An important part of the research lead by
SAM, SVH, Mistis and IPS teams is motivated by biology or medecine problems.
Applications are found in pharmacometrics, cancer, neurosciences and ecology.
Figal developped also multi-state models to predict the nosocomial pneumonia in inensive care unit.
Fruitful links have been established with
the LECA (Laboratoire d'ecologie alpine) and the GIN (Grenoble institute of neurosciences) .
SVH team has been involved in the Laboratoire
d'Excellence TOUCAN (Toulouse Cancer), an interdisciplinary project
mainly involving biology and biomedicine teams from the Canceropole
in Toulouse. The mathematical part from the LJK is meant as a
methodological support for researchers in biology and medicine. Since
2012, mathematics have been developed around two themes, data mining in
genomic databases, and probabilistic
modeling of population growth with mutations. The first theme, developed by Bernard Ycart (SVH) deals with the statistical treatment of open
source databases either of genesets such as the MSigDB C1 to C7, or the
genomewide expression data from the GEO depository. In the
second theme, developed with Agnes Hamon (SVH), statistical techniques for
estimating mutation probabilities have been produced.
F. Letue and S. Lambert-Lacroix
(TIMC-IMAG) proposed a Partial Least Squares algorithm in the Cox
model for dimensionality reduction of covariates. M. Giacofci (SAM),
S. Lambert-Lacroix (TIMC-IMAG) and F. Picard (LBBE Lyon) developed an
unsupervised clustering method for sets of curves from omics
data. This uses E-M over a wavelet representation, with a mixed
effects setting to take account of inter-individual variability.
A second and growing aspect of our cancer work is probabilistic
modeling of population dynamics. Notably, B. Ycart (IPS) and A. Hamon worked
on statistical techniques for estimating mutation probabilities.
Much of our work in this area
focuses on nonlinear mixed effects models. C. Bazzoli developed a
Fisher information based methodology for these that allows the
population designs of clinical studies to be evaluated and optimized.
An R module for this, `PFIM', was developed and a scientific/technical
user-support group for it was created in collaboration with INSERM UMR
C. Bazzoli also worked on estimating cross-patient response
variability and its links to drug efficacy and toxicity, notably for
antiretroviral drugs. She proposed a
model for the plasma and intracellular pharmacokinetics of zidovudine
(ZDV) and lamivudine (3TC) nucleoside reverse transcriptase inhibitors
in HIV patients.
Input-output relations are
a critical aspect of neural modelling. With R. Berg and S. Ditlevsen
(Copenhagen University), A. Leclercq-Samson developed methods for
estimating synaptic inputs from intracellular recordings of membrane
potentials, using stochastic differential equation models and
estimators that couple stochastic approximation and particle filtering.
With LECA Grenoble, LMAP Pau
and INR, P. Garat worked on statistical models for the `mast'
(intermittent) seeding strategies of the European larch in relation to
its dominant seed predator, Strobilomyia laricicola.