The 3rd Franco-Japanese Symposium
on Knowledge Discovery in Systems Biology (FJ'09)
Abstracts

======================================================================
"Inferring Rules and Facts by Meta-level Abduction on SOLAR"
Katsumi Inoue and Hidetomo Nabeshima
======================================================================
ABSTRACT:
This work addresses discovery of unknown relations from
incomplete network data using abduction.
Given a network information such as causal relations and
metabolic pathways, we want to infer missing links and nodes
in the network to account for observations.
To this end, we introduce a framework of meta-level abduction,
which is implemented in SOLAR using a first-order representation
for algebraic properties of causality and full-clausal form of
network information and constraints. Meta-level abduction by
SOLAR is powerful enough to infer missing rules and facts.
Inferring unknown causes is also realized by predicate invention
in the form of existentially quantified hypotheses, in which
new variables represent unknown nodes to be added.



======================================================================
"Efficient Equational Consequence Finding Calculus on SOL with Ordering Constraints"
Koji Iwanuma and Katsumi Inoue
======================================================================
In this talk, we study some issues of equational consequence finding problems. Equational inference is an essentially important not only in traditional theorem proving but also for consequence finding. Efficient inference of equality, however, is a substantially difficult problem, so huge amounts of researches have been conducted so far. Among of them, some ordering constraint methods clearly succeeded in improving the inference performance for equality. SOL is a well-known efficient calculus for consequence finding over non-equational problems . In this talk, we introduce into SOL a novel mechanism for efficient equational consequence finding. The proposed mechanism is based both on Bachmair et. al's modification method and Paskevich's connection tableaux for equality using ordering constraints.


===========================================================================
"Applying a Declarative Constraint-based Method to the Re-examination of a Discrete Genetic Regulatory Network"
L.Trilling , joint work with F. Corblin, S. Tripodi, D. Ropers and E. Fanchon.
==========================================================================
ABSTRACT:
Dynamical modeling has proven useful for understanding how complex biological processes emerge from the many components and interactions composing genetic regulatory networks (GRNs). However, the development of models is hampered by large uncertainties in both the network structure and parameter values. To remedy this problem, the models are usually developed through an iterative process based on numerous simulations, confronting model predictions with experimental data and refining the model structure and/or parameter values to repair the inconsistencies. We will present an alternative to this generate-and-test approach through a four-step method for the systematic construction and analysis of discrete models of GRNs by means of a declarative approach. Instead of instantiating the models as in classical modeling approaches, the biological knowledge on the network structure and its dynamics is formulated in the form of constraints. The compatibility of the network structure with the constraints is queried and in case of inconsistencies, some constraints are relaxed. Common properties of the consistent models are then analyzed by means of logical languages that we will introduce. Removing questionable constraints or adding interesting ones allows to further analyze the models. We will illustrate the feasibility of our approach by applying it to the re-examination of a model describing the nutritional stress response in the bacterium Escherichia coli. (see Corblin, F., et al., A declarative constraint-based method for analyzing discrete genetic regulatory network. BioSystems (2009), doi:10.1016/j.biosystems.2009.07.007)


======================================================================
"Toward an Efficient Consequence Synthesis in SOL Tableaux"
Hidetomo Nabeshima, Koji Iwanuma and Katsumi Inoue
======================================================================
TBA


==========================================================================
"Non-monotonic Reasoning and Abduction"
Pierre Siegel
===========================================================================
ABSTRACT:
Classical logic has the monotonic property: any information deductible from a knowledge C, are deductible from C if C is increased. On other hand, an essential component of the intelligence is to take into account incomplete information (uncertain information, revisable information). For nonmonotonic logics a fact deductible from C (from knowledge, from belief) is not true but only probable in the sense that it can be invalidated or revised when adding new information. The Abductive reasonning is close from nonmonotonic logics. For example, most of proof procedures for these logics are using abductive algorihms. In this presentation, we present some links between monmonotonic logics (default logic, preferential logic, hypothesis theory, X-logic) and abduction.


======================================================================
"Protocols for Multi-agent Diagnosis using SOLAR"
Gauvain Bourgne and Katsumi Inoue
======================================================================
ABSTRACT:
TBA


======================================================================
"An Incremental Way for Finding Characteristic Hypotheses in CF-induction"
Yoshitaka Yamamoto, Katsumi Inoue and Koji Iwanuma
======================================================================
ABSTRACT:
CF-induction is one of the hypothesis-finding procedures in Inductive Logic Programming (ILP).
Compared with the other ILP systems, CF-induction has several theoretical features like completeness for finding hypotheses. However, It is still far from our goal that the hypotheses which we wish to obtain can be automatically computed.
One problem is that CF-induction consists of several non-deterministic procedures, each of which needs some selections of users by hand.
In this talk, we first present an incremental way for finding hypotheses which enables us to dramatically reduce the non-determinisms in CF-induction. We implemented CF-induction with this way by JAVA. We then show the performance of the current version of this implementation.


