Whole genome prediction (WGP) based on GBLUP and Bayesian models (e.g. A, B, C and R) are routinely used in animal breeding but have not been well tested in a genetic mapping population that segregates for QTLs. In our pig model experiment, purebred Duroc and Yorkshire sows were crossed with Göttingen minipig boars to obtain an F2 intercross resource population (n=566) that is known to segregate for QTLs affecting several obesity and metabolic phenotypes. All pigs were genotyped using Illumina Porcine 60k SNP Beadchip. The objective of this study was to test the differences in predictive accuracy and bias of WGP using Bayesian methods (Bayes Cpi, Bayes R and Bayes Power Lasso) and a polygenic inheritance (GBLUP) model in a resource population with segregating QTLs. Initially, we are applying Bayesian Power LASSO (BPL) model to investigate genetic architecture of obesity phenotypes in order to partition genomic variances attributable to different SNP groups based on their biological and functional role via systems genetics / biology approaches. We apply different methods to group SNPs: (i) functional relevance of SNPs for obesity based on data mining, (ii) groups based on positions in the genome, and significance based on previous genome-wide association study in the same data set. Preliminary results from our studies in production pigs indicate that BPL models have higher accuracy but more bias than GBLUP method, although using different power parameters has no effect on predictive ability of the models. Future work involves applying these and other WGP methods to the F2 intercross resource population.