others, AAS abstracts the

amino acid synthesis process. It summarizes the ele-

ments necessary for the de novo production of amino

acids (Berg et al., 2002): nitrogen and carbon donors

(DNC > 2), a large amount of NADH (NADH > 2)

and ATP (ATP > 1) (necessary for these anabolic pro-

cesses). This can be expressed by the following for-

mula: (DNC > 2)&(AT P > 1)&(NADH > 2).

Figure 1: A part of inﬂuence graph: representation of vari-

ables and multiplexes (focus on BM in Figure 5).

Each variable has a domain of variation made of a

small number of discrete levels and the dynamics of

the model is deduced from the regulatory network and

a set of kinetic parameters K

...

(simply “parameters”

in the sequel). Formally, the presence of an activator

or the absence of an inhibitor of a variable x is called

a resource, and the exhaustive list ω of its resources

deﬁnes the discrete level towards which x is temporar-

ily attracted, denoted K

x,ω

. If x has p possible re-

sources then it has 2

parameters. Lastly, when the

set of effective resources of a variable increases, this

cannot lead to decreasing its level (Snoussi, 1989):

ω ⊂ ω

⇒ K

x,ω

6 K

x,ω

, which can be used to reduce

the number of parametrisations to consider.

2.2 Selection Technics

TotemBioNet inherits from SMBioNet (Bernot

et al., 2004) and combines two ﬁltering approaches to

identify the exhaustive set of models consistent with

biological knowledge: “genetically modiﬁed Hoare

logic” and temporal logic.

Hoare Logic. Biological experiments prove that

a set of traces must exist in the model. The ge-

netically modiﬁed Hoare logic (Bernot et al., 2019;

Folschette, 2019) produces the constraint on param-

eters that characterizes the models in which these

traces exist. We ﬁrstly transcript the observed ex-

perimental traces into a so-called Hoare triple (ob-

served pre-condition, path and post-condition) and

Hoare logic constructs the weakest pre-condition that

must be fulﬁlled. This constraint throws away param-

eter values which do not enable the observed traces.

Temporal Logic. Most of biological knowledge

do not translate directly into Hoare logic, such as epi-

genetic phenomena, homeostasis, (non-)reachability

of certain states, some events that always happen af-

ter others (but not necessarily right after), etc. To for-

malize this general knowledge, it is preferable to use

temporal logic. (Bernot et al., 2004; Goldfeder and

Kugler, 2019) and several other authors have chosen

CTL due to the efﬁciency of its model checking algo-

rithm, and its ability to capture the non-deterministic

choices of Thomas’ theory.

Mixing Both Approaches. TotemBioNet in-

puts are: an inﬂuence graph, any knowledge on the

parameter values, and properties on the dynamics of

the system expressed using CTL or Hoare logic. It

ﬁrst computes the weakest pre-condition (wp) us-

ing genetically modiﬁed Hoare logic. Then, it ef-

ﬁciently enumerates all parametrisations satisfying

wp, and generates input ﬁles for the model checker

NuSMV (Cimatti et al., 2002). Each ﬁle contains the

conjunction of CTL formulas and an automaton that

encodes the model for the considered parametrisation.

TotemBioNet

, comes with many examples, in-

cluding those of the present paper. It allows one to

describe the inﬂuence graph with yEd graph editor

A typical session consists in building the inﬂuence

graph using yEd, in automatically generating the as-

sociated input ﬁle and then adding temporal proper-

ties and Hoare triples. TotemBioNet generates all

parametrisations, labelled with “OK” when the dy-

namic properties are veriﬁed, and if not with all the

properties which are not satisﬁed (see Fig. 2.).

https://gitlab.com/totembionet/totembionet

https://www.yworks.com/products/yed

BIOINFORMATICS 2021 - 12th International Conference on Bioinformatics Models, Methods and Algorithms

Behavioural

properties

{Pre} x+;y−;... {Post}

Interaction graph

G PC

0 1

Set of dynamics satisfying the specifications

Hoare Logic

Construction of

Model Checking

temporal logic formula (via NuSMV)

Selection of dynamics which satisfy the

temporal

formula

of parameterizations

Enumerator

R. Thomas’ modeling framework

Φ: a temporal

logic formula

G ≥ 1

¬(PC≥1 ∧¬(L≥1))

yes

simple

contraints

contraints wp

|= Φ)

|= wp

Figure 2: TotemBioNet processing ﬂow (Boyenval et al.,

2020).

