See this nice blog for how to choose which genome version for NGS data analysis:
http://genomespot.blogspot.com/2015/06/mapping-ngs-data-which-genome-version.html
It is recommended to use Ensembl rather than UCSC or NCBI annotation. For Ensembl genome files, primary assembly is preferred than toplevel as the latter contained haplotype sequences which are not properly handled by the current sequencing mappers.
