Financing: Internal Funding KU Leuven (KU Leuven)
Start: 2014-08-01
End: 2015-07-31

Description:

In the future, large amounts of complex clinical data are expected to become available in real time. Currently this is mainly limited to clinical measurements which can largely be analysed using fast signal processing techniques (heart rate, respiratory signals, ...). A paradigm shift will occur once new omics data for patients will become available in real time. For instance in adaptive clinical trials, it is expected that omics data are routinely collected, which then need to be analysed quickly in order to make evidence-based clinical decisions on trial design. Patients' omics-based molecular profiles may be linked to treatment outcomes (or intermediate trial observations), which in turn may lead to improved stratification of study groups.

 The challenge, then, is to obtain precise, up-to-date models at a high speed to uncover potential associations between incoming molecular profiles and treatment outcomes / observations. Online learning is a promising technique to achieve this goal. Online learning aims to update existing models when new data becomes available. This is faster than retraining from scratch and makes processing complex data in real time feasible, even when processing necessitates non-linear modeling.

 Another challenge is coping with large amounts of omics data, stressing the need for large scale computing and database infrastructures. Finally, in order to be able to test modeling methodologies for online and large scale learning, the project involves generating simulated omics datasets in anticipation of large volumes of real data.

 Therefore, this project will address the challenges of online learning, kernel methods for large-scale problems, and simulated datasets, as described below. This entails the following research tasks:

·       Further development of online learning with kernel methods

·       Development of HPC-implementations of kernel methods

·    Generation of simulated genome-wide datasets to test the scalability of aforementioned methods 

SMC people involved in the project:

• Maira Ribeiro Rodrigues (Grantee)
• Bart De Moor (Promoter)
• Inge Thijs (Team member)