3 TotemBioNet ENRICHMENT

FRAMEWORK

3.1 Modelling Context

We proceed by successive small enrichments of the

model under design. At each step, we start from a

regulatory network model together with a set of pa-

rameter settings (non empty but of reasonable size)

that is considered relevant w.r.t. biological knowledge

and expertise. For the new enrichment, it is essen-

tial to keep track of the fundamental properties of the

previously designed network models and to guarantee

that they are preserved (possibly reformulated or en-

riched). Our TotemBioNet enrichment framework

relies on a Modelling context, formed by:

IG: An Inﬂuence Graph that describes the inﬂu-

ence of each variable and multiplex in the system,

BK: Some Biological Knowledge on the dynamics

of the system, collected from biologists and literature,

and given in natural language,

PM: A Property Matrix that formalizes, (CTL or

Hoare logic) the dynamics under different conditions.

These formalisations are a translation of BK,

PS: A set of Parameter Settings that govern sys-

tem’s dynamics. Some of them may be identiﬁed

from BK by the modeller, others are automatically se-

lected by TotemBioNet to satisfy PM formulas.

The couple (BK,PM) is the cornerstone for the re-

liability of the enrichment process: all along this pro-

cess, PM is updated according to BK, and formal ver-

iﬁcation methods are applied to PM.

3.2 Property Matrices

Each column of a Property Matrix (Fig. 3-top) is a

variable

whose behaviours are described in the col-

umn. Each row is an experimental condition and a

box in the matrix can be ﬁlled by a conjunction of

CTL formulas and Hoare triples that formalizes BK.

The experimental conditions are ﬁxed via a set of en-

vironment variables which are sources of regulations

but never target of a regulation. Their constant values

can differ from one experimental condition to another.

Along our enrichment workﬂow, Property Matri-

ces (PM) have two different usages, validation or pre-

diction, as detailed below:

(1)Validation Property Matrices (Fig. 3-middle) are

PM where the biologist has certainty on the proper-

ties, via literature or wet experiments, and the mod-

eller is conﬁdent in the translation into logical formu-

las. When an enrichment is done, all the properties of

the enriched version of a Validation PM must be sat-

isﬁed. By analogy with software engineering, a Vali-

dation PM can be seen as a set of non regression tests,

(2)Prediction Property Matrices (Fig. 3-bottom) fo-

cus on ﬁxed parameter settings. They are PM whose

properties have been automatically generated and val-

idated with TotemBioNet from simple property

patterns such as oscillations, homeostasis, attraction

basins and so on. There are as many prediction prop-

erty matrices as parameter settings: a subset of se-

lected prediction property matrices can then become

the starting point of a new enrichment process.

Fig. 3 shows a schematic vision of the usage of

PMs along an enrichment of a model, where some

variables have been added/removed. A Validation PM

may be sparse, as the set of established biological

properties is generally small. At the opposite, a Pre-

diction PM is entirely ﬁlled, but some of the discov-

ered properties can be of little relevance. This is why

the multidisciplinary dialogue is crucial and the in-

crements of BK are often suggested by the prediction

PMs, possibly asking for new biological experiments.

3.3 Enrichment Workﬂow

Initial Model. The TotemBioNet enrichment

workﬂow starts with an initial modelling context (top

of Fig. 4). It contains a Validation PM that will be en-

riched. The challenge is to design an enriched mod-

elling context that ensures non regression w.r.t. the

initial validation matrix.

Enrichment Design. The design of an enriched

modelling context lies on a dialogue between the biol-

ogist and the modeller (Fig. 4-middle). The biologist

Or possibly a few variables

TotemBioNet Enrichment Methodology: Application to the Qualitative Regulatory Network of the Cell Metabolism

Figure 3: Schematic vision of Property Matrices (PM).

