You basically take all the dna in the sample and chop it all into small pieces and all of these small pieces are read into computer memory, and then an algorithm goes through them all and merges the pieces that fit together back together into the original long form of the dna. Only pieces that have overlaps with each other will merge together, and there are many redundancies, so if any pieces accidentally fit together that aren't actually supposed to, then they are discarded, because this won't happen nearly as often as the real merge.

So maybe a bunch of human dna gets mixed up with bacterial dna - I dunno really - but it really doesn't matter because of the way they put back together.