Summary Genome-wide association studies explain at most 5%–10% of the heritable components of type 2 diabetes. Some of the “missing type 2 diabetes heritability” could be explained by low frequency variants. Pooling low frequency variants by genes into a composite test provides an alternative strategy to conventional analysis of individual SNPs for testing susceptibility genes. We examined the associations between low frequency variants and type 2 diabetes, using data from 2,538 diabetic and 2,977 non-diabetic subjects in the publicly available database of Genotypes and Phenotypes. We hypothesized that applying two approaches in combination would enable the identification of promising candidate genes. First, we combined information from all low frequency (1%–5%) variants at a locus in gene-centric analysis of associations with diabetes. Next, we searched for gene ontology (GO) biological processes that were enriched for gene-centric associations, after correcting for multiple testing to control the false discovery rate (FDR). We found three GO biological processes that were significantly enriched for associations to type 2 diabetes: ‘response to superoxide’ (FDR)-adjusted p-value =2.7×10−3), ‘response to oxygen radical’ (FDR-adjusted p-value =2.7×10−3), and ‘heart contraction’ (FDR-adjusted p-value =2.6×10−2). The ‘response to superoxide’ pathway was at the deepest level of the GO hierarchy. There were three genes which contributed to this pathway enrichment for association with type 2 diabetes, including SOD1 a gene which is involved in superoxide and oxygen radical pathways. Gene-centric tests of association with low-frequency variants, followed by analysis to evaluate which biological pathways are enriched for these associations has potential to recover, at least some proportion of, the “missing heritability” of type 2 diabetes.