Top: initial validation PM. Middle: enriched validation PM

where var-3 has been divided into var-3A and var-3B lead-

ing to the generation of purple properties prop-3A-1 and

prop-3B-1 from the initial property prop-3-1. Bottom:

prediction PM where yellow properties have been automat-

ically generated.

describes the Enriched Biological Knowledge (EBK)

and the modeller propagates it by enriching the Inﬂu-

ence Graph (IG) or the Property Matrix (PM) (they

become EIG and EPM). A modiﬁcation of IG may

induce a modiﬁcation of the validation matrix, for ex-

ample by changing the level of an homeostasis. It may

also induce new parameters, thus TotemBioNet is

run to identify the unknown parameters that validate

EPM. If there is no such parameters, the modelling

is inconsistent and has to be corrected. If there are

too many parameter settings, the modelling is not pre-

cise enough to be a good candidate for predicting new

relevant biological properties (Occam’s razor). Thus

the process goes back to a dialogue for reﬁning the

model. When a reasonably small number of parame-

ter settings validate EPM, the biologist takes the ﬁnal

decision: return to improve the models and design a

better enrichment, or continue the workﬂow.

Test and Validation of an Enriched Modelling

Context. When a set of parameter settings has been

selected as the best compromise, tests and validations

begin (Fig. 4-bottom). An instanciated modelling

context is then built for each parameter setting. The

dynamic behaviours controlled by these settings en-

sure a non regression of the initial modelling, but they

may also exhibit some behavioural properties that dif-

fer from one model to another. Thus the matrix com-

pletion functionality of TotemBioNet is called on

each instantiated model. TotemBioNet establishes

a set of properties on each variable in each context

which are added to the matrix, as predictions. Lastly,

the biologist and the modeller select the most interest-

ing contexts according to phenotypes, possibly with

the help of new biological experiments. This set of

contexts is the starting point of another enrichment.

4 ILLUSTRATING EXAMPLE

This section illustrates the enrichment steps of Fig 4.

The cell metabolism can be viewed as an energy bal-

ance where all metabolic pathways (linked series of

chemical reactions) are regulated according to nu-

trient cell intake. They can be anabolic (produc-

ing biomass compounds) and catabolic (that degrade

large compounds to produce energy). The anabolic

and catabolic balance is ﬁne-tuned in the cell.

Within this context, an abstract and qualita-

tive model of the metabolism regulation was devel-

oped (Khoodeeram et al., 2017) to understand how

activity level of biological pathways in combination

with nutrient can inﬂuence Warburg/Crabtree effect.

This effect appears in all cell types from yeast to hu-

man: It leads the cells to go from one catabolic pro-

cess (respiration) to another (fermentation). Since

2017, in order to focus on cancer in human cells in

correlation with the cell cycle, we made several en-

richments of the model to better reﬂect key metabolic

pathways (glycolysis, oxidative respiration, Krebs cy-

cle), nutrient (oxygen or glucose), key molecular

components (biomass) and regulators (key-cofactors).

The following subsections correspond respec-

tively to the top, middle and bottom parts of Fig. 4.

4.1 Initial Model and its Limits

The enrichment starts from the model whose IG is

given in Fig. 5: biomass was represented by a unique

variable BM that covered all cell growth components

(lipids, nucleic acids for genetic material and amino

acids for proteins). Fig. 1 is an extract from Fig. 5 that

focuses on BM: the multiplexes PPP, AAS and LS ab-

stract anabolic pathways respectively producing nu-

cleic acids, amino acids and lipids (Axelrod, 1967;

Berg et al., 2002; Stein and Stein, 1967). If one of

them is activated, biomass production (PBM) is on:

PBM is the inclusive disjunction of PPP and AAS and

Currently, only oscillations, homeostasis, and stable

states are implemented in prediction property matrices but

enriching this range of properties is not difﬁcult.

BIOINFORMATICS 2021 - 12th International Conference on Bioinformatics Models, Methods and Algorithms

Figure 4: TotemBioNet Enrichment Framework. The quadruplet of a modelling context is represented by a box containing:

a yellow trapeze (IG), a blue bubble (BK), a grey rectangle (PM) and a green parallelogram (PS). The preﬁx E means an

Enrichment of one part of the quadruplet.

LS (AAS|PPP|LS). When PBM is a resource of BM,

biomass will become present in the cell

Table 1: Subpart of the Initial validation matrix. “0” means

tend toward 0, “Osc” means oscillations with arbitrary

boundaries and “!” stands for the negation.

Initial validation matrix

Environment Behaviour

OXYG GLC IN AA ATP BM FERM KREBS PHOX

0 0 0 0 0 0 0 0

0 1 0 Osc !0 0

1 1 1 Osc !0 0 Osc Osc

1 2 2 Osc !0 !0 Osc

Table 1 is a subpart of the validation matrix of this

initial model. The column BM, for example, requires

obvious properties, e.g. absence of biomass if the cell

does not receive any nutrient, and so on.