======================================================================
"An Implementation of a Model-based Abduction and its Application to Systems Biology"
Takehide Soh and Katsumi Inoue
======================================================================
ABSTRACT:
Several studies have applied abduction to systems biology such as predicting gene expression regulation, predicting the effects of a toxin and completing a model of metabolic pathways. In those approaches, abduction is achieved with resolution-based proof procedure and its computation time tends to be large because intended organisms often contain a lot of entities to be considered. To overcome this problem, we introduce a model-based abduction by SAT solver. Recent advance of satisfiability (SAT) technologies has been tremendous and several problems such as planning, scheduling and packing problems are successfully solved with SAT solvers. In our approach, we compute the set of all Herbrand models of the given background knowledge and observation. Then we choose correct hypotheses from the models. In this talk, we show this procedure through simple example and introduce its application to systems biology.


==========================================================================
"Robustness of Threshold Boolean Automata Networks to Iteration modes. Application to Genetic Regulation Networks Modelling"
Adrien Elena
==========================================================================
ABSTRACT:
We study the influence of a change of iteration mode on the attractors of a threshold boolean automata network. These networks are discrete mathematical object classicaly used to model biological regulation systems. The aim is to underline the importance of the choice of iteration mode for the dynamics of these networks, especially for the observed limit cycles. First we run simulations of the dynamics of unbiased samples of networks with size varying from one node to seven nodes. The result of these simulations show that as the size of these networks increases their dynamics becomes more and more sensitive to the choice of iteration mode. Next we show a theoretical result that links the observed limit cycles for a network iterated with parallel mode and other iteration modes with limit cycles.


======================================================================
"An EM Algorithm Spetialized for Shared BDDs"
Ishihata Masakazu, Kameya Yoshitaka and Sato Taisuke
======================================================================
Abstract:
The BDD-EM algorithm, which is an expectations-maximization (EM) algorithm working on binary decision diagrams (BDDs), is used to learn probabilities of statistical models described by Boolean formulas. We extend the BDD-EM algorithm to be able to deal with shared BDDs (SBDDs). A SBDD represents multiple BDDs as a single graph, and it uses less memory than the sum of memories required for each of the BDDs when they have common sub-graphs. We apply the extended algorithm to a task to predict intermittent faults which probabilistically occur in logic circuits. We show execution both time and space are less consumed than ones of the BDD-EM algorithm, and also show the reduction rate increases with the number of observations.


==========================================================================
"Role of the Rho (A,B,C) GTPases in Cancer Progression"
G. Favre and J-C. Faye
==========================================================================
ABSTRACT:
The Rho GTPases proteins are members of a large superfamily of regulatory proteins whose activities are controlled by regulated GDP/GTP cycling. To date, a total of 20 Rho family members have been suggested by the data available from the human genome sequence project. Rho GTPases control a wide variety of signal transduction pathways regulating many fundamental processes of cell biology, such as organization of the actin cytoskeleton, gene expression, cell proliferation and survival. Rho proteins are also implicated in participating in several steps of tumor progression and development of metastasis.. They function in cell cycle regulation by the modulation of cyclin D1 and by their involvement in endocytic traffic, such as in regulation of epidermal growth factor receptor. Furthermore, Lacal et al have shown that Rho GTPases are directly involved in signalling pathways that trigger either proliferation or cell death. These studies linking Rho proteins to many aspects of cellular proliferation are further extended by the study by Gomez del Pulgar, which revealed that several human tumors contained aberrant expression and activation of Rho GTPases. Elevated expression of RhoA and RhoC was found in breast, lung, ovarian, gastric, and bladder cancers. The involvement of RhoA in testicular human tumors was demonstrated by increased RhoA mRNA levels in relation to tumour grade. Overexpression of the rhoC gene in adenocarcinoma of pancreas correlated with poorer prognosis of patients, whereas we have shown that RhoB expression is lost in several tumors. To identify proteins involved in the signal transduction pathways of these three Rho GTPases, is an important challenge in the knowledge of their various biological properties, and would open out onto new cancer treatments.


======================================================================
"A Comparison Among Unsupervised Discretization Methods for Time Series Data"
Yoshitaka Kameya
======================================================================
Abstract:
Discretization is a key preprocessing step in symbolic data mining, e.g. Apriori-style frequent pattern mining or inductive logic programming, from real-valued data. This talk focuses on unsupervised discretization for time series data. Many of the existing methods (SAX, Persist, hidden Markov models and so on) work under a common strategy, in which we perform smoothing at the time axis, and binning/clustering at the measurement axis. We review these existing methods and report some results of comparative experiments on artificial and real data used in previous work.


