This course will be delivered ONLINE: 20 hours of online live lessons, plus 12.5 hours of recorded classes and assignments. A good internet connection is required to follow the course.
Most researchers in life sciences are exposed in their research to a multitude of methods and algorithms to test hypotheses, infer parameters, explore empirical data sets, etc.
Bayesian methods have become standard practice in several fields, (e.g. phylogenetic inference, evolutionary (paleo)biology, genomics), yet understanding how these Bayesian machinery works are not always trivial.
This course is based on the assumption that the easiest way to understand the principles of Bayesian inference and the functioning of the main algorithms is to implement these methods yourself.
The instructor will outline the relevant concepts and basic theory, but the focus of the course will be to learn how to do Bayesian inference in practice. He will show how to implement the most common algorithms to estimate parameters based on posterior probabilities, such as Markov Chain Monte Carlo samplers, and how to build hierarchical models.
He will also touch upon hypothesis testing using Bayes factors and Bayesian variable selection.
The course will take a learn-by-doing approach, in which participants will implement their own MCMCs using R or Python (templates for both languages will be provided).
After completion of the course, the participants will have gained a better understanding of how the main Bayesian methods implemented in many programs used in biological research work. Participants will also learn how to model at least basic problems using Bayesian statistics and how to implement the necessary algorithms to solve them.
Participants are expected to have some knowledge of R or Python (each can choose their preferred language), but they will be guided “line-by-line” in writing their script. The aim is that, by the end of the week, each participant will have written their own MCMC – from scratch! Participants are encouraged to bring own datasets and questions and we will (try to) figure them out during the course and implement scripts to analyze them in a Bayesian framework.
Basic knowledge of Python or R and Statistics. All participants must have a personal computer (Windows, Macintosh), with webcam if possible, and a good internet connection.
- Introduction to probabilistic models and Bayes theorem. We’ll learn:
- How to calculate the likelihood of any dataset under a simple model
- The Bayes principles (what is a prior? What is a posterior probability?)
- Write an R (or Python) script to compute the likelihood of data under Normal and Gamma models. 3D plots of the likelihood surface.
- Basic structure of Markov Chain Monte Carlo, the most popular algorithm in Bayesian analysis.
- MCMC, how it works, how to implement it. Based on the likelihood functions written on Monday, implement an MCMC to fit normal and gamma distributions.
- What is the difference between modeling a pattern and modeling a process? When should we prefer one or the other? (practical) Analysis of global temperature data (provided) to estimate the existence of any climatic trends.
- Hypothesis testing using marginal likelihoods. (practical) How to interpret and summarize the results of an MCMC, how to assess if it worked.
- Bayesian tricks to avoid model testing: Hierarchical modeling, shrinkage, and Bayesian variable selection.
- Continue working on the MCMC script and with own data.
- Alternative algorithms in Bayesian analyses: Gibbs sampling and Approximate Bayesian Computation (ABC). Basic principles of machine learning.
- Finalize the MCMC script and (if applicable) plan the future development of the methods implemented for analysis of own data.
- Course Fee
- Early bird (until July 31st, 2020):
- 535 € *
(428 € for Ambassador Institutions)
- Regular (after July 31st, 2020):
- 644 € *
(515.20 € for Ambassador Institutions)
- This includes course material (VAT included).
* Participants from companies/industry will have an extra charge of 100 €.