======================================================================
TITLE:
"Evaluating Abductive Hypotheses using an EM Algorithm on BDDs"
AUTHORS:
Katsumi Inoue, Taisuke Sato, Masakazu Ishihata, Yoshitaka Kameya
and Hidetomo Nabeshima
======================================================================
ABSTRACT:
Abductive inference is an important AI reasoning technique to find
explanations of observations, and has recently been applied
to scientific discovery.
To find best hypotheses among many logically possible hypotheses,
we need to evaluate hypotheses obtained from the process of hypothesis
generation. We propose an abductive inference architecture combined
with an EM algorithm working on binary decision diagrams (BDDs).
This work opens a way of applying BDDs to compress multiple hypotheses
and to select most probable ones from them. An implemented system
has been applied to inference of inhibition in metabolic pathways
in the domain of systems biology.
(This work will be presented at IJCAI-2009, in Pasadena, CA, July 2009.)
======================================================================
TITLE:
"Abductive Reasoning to Model Some Aspects of Breast Cancer Therapy"
SPEAKER:
Andrei Doncescu
======================================================================
A diagnosis of cancer therapy is one of the most difficult experiences
we can face due to the complex decisions on treatment with potentially
life-threatening illness. The prediction of the cancer evolution in
a frame of personalized therapy requests an exhaustive analysis of a patient
profiles by efficient machine learning techniques. Using techniques
as Inductive Logic Programming (ILP) which can find hypotheses that
account for given observations with a background theory. The Hypothesis
Finding applied on a relational database containing 250 profiles of cancer
patient : relapse (or recover), emergence of 5 proteins SC35, 9G8, hnRNPA1,
ASF-SF2 and SRp20 as well as the clinical information for each patient
figure out several causal relations between treatment and the life extension.
-->
======================================================================
TITLE:
"A Logical Modeling on Dynamic Transitions of
the High Flux Backbone in Metabolic Pathways"
AUTHORS:
Yoshitaka Yamamoto, Katsumi Inoue, and Koji Iwanuma
======================================================================
ABSTRACT:
The flux distribution in metabolic pathways has two features.
First is its inhomogeneity. Whereas most metabolic reactions
have low fluxes, the overall activity of the metabolism is dominated
by several reactions with very high fluxes. The second is its
scale-freeness. Because this inhomogeneity is preserved not only for
each growth condition or for each pathway but also for each individual
reaction.
In this talk, based on these two features, we propose a logical model
which enables us to estimate dynamic transitions of dominant source
reactions in metabolic pathway, called High flux backbone (HFB).
Previously, HFBs at the steady state are computed based on Flux-balance
Analysis (FBA). However, due to the constraints on applications of FBA
(Ex. FBA cannot be applied in case that the steady state does not hold),
it is difficult to estimate dynamical transitions with only FBA.
Using our proposed logical model, we can compute the past HBAs before
the steady state. This computation is based on abductive-inference.
Besides, it is known that dynamic transitions of HFBs are caused by some
particular generic signals. Our long term goal is to detect these signals
that are still unknown. For this goal, we need to use inductive-inference.
Here, we show a possibility that both abductive and inductive inferences
will be used for analyzing the mechanism of dynamic transitions of HFBs in
metabolic pathways.
======================================================================
TITLE:
"Numerical Values in Logic Modeling"
SPEAKER:
Gabriel Synnaeve
======================================================================
ABSTRACT:
We present the steps required to obtain abductive or inductive
hypothesis from experimental data. We will explain the use of the
tools that have been developed for discretizing experimental data
and building symbolic models. We will then explain our way to handle
several levels with a special "compute" predicate and how we used it
to build a Kinetic Model as an application example of this approach.
======================================================================
TITLE:
"A SAT-based Approach for Analyzing Biochemical Pathways"
AUTHORS:
Takehide Soh and Katsumi Inoue
======================================================================
ABSTRACT:
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 this paper, we show a SAT-based approach for analyzing biochemical pathways.
We review a previously proposed approach where a weighted Max-SAT solver was
used to compute possible reaction states in biochemical pathways.
We show how this problem can be solved using a SAT solver. In our approach, we
translate a given biochemical pathway into a SAT problem.
One of the advantages of this method is that it generates only feasible
solutions by utilizing minimal models. In the experiments, we attempt to compute reaction
states in the glycolytic pathway and compare them with the previous result from the literature.
======================================================================
TITLE:
"ILP and Grammatical Evolution"
SPEAKER:
Petr Buryan
======================================================================
ABSTRACT:
I will briefly show my first achievements with a novel approach to
combining ILP with Genetic Algorithms based on Grammatical Evolution.
Current combined GA-ILP systems either employ a user-defined template for
mapping first-order rules into bit strings (e.g. GA-SMART or G-NET) or
use hierarchical representations instead of fixed length bit-strings and
evolve a population of logic programs in a Genetic Programming (GP) manner
(e.g. GILP or GLPS).
However, even though some of these systems use background knowledge for
generating the initial population or seeding the population, most of these
systems cannot benefit from intentional background knowledge in the same
way as in usual first-order learning systems.
The approach proposed should come over this basic problems of mentioned
algorithms by utilisation of grammatical evolution working with
user-defined problem-based grammar, that describes and considerably limits
the search space.
======================================================================
Title:
"Automatic design of experimental plans"
SPEAKER:
Gauvain Bourgne
======================================================================
Abstract:
In order to apply knowledge discovery processes to biological problem,
a qualitative model as well as a data set of experimental observations
is usually needed. However, observations can sometimes be too scarce to
produce good hypotheses, and there is a need to gather new data.
In this case, rather than doing random experiments, it might be useful
to use these incomplete hypotheses as a way to determine which data would
be useful to improve them, and design a flexible experimental plan.
Different strategies could be used, and confronting the results obtained
with each them should make it possible to refine the hypotheses in a more
comprehensive way, while reduced the biases.
These processes would first be tested using a simulator.
Takehide Soh: soh at nii.ac.jp
Back