===========================================================================
"Comparative Analysis of the Model Based Diagnosis Approach From the Articial Intelligence and Automatic Control Perspectives"
Louise Trave-Massuyes
==========================================================================
ABSTRACT:
Two distinct and parallel research communities have been working along the lines of the model-based diagnosis approach: the fault detection and isolation (FDI) community and the diagnostic (DX) community that have evolved in the elds of automatic control and articial intelligence, respectively. This presentation claries and links the concepts and assumptions that underlie the FDI analytical redundancy approach and the DX con- sistency-based logical approach. A formal framework is proposed in order to compare the two approaches and the theoretical proof of their equivalence together with the necessary and sufcient conditions is provided.


======================================================================
"Interacting Answer Sets"
Chiaki Sakama and Tran Cao Son
======================================================================
ABSTRACT:
We consider agent societies represented by logic programs. Four different types of social interactions among agents, cooperation, competition, norms, and subjection, are formulated as interactions between answer sets of different programs. Answer sets satisfying conditions of interactions represent solutions coordinated in a multiagent society. A unique feature of our framework is that answer set interactions are specified outside of individual programs.
This enables us to freely change the social specifications among agents without the need of modifying individual programs and to separate beliefs of agents from social requirements over them. Social interactions among agents are encoded in a single logic program using constraints. Coordinated solutions are then computed using answer set programming.


===========================================================================
"Applications of a Constraint-based Approach to Build and Re-examine Biological Regulatory and Formal Neural Networks"
Fabien Corblin, Hedi Ben Amor, Eric Fanchon, Laurent Trilling, Jacques Demongeot and Nicolas Glade
==========================================================================
ABSTRACT:
The biological knowledge obtained from numerous experiments is usually incomplete. Two kinds of knowledge can be distinguished concerning the biological regulatory network of interest : knowledge about its structure (usually formalized as an interaction graph) and about its behaviour. Building a meaningful model of biological regulatory network is usually done by specifying the components (molecular species, genes, ...) and their interactions, by comparing the observed behaviours to the predicted ones, and by modifying (architecture or parameters) it in order to reach the optimal _tness. The idea in this work is to avoid such a trial-error process (or learning process) by building automatically the set of all models coherent with all the available knowledge (structural and behavioural) and by characterizing this set. We propose to use the formalism of Hop_eld-like networks (a usual model for neural networks and often for regulatory networks) where the structure and the dynamics are formalized with formal constraints, so as to construct and analyse the biological models by reverse engineering methods. These formal constraints are compiled into boolean formula in conjunctive normal form (CNF) and then submitted to a satis_ability solver (SAT). Using this method, we construct queries in order to build or re-examine a biological regulatory networks and formal neural networks.


==========================================================================
"Pronostic of Breast Cancer Based on a Fuzzy Classification"
L. Hedjazi, T. Kempowsky-Hamon, M.-V. Le Lann and J. Aguilar-Martin1
==========================================================================
ABSTRACT :
The techniques of learning and classification can help with data analysis from ana-cyto-pathological cancerous tissue to develop a tool for the diagnosis or prognosis of cancer. This paper presents the application of the classification LAMDA with recent developments that allow the processing of information represented by intervals of numerical values. This study includes the selection of the attributes based on several preliminary analysis. This method was applied to the prognosis of breast cancer from two databases and compared with results published previously.


======================================================================
"Generative Modeling by PRISM"
Taisuke Sato
======================================================================
ABSTRACT:
PRISM is a probabilistic extension of Prolog. It is a high level language for probabilistic modeling capable of learning statistical parameters from observed data. After reviewing it from various viewpoints, we examine some technical details related to logic programming, including semantics, search and program synthesis.


==========================================================================
"Knowledge Representation in Systems Biology"
Andre Doncescu
==========================================================================
ABSTRACT:
The development of biological databases provides an opportunity to recognize patterns not considered in the past and to use these patterns for diagnosis and for knowledge science discovery. ** The different approaches are related to the type of reasoning. The physical models basically represented by differential equations "mime" physical structure and give a synoptic view. The engineering aspect is defined by functional models which describe the chain of functions realized by the system. The modelling of some systems demands an informational representation, which is supposed to gather signals and find out the relations causality/effects. The informational model is based on the fusion of information, which means the ability to combine sources of information describing the same phenomena. But more of that, the sources of information deliver data, decisions and models. To accomplish these tasks the central concept is *knowledge representation which becomes fondamental for knowledge discovery in systems biology.


Contact

Takehide Soh: soh at nii.ac.jp

Back
Webmaster: Takehide Soh (E-mail soh at nii.ac.jp)