Because the applicable parameter is K

BM,PBM

= 1 in

the current parameter setting

BM participates in Box, which abstracts the β-

oxidation pathway, referring to the catabolic process

of lipids (Wakil, 1970). Unfortunately, this IG implies

that β-oxidation could be activated without any lipids

input. Indeed Box is satisﬁed when BM is activated

by PPP (nucleic acids) or AAS (amino acids) through

PBM: this behaviour proves that considering biomass

as a whole is a too coarse abstraction for an adapta-

tion to human cells. So, separating lipid biomass from

the other compounds is our next enrichment.

4.2 Enriched Modelling Context Design

EBK: Enriched Biological Knowledge. It lists all

biological information needed to separate lipids from

the rest of the biomass: lipids are produced by syn-

thesis and degraded by β-oxidation. Amino acids syn-

thesis produces proteins, and PPP (Pentose Phosphate

Pathway) produces nucleic acids for genetic material.

TotemBioNet Enrichment Methodology: Application to the Qualitative Regulatory Network of the Cell Metabolism

Figure 5: Inﬂuence graph (IG) of the initial modelling context: variables in red are the main metabolic processes (glycolysis,

Krebs cycle, oxydative phosphorylation), yellow ones are the main cofactors (ATP/ADP and NADH/NAD+ ratios, oxygen,

amino acid donors), biomass is in blue, environment variables are white (oxygen intake, glucose, nutrients...) and rectangle

multiplexes formalize their interactions. For more details see (Khoodeeram et al., 2017) who deﬁned the ﬁrst IG of the cell

metabolism regulation and its whole parameter setting.

Each kind of biomass production needs ATP for en-

ergy. All of this is added to initial BK to obtain EBK.

EIG: Enriched Inﬂuence Graph. Accordingly,

we modify Fig. 1 into Fig. 6. It focuses on what has

been modiﬁed: the biomass variable and its surround-

ing multiplexes. In the EIG (Fig. 6) BM has been split

into BL (lipids biomass) and BnL (non lipidic com-

pounds). BL participates in Box and is regulated by

LS. BnL resources are PPP and AAS. In the multiplex

Box the variable BM is replaced by the variable BL,

which solves the problem mentioned in Subsection

4.1. Lastly, both BL and BnL productions consume

ATP, thus regulate it negatively.

EPS: Enriched Parameter Settings. Our EIG

leads to new parameters that need to be identiﬁed. BL

and BnL have two resources each, so there are (2

= 8) new parameters (see Section 2.1). Moreover

ATP gains one resource (5 resources in the EIG in-

Figure 6: Enrichment of the inﬂuence graph of Fig. 1: BM

has been split into Bnl and BL.

stead of 4 in IG). It increases the number of ATP pa-

rameters to 2

= 32. Among them, 16 are ineffective

due to unsatisﬁability of some resource combinations,

and the 6 identiﬁed parameters which do not involve

new resources are kept. Finally, it remains 18 parame-

BIOINFORMATICS 2021 - 12th International Conference on Bioinformatics Models, Methods and Algorithms

ters to identify for ATP. Thus, we have 26 parameters

to identify.

Thought experiments can be useful. For instance, to

identify the parameter K

BnL,∅

, we assume that both

pentose phosphate pathway (PPP) and amino acids

synthesis (AAS) are inactive (the set of Bnl resources

is ∅). We try to foresee toward which value would

BnL go if these conditions persist: neither genetic

material nor proteins are produced, thus K

BnL,∅

= 0.

Most of the time EBK is not sufﬁcient to allow the

parameter identiﬁcations (many thought experiments

are inconclusive). In such cases TotemBioNet will

test all possible parameter settings. All in all 5 out of

the 26 parameters were not identiﬁed.

EPM: Validation Enriched Properties Matrix.

It summarizes the dynamical behaviours of variables,

according to EBK. The only modiﬁcations affect BM

(Table 2). It is replaced by both columns BnL and BL

and they are ﬁlled from EBK: For example without

any intake of glucose, no compounds are synthesized

so BnL and BL tend towards 0.

Table 2: Subpart of the validation EPM.

