Understanding and using Approximate Bayesian Computation
Course offered on May 24, 25, 28 2021.
Contact. Miguel Navascués (IOB), firstname.lastname@example.org
To register, please email the contact above before May 9. Indicate your level of experience on approximate Bayesian computation (ABC) or other simulation-based methods (none is OK), population genetics and programming (see requirements below). State whether you are interested in attending the whole course or only some subjects.
Course aims. 1) Acquire the capacity to critically asses an ABC analysis from a published (or under review) work. 2) Acquire the capacity to set up an ABC analysis for simple models to address classical questions in population genetics (e.g. population size changes, isolation vs. migration).
Course content. The exact content of the course can be tailored to the level of experience of participants. Assuming participant with no experience, we will start with a brief introduction to Bayesian statistics. Then we will work on classical approaches in ABC, their limitations and potential/usual blunders. Then we will work on a modern implementation of ABC that makes use of random forests (a machine learning approach). Finally we will discuss current trends and developments on simulation-based inference (e.g. use of machine learning approaches, inference on models with selection). All examples will be based on population genetic data.
Course organization. Giving the current situation, this will be a virtual workshop via Zoom. The course will be organized over three days (Monday, Wednesday and Friday), one 3-hours session each day (including break(s)). The course will be based entirely on R, with some parts being “type along” and some time for personal exercises. Participants could also invest around 1-2 hours of personal work for more advanced exercises between sessions. All examples will be based on population genetics thus some minimum knowledge is required on the processes determining genetic diversity, the concept of coalescent and coalescent simulations. Working knowledge of R is a must, knowledge of Python can allow more advanced exercises but it is not required. A maximum of 15 students will be admitted.