This dataset of 794 genes (and their protein-level translation) is a reference resistome that comes from contigs functionally associated with antibiotic resistance in preterm infants (Gibson et al. 2016) to overcome inconsistencies in existing AMR gene databases. 79% (n=627 out of 794) of these AMR genes had no clear homology to known ones. This reference resistome came from 401 stool samples longitudinally collected from 84 infants undergoing antibiotic treatment, and were assembled as 2,004 redundant AMR contigs experimentally tested in vitro for resistance to 16 antibiotics (Gibson et al. 2016).