High throughput methods are powerful tools to develop predictive models for assessing drug-induced liver injury (DILI). The development of predictive models, however, requires a drug reference list with an accurate annotation of DILI risk in humans. We previously developed a DILI annotation schema based on the information curated from the FDA-approved drug labeling for 287 drugs. In this paper, we refined the schema by weighing the evidence of causality (i.e., a verification process to evaluate a drug as the cause of DILI) and generated a dataset that rank the DILI risk (DILIrank) in humans for 1036 FDA-approved drugs, providing the largest annotated dataset in public domain.