12-16 February 2018
Jaap Kaandorp (UvA), Jaap Molenaar (WU), Hans Stigter (WU), Cajo ter Braak (WU)
Many problems in bioinformatics and systems biology require at a certain moment application of some optimisation technique. When facing optimisation problems, many researchers suffice with applying a simple algorithm in Excel, coming to the conclusion that it does not work properly or runs into convergence problems. It is not necessary to stick to that stage, since optimisation is a rapidly expanding field of research, that generates more and more efficient, successful, and easy-to-use algorithms.
This course is intended to provide you with the main ideas underlying optimisation, and to get you acquainted with the optimisation techniques that are nowadays available and widely used. In addition, a great variety of examples of optimisation problems in life sciences will be presented and discussed.
The course has the following structure:
- Day 1: Math refresher and univariate optimisation (Jaap Molenaar+ Hans Stigter)
- Dag 2: Local multivariate optimisation (Jaap Molenaar + Hans Stigter)
- Dag 3: Global stochastic optimisation (simulated annealing, evolutionary algorithms) (Jaap Kaandorp)
- Dag 4: Parameter estimation, systems control and optimisation (Hans Stigter + Jaap Molenaar)
- Dag 5: MCMC techniques (Cajo ter Braak)
The first day is devoted to study simple problems that allow the use of algorithms that are intuitively clear. On day 2, we extend to problems that are bit more complicated. On day 3, we introduce techniques that are based on stochastic searching, while day 4 is spent on the frequently met problem of finding values for the parameters in a model, given a dataset. This challenging problem requires a bit more sophisticated optimisation tools. You will learn about uncertainty and identifiability of the parameters. In addition, you will get some experience with using appropriate parameter estimation tools. On the last day, the state-of-the-art will be presented, since then the popular and powerful Markov Chain Monte Carlo (MCMC) technique will be dealt with. This method is based on a Bayesian approach. So, you will learn about prior distribution, posterior distribution, likelihood, and how in MCMC likelihood and prior distribution are combined to arrive at a posterior distribution.
We assume that the participants have some basic knowledge of modelling in the life sciences. However, it is not necessary to be an expert in mathematics, since the course starts with a session to refresh the mathematical background needed for optimisation.