Machine Learning for Personalized Medicine

Marie-Curie Action: "Initial Training Networks"

Project: Adverse drug reaction discovery


Victor Bellon
Jean-Philippe Vert
Mines ParisTech
Paris, France


Victor Manuel Bellon Molina was born in Barcelona, Spain, in 1989. He studied Mathematics at the Autonomous University of Barcelona, and obtained his MSc in 2013 working on Bayesian Gaussian network classifiers for mass spectra classification.

Project description

Drug side effects can lead to serious health problems and each year the number of reported critical incidents increases. Moreover, it is known that side effects vary between different individuals. These interpersonal variations justify tackling this problem in the context of personalized medicine. Indeed, they indicate that drug side effects could be modulated not only by environmental parameters, but also by genetic characteristics of the patient. The field that studies the relation between drugs and genome is called chemogenomics. I work within this framework to predict the response of patients to different drugs. Patient response includes both the therapeutic response (whether the treatment is working or not) and unwanted side effects. For that purpose, I use machine learning techniques to create models that use various types of information such as

  • genetic and clinical data about the patients,
  • the chemical structure and the chemical properties of the drugs, and
  • information about the targets of the drugs.

Motivation for participating in the network

I have always been interested in mathematics and machine learning. I think that they offer a powerful set of techniques to help us creating models and solving complex problems. Over the last years almost every field has experienced an increase in the amount of available data and in the capacity to generate these data, which cannot be analyzed without the help of statistical and mathematical models. The biological field is no exception and presents some of the most challenging and interesting questions for humanity. I am interested not only in machine learning in general, but also in using it to solve biological and medical problems. The Marie Curie Initial Training Network offers me the possibility to learn more about these topics from expert researchers, and to collaborate with and learn from companies working in this field.


Duration of fellowship: from January 2014 to December 2016


MLPM2012 Publications:


Victor Bellon, Veronique Stoven, Chloe-Agathe Azencott. Multitask Feature Selection with Task Descriptors. Pacific Symposium on Biocomputing (PSB) 2016 


- Victor Bellón, Chloé-Agathe Azencott, Véronique Stoven, Olivier Collier, Azadeh Khaleghi, Valentina Boeva, Jean Philippe Vert. DREAM Rheumatoid Arthritis Responder Challenge 2014: Team Lucia.

- Víctor Bellón, Chloé-Agathe Azencott and Véronique Stoven. A Multiplicative Multitask Lasso approach with task descriptor variables. FEAST 2015: ICML Workshop on Features and Structures.

Other Publications:

- Bellón V, Cerquides J, Grosse I. Gaussian Join Tree classifiers with applications to mass spectra classification. The Sixth European Workshop on Probabilistic Graphical Models (2012), p. 19-26