Enriched validation matrix

Environment Behaviour

OXYG GLC IN AA ATP BnL BL FERM KREBS PHOX

0 0 0 0 0 0 0 0 0

0 1 0 Osc !0 0

1 1 1 Osc !0 !0 0 Osc Osc

1 2 2 Osc !0 !0 !0 Osc

4.3 TotemBioNet Test and Validation

We test with TotemBioNet the enriched modelling

context created above, taking into consideration EBK,

EIG, EPS, EPM. The 5 unidentiﬁed parameters from

EPS give rise to only 15 instantiated modelling con-

texts (where each parameter has a unique value). We

obtain 15 prediction matrices such as Table 3 where

all empty boxes of validation-EPM are automatically

completed using TotemBioNet. They allow mod-

ellers and biologists to eliminate parameter settings

which exhibit non credible behaviours.

Table 3: Sub part of one of the 15 prediction EPMs.“Osc(0-

1)” means oscillations between level 0 and 1 of the variable.

Enriched Prediction matrix

Environment Behaviour

OXYG GLC IN AA ATP BnL BL FERM KREBS PHOX

0 0 0 0 0 0 0 0 0

0 1 0 Osc(0-2) Osc(0-1) 0 Osc(0-1) 0 0

1 1 1 Osc(0-2) Osc(0-1) Osc(0-1) 0 Osc(0-1) Osc(0-1)

1 2 2 Osc(0-2) Osc(0-1) Osc(0-1) Osc(0-1) Osc(0-2) Osc(0-1)

For our example, the 15 prediction EPMs exhibit

interestingly different behaviour for the variables BL

and BnL. Here, none of the models have aberrant pre-

dictions, so we keep this enriched modelling context

for the next enrichment step. Nevertheless, in gen-

eral some of these different behaviours can contradict

established biological knowledge or suggest new bio-

logical experiments.

5 CONCLUSION

We deﬁned an incremental methodology for develop-

ing large formal models of biological regulatory net-

works. Starting from an initial formal model, the en-

richment is ﬁrst described in natural language EBK,

then the formalisation of this information leads to

the EIG, EPS and validation EPM. Our process in-

volves, at each round of this methodology, a manage-

able number of parameter settings.

In practice, we observe that after a few enrichment

steps, the ab initio design of the enriched model, with-

out using the proposed methodology, would be im-

possible. The example outlined in this article (that

contains notably more than 100 parameters) would

lead to a number of parametrizations (1.45 × 10

)

that is so huge that TotemBioNet or any similar

platform would have taken more that 4.6 × 10

years

of computation to model check them, at the rate of

100 proofs per second! One of the key points of our

incremental methodology consists in limiting, at each

enrichment step, the number of parameter settings to

enumerate, preserving a maximum number of known

parameters from the previously validated model. EPS

and validation EPM in an enriched modelling context

allow TotemBioNet to drastically reduce the num-

ber of total parameters to identify, with a negligible

computation time: The parameters for which the re-

sources do not change (according to the new interpre-

tation) remain at the same value, and the new ones can

partly be identiﬁed by thought experiments.

Moreover, following our methodology, getting too

many parameter settings for an enriched modelling

context, simply means that the enrichment step is too

ambitious to be manageable. In such a case we re-

assess the informal enrichment EBK to address a less

ambitious enrichment. Conversely, when no coherent

parameter setting is obtained, this is generally due to

a construction error in the enriched modelling context

which has also to be reassessed. In our methodology,

this loop of reassessment is made possible owing to

the remarkably efﬁcient TotemBioNet platform.

We currently use this methodology for the

metabolism regulation in cancer cells, when cells

acquire new functioning during the Epithelial-

mesenchymal transition. In practice, the informal BK

plays a crucial role for assisting interactions between

biologists and modellers, not only in the design of the

enrichment but also, in order to keep track of interpre-

tations made in previous contexts. Last but not least,

TotemBioNet Enrichment Methodology: Application to the Qualitative Regulatory Network of the Cell Metabolism

at each step the choice of a sensible subset of parame-

ter settings often suggest new biological experiments

that are particularly revealing.

ACKNOWLEDGEMENTS

We are fully indebted to Sophie Vasseur and Fabi-

enne Marchai-Guillaumond from INSERM CRCM,

for fruitful collaborations which helped us to design

the methodology. We warmly thank Jean-Yves Tros-

set and Rajeev Khoodeeram without whom the initial

model of metabolism regulation would not exist.

REFERENCES

Axelrod, B. (1967). Chapter 3 - glycolysis. In Greenberg,

D. M., editor, Metabolic Pathways (Third Edition),

pages 112–145. Academic Press.

