Description
Date depot: 7 avril 2021
Titre: Monotonic Classifiers for Systems Biology
Directrice de thèse:
Carola DOERR (LIP6)
Encadrant :
Benno SCHWIKOWSKI (BCS)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Aide à la décision et recherche opérationnelle
Resumé: This PhD project is located at the interface between theory of optimization algorithms and systems biology. It will extend and strengthen the budding collaboration between the Operations Research team at LIP6 and the Systems Biology group at Institut Pasteur.
The starting point of this PhD project is the recent work of the Systems Biology group proposing a monotonic ensemble model to predict the severity of dengue infections from blood transcriptomes. The classification model was shown to be similarly predictive as the best state-of-the-art machine learning models, while at the same time showing better interpretability -- an important feature in biomedical applications.
In this PhD project we aim to extend this work by addressing a number of fundamental questions around these monotonic ensemble models,
to improve their predictive accuracy, the ``smoothness'' of the classification model, the speed of computation, extensions to predict probabilities instead of 0/1-outcomes, etc.
We also aim to explore the broader applicability of monotonic models to different biomedical problems.
By the end of the project, we envisage that increased theoretical insight into monotonic ensemble models will provide a solid picture of their value for biomedical research.
Résumé dans une autre langue: This PhD project is located at the interface between theory of optimization algorithms and systems biology. It will extend and strengthen the budding collaboration between the Operations Research team at LIP6 and the Systems Biology group at Institut Pasteur.
The starting point of this PhD project is the recent work of the Systems Biology group proposing a monotonic ensemble model to predict the severity of dengue infections from blood transcriptomes. The classification model was shown to be similarly predictive as the best state-of-the-art machine learning models, while at the same time showing better interpretability -- an important feature in biomedical applications.
In this PhD project we aim to extend this work by addressing a number of fundamental questions around these monotonic ensemble models,
to improve their predictive accuracy, the ``smoothness'' of the classification model, the speed of computation, extensions to predict probabilities instead of 0/1-outcomes, etc.
We also aim to explore the broader applicability of monotonic models to different biomedical problems.
By the end of the project, we envisage that increased theoretical insight into monotonic ensemble models will provide a solid picture of their value for biomedical research.
Doctorant.e: Fourquet Océane