Fundamentals of computational Bayesian inference and parameter estimation
May 21-23, 2012, Lausanne
Bayesian methods for population genomics
May 23-25, 2012, Lausanne
Jérôme Goudet, University of Lausanne
Christian Lexer, University of Fribourg
Alex Buerkle, University of Wyoming
Zach Gompert, Texas State University
The overall objective of these modules is to give biologists access to the powerful inferential tools that are offered by Bayesian analysis. Bayesian methods make possible the estimation of large numbers of parameters, in some cases with complex hierarchical relationships, and the proper modeling of uncertainty throughout an analysis. However, utilizing Bayesian methods to their full advantage requires computer programming and understanding the underlying components of the estimation procedures. Unfortunately, most educational materials introduce Bayesian methods with only relatively simple models and methods that are rarely applicable for biological research. For example, several key textbooks in Bayesian methods require more than 200 pages to get to what a researcher might use in practice and instead wallow first in historical philosophical debates and closed-form solutions for relatively simple probability models. The requisite theory for a well-informed practitioner is much more compact than this and will be presented as such in these modules.
Module 1 -- a concise introduction to Bayesian estimation procedures that rely on Monte Carlo methods and the details of implementation of Bayesian estimation in computer code. The focus will be on problems in evolutionary biology, ecology and genetics. Algorithms will be implemented and studied in R. Exercises will include specifying models for various estimation problems (e.g., linear models) and implementing these in computer code (R).
Module 2 -- builds on knowledge from Module 1 and will focus on estimation problems in population genomics. Learning objectives will include increasing knowledge of both Bayesian methods and contemporary issues in population genomics. Several applications will be studied, including population genomics with genotype uncertainty, and commonly used Bayesian models (e.g., structure, F-model, etc.). Exercises will include specifying models for various parameters in population genomics, studying existing models in detail, and an introduction to implementing these models in computer code (C; previous knowledge of C is not necessary; code will be used to illustrate the algorithms). Discussions will include students studying applications to their own work and future directions for Bayesian estimation in population genomics.
Module 1 - May 21-23, 2.5 day course
Module 2 - May 23-25, 2.5 day course
* Module 1 ends at 12:30 on May 23
** Module 2 begins at 14:00 on May 23 and ends at 18:00 on May 25.
10:00-12:30 Class session 1
14:00-18:00 Class session 2
Informal dinner and evening discussion for participants staying in Lausanne
Location: University of Lausanne, Biophore Building, Room 2107
Number of participants: maximum of 25
Educational background of students:
i) Ph.D. students and postdoctoral researchers
ii) Module 1 -- interest in model-based inference in biology and learning the underlying mechanics of Bayesian analysis of real questions in evolutionary biology, ecology and genetics. Facility with basic procedural programming in R. Knowledge of basic population genetics and evolutionary biology, and familiarity with basics of probability.
iii) Module 2 -- interest in answering questions in population genomics on the basis of appropriate hierarchical probability models. Familiarity and facility with basic components of Bayesian analysis, including implementation of models in procedural programs for estimation with MCMC (in R or other language). Knowledge of basic population genetics and evolutionary biology.
Computers: students should provide their own laptops, with R installed
Day 1 - Estimating proportions (binomial-beta, multinomial-Dirichlet)
- examples: host choice experiments, allele frequency estimation
- Sampling from the posterior
Day 2 - Simple linear models (Normal-gamma) and hierarchical models
- examples: modeling quantitative phenotypes
Day 3 - Model selection
- MCMC diagnostics
Day 1 - Estimating allele frequencies
- Estimating genotype probabilities when there is genotype uncertainty
Day 2 - Hierarchical models
- The F-model
- The 'structure' model(s)
- Where does Approximate Bayesian Computation fit?
Day 3 - Modeling genotypic effects on phenotypes
- Discussion of future applications
Please note that students may choose to attend both modules or one of them due to their background and experience. An additional advanced module on "Bayesian Statistics and Applications for Phylogenetics" will be held in June -> read more. You may directly register to this 'module 3' on-line at the the same registration page.
This is a course of the SNSF Doctoral Program in Population Genomics. Therefore, priority is given to PhD students enroled in this doctoral program until April 23, 2012.
Le Biophore, UNIL-Sorge
University of Lausanne
Phone: +41 (0)21 692 4207