Batt, G. et al. (2004). Model checking genetic regulatory

networks using GNA and CADP. In Graf, S. and

Mounier, L., editors, Model Checking Software, pages

158–163. Springer Berlin Heidelberg.

Berg, J. M., Tymoczko, J. L., and Stryer, L. (2002). The

biosynthesis of amino acids.

Bernot, G., Comet, J.-P., Khalis, Z., Richard, A., and Roux,

O. F. (2019). A genetically modiﬁed Hoare logic. The-

oretical Computer Science, 765:145–157.

Bernot, G., Comet, J.-P., Richard, A., and Guespin,

J. (2004). Application of formal methods to bi-

ological regulatory networks: extending Thomas’

asynchronous logical approach with temporal logic.

J.Theor.Biol, 229(3):339 – 347.

Bernot, G. and Tahi, F. (2009). Behaviour preservation of a

biological regulatory network when embedded into a

larger network. Fundam. Inform., 91(3-4):463–485.

Boyenval, D., Bernot, G., Collavizza, H., and Comet, J.-P.

(2020). What is a cell cycle checkpoint? the Totem-

BioNet answer. In CMSB, pages 362–372.

Cimatti, A. et al. (2002). NuSMV 2: An opensource tool

for symbolic model checking. In CAV’02, pages 359–

364, London, UK, UK. Springer-Verlag.

Folschette, M. (2019). The Hoare-fol Tool. Technical re-

port, CNRS, Centrale Lille, UMR 9189, France.

Gebser, M. et al. (2010). Repair and prediction (under in-

consistency) in large biological networks with answer

set programming. In 12th Intl Conf. on the Principles

of Knowledge Representation and Reasoning.

Glass, L. (1975). Classiﬁcation of biological networks by

their qualitative dynamics. J.T.B, 54(1):85–107.

Goldfeder, J. and Kugler, H. (2019). BRE:IN - A Back-

end for Reasoning About Interaction Networks with

Temporal Logic. In CMSB, LNCS, pages 289–295.

Springer International Publishing.

Gonzalez, A. G., Naldi, A., Sanchez, L., Thieffry, D., and

Chaouiya, C. (2006). GINsim: a software suite for

the qualitative modelling, simulation and analysis of

regulatory networks. Biosystems, 84(2):91–100.

Gouveia, F., Lynce, I., and Monteiro, P. T. (2018).

Model Revision of Logical Regulatory Networks Us-

ing Logic-Based Tools. In Technical Communications

of the 34th ICLP 2018, pages 23:1–23:10.

Khalis, Z., Comet, J.-P., Richard, A., and Bernot, G. (2009).

The SMBioNet method for discovering models of

gene regulatory networks. G.G.G, 3:15–22.

Khoodeeram, R., Bernot, G., and Trosset, J.-Y. (2017). An

Ockham razor model of energy metabolism. advances

in Systems and Synthetic Biology, pages 81–101.

Mabrouki, M., Aiguier, M., Comet, J.-P., Le Gall, P., and

Richard, A. (2011). Embedding of biological regula-

tory networks and property preservation. Mathematics

in Computer Science, 5(3):263–288.

Naldi, A. et al. (2018). The CoLoMoTo interactive

notebook: accessible and reproducible computational

analyses for qualitative biological networks. Frontiers

in physiology, 9:680.

Paulev

e, L. (2017). Pint: A static analyzer for transient dy-

namics of qualitative networks with ipython interface.

In CMSB, pages 309–316. Springer.

Siebert, H. (2009). Deriving behavior of boolean bioregu-

latory networks from subnetwork dynamics. Mathe-

matics in Computer Science, 2(3):421–442.

Snoussi, E. H. (1989). Qualitative dynamics of piecewise-

linear differential equations: a discrete mapping ap-

proach. Dynamics and stability of Systems, 4(3-

4):565–583.

Stein, O. and Stein, Y. (1967). Lipid synthesis, intracel-

lular transport, storage, and secretion. J. Cell. Biol.,

33(2):319–339.

Thomas, R. (1973). Boolean formalization of genetic con-

trol circuits. J.Theor.Biol, 42(3):563–585.

Wakil, S. J. (1970). Chapter I - fatty acid metabolism. In

Wakil, S. J., editor, Lipid Metabolism, pages 1–48.

Academic Press.

BIOINFORMATICS 2021 - 12th International Conference on Bioinformatics Models, Methods and Algorithms