Approximate Bayesian Computation (ABC) analysis is a computational method that is increasingly being applied to estimate parameter values and model selection in population genetics and phylogenetic. With its rising popularity in genetic studies, many software programs are being designed for ease of use for the biologist. So how does ABC work? Here are the basic steps to think about when designing an ABC analysis:
1.) Model selection: Choose competing models that incorporate different demographic scenarios. Include in these models parameters you wish to estimate, including both demographic (population size, migration rates, etc.) and genetic parameters (mutation rate).
2.) Prior Information: Choose prior distributions that are large enough to include all possible values of the parameter of interest.
3.) Choose the summary statistics: Choose summary statistics that you consider to be a good representation of the data you are analyzing (typically 5-20 summary stats are used).
4.) Simulate Data: Do a large number of simulations (typically several hundred thousand) for each of the indicated models.
5.) Filter the simulations: Choose a threshold value, and simulations are retained when the multivariate distance between observed summary statistics from your dataset and the simulated summary statistic are below the threshold. Several different threshold values should be evaluated to see how it changes model and parameter selection.
6.) Model Selection: The simulations from the most probable model for your dataset will be overrepresented in final retained simulations. Model comparison is evaluated with the Bayes factor.
7.) Assessing Model Choice: Assess the choice of model by re-evaluating the sensitivity of model choice with different summary statistics, threshold values, number of simulations, etc.
This is by no means a comprehensive guide to ABC analysis, but I always like to see the basic steps of an analysis before delving into the nitty-gritty of a method.