Estimation of demo‐genetic model probabilities with Approximate Bayesian Computation using linear discriminant analysis on summary statistics

• DOI: 10.1111/j.1755-0998.2012.03153.x
• Publication Date: 2012-05-09T11:49:37Z
• Authors: ARNAUD ESTOUP; ERIC LOMBAERT; JEAN‐MICHEL MARIN; THOMAS GUILLEMAUD; PIERRE PUDLO; CHRISTIAN P. ROBERT; JEAN‐MARIE CORNUET
• Keywords: Approximate Bayesian Computation

Abstract Comparison of demo‐genetic models using Approximate Bayesian Computation (ABC) is an active research field. Although large numbers populations and (i.e. scenarios) can be analysed with ABC molecular data obtained from various marker types, methodological computational issues arise when these become too large. Moreover, Robert et al. ( Proceedings the National Academy Sciences United States America , 2011, 108, 15112) have shown that conclusions drawn on model comparison cannot trusted per se required additional simulation analyses. Monte Carlo inferential techniques to empirically evaluate confidence in scenario choice are very time‐consuming, however, summary statistics (Ss) scenarios We here describe a innovation process efficient probability computation linear discriminant analysis (LDA) Ss before computing logistic regression. used simulated pseudo‐observed sets pods ) assess main features method (precision time) traditional estimation raw not LDA transformed) Ss. also illustrate real microsatellite produced make inferences about invasion routes coccinelid Harmonia axyridis . found probabilities computed LDA‐transformed were strongly correlated. Type I II errors similar for both methods. The faster we observed (speed gain around factor 100 Ss) substantially increases ability practitioners analyse hence provides manageable way power available discriminate among set complex scenarios.