7 Sep 2004 - Keynote address Relational Data Mining conference
On 7 September 2004, PharmaDM co-founder and CSO Luc Dehaspe gave an invited talk at the annual International Conference on Relational Data Mining ( ILP-2004, Portugal): "From Promising to Profitable Applications of Relational Data Mining: a Case Study in Drug Discovery".
From promising to profitable
applications of ILP: a case study in drug discovery.
Luc Dehaspe
PharmaDM
AbstractPharmaDM was founded end 2000 as a spin-off from three European
universities (Oxford, Aberystwyth, and Leuven) that participated in two
subsequent EC projects on Inductive Logic Programming (ILP I-II,
1992-1998). Amongst the projects highlights was a series of
publications that demonstrated the added-value of ILP in applications
related to the drug discovery process. The mission of PharmaDM is to
build on those promising results, including software modules developed
at the founding universities (i.e., Aleph, Tilde, Warmr,
ILProlog), and develop a profitable ILP based data mining product
customised to the needs of drug discovery researchers. Technology
development at PharmaDM is mostly based on "demand pull", i.e., driven
by user requirements. In this presentation I will look at the way
ILP technology at PharmaDM has evolved over the past four years and the
user feedback that has stimulated this evolution.
In the first part of the presentation I will start from the general
technology needs in the drug discovery industry and zoom in on the data
analysis requirements of some categories of drug discovery researchers.
One of the conclusions will be that ILP -via its ability to handle
background knowledge and link multiple data sources- offers
fundamental solutions to central data analysis problems in drug
discovery, but is only perceived by the user as a solution after is has
been complemented with (and hidden behind) more mundane technologies.
In the second part of the presentation I will discuss some research
topics that we encountered in the zone between promising prototype and
profitable product. I will use those examples to argue that ILP
research would benefit from very close collaborations, in a
"demand-pull" rather than "technology push" mode, with drug discovery
researchers. This will however require an initial investment of
the ILP team to address the immediate software needs of the user,
which are often not related to ILP.
