Discrepancy in taxonomic assignment between reads and gene-centric assemblies

Hi there,
when I take the reads from a certain taxon (Staphylococcus aureus) to get them gene-centric assembled, some of the resulting contigs show best BLASTX hits to other species than S. aureus . How could this be explained?

Thanks,

Ralf