Confidence intervals obtained from different methods using simulated data and their evaluation through artificial neural network

Abstract

Determination of confidence intervals (CI) using different methods at different levels of population size (Ps), marker space (Ms), standard deviation of QTL effect (SDQ), ratio of additive to dominance SD (Rad) and QTL position relative to flanking markers (rpQ) were investigated by simulation. The simulation conducted by F2 design and analyzed with Haley and Knott (HK) method. Moreover an ANN model trained by backprobagation algorithm obtained to predict CIs of different methods at combinations of simulated parameters. After obtain of best ANN model with optimal adequacy parameters we used the artifitial neural network (ANN) model to prediction of CIs at very large-scale combination of simulated parameters comparing actual simulation study. Bootstrap method had more per cent of acurate intervals but average size of the intervals was very high in more scenarios. 1 LOD support interval and bayesian credible interval resulted to be preferable with high per cent of acurate and small confidence intervals, moreover they weekly affected by parameters such as population size and SD of QTL. This study investigated that we can predict CIs for more combination of simulated parameters using best trained ANN. By this study it is sugestive to consideration of more combinations of simulated parameters using the model obtained by best structured ANN to expanding of original study.