postdoc position in the Orpailleur group of the Loria-Inria Nancy

When:
01/03/2020 – 02/03/2020 all-day
2020-03-01T01:00:00+01:00
2020-03-02T01:00:00+01:00

Annonce en lien avec l’Action/le Réseau : aucun

Laboratoire/Entreprise : Loria – Inria Nancy, Orpailleur group
Durée : 2 years
Contact : adrien.coulet@loria.fr
Date limite de publication : 2020-03-01

Contexte :
The postdoc will be at the Loria in Nancy, France
It is funded for 1 year, jointly by the ANR project PractiKPharma, and the I-site LUE (Lorraine University of Excellence)
The start date is fexible, but preferably not later than March 1st, 2020. It is extendable for a second year.
The postdoc will work with Adrien Coulet (Orpailleur team, Loria, Inria,Nancy) and Anne Gégout-Petit (Bigs team, Institut Elie Cartan, Inria, Nancy).
The project is also an opportunity to maintain an existing collaboration with the Shah Lab at Stanford University. Part of the experiments will be conducted with
data from the Shah Lab. Accordingly few travels to Stanford will be planed.

Sujet :
The aim of this postdococtoral project is to build upon previous results by first developing approaches that from EHR data identify without a priori groups of patients with distinct response profiles to particular drugs. Second, discovered groups will be used to identify predictors of drug response profiles.
We want to study the use of the causal inference framework (Hernàn and Robin, 2019), and in particular of double robust approaches to identify groups with heterogeneous
treatment effect, or in other words with significantly different drug response profiles. Through others, (Athey and Imbens, 2016) introduced Causal Trees for subgroup analyses. Those are regression trees with modied splitting rules that maximise the difference in treatment effect between groups. Causal Trees have
been reused to propose an ensemble method named Causal Forest that has the advantage of being non-parametric and consistent. These models will allow us to estimate the causal effect of the phenotype on the drug response.
A second objective of the project is to develop high performance predictive models for drug response profiles that we aim at identifying with causal methods.
These models can be seen as classifiers that assign individuals to a specific profile. A first challenge here is to develop real predictive models, i.e., models that are trained only on data collected prior to the prescription of the drug. A second challenge is to identify a subset of good predictive features that may help in interpreting group belonging and heterogeneous drug responses.
The set of drug studies will first be pharmacogenomic drugs, i.e, drugs known to present a variability in the population for genomic reasons, and will potentially be extended in a second time to other drugs of interest.

Profil du candidat :
Candidate’s background can be a PhD in machine learning, data mining, stats, biomedical informatics, bioinformatics or epidemiology (or other).

Formation et compétences requises :
Idéalement Apprentissage automatique, apprentissage profond, mais ne vous censurez pas.

Adresse d’emploi :
Loria – Inria Nancy, http://www.loria.fr/

Document attaché : postdoc_subject_adrien.pdf