Figure 1.

Overview of the OARD pipeline, data source, and architecture

Briefly, we first extracted individual levels of HPO and MONDO concepts from two institutions, CUIMC and CHOP, EHR databases. For each dataset, we also generated stratified subsets (as described in the material and methods) when it was applicable. Only the resulting frequency and co-occurrence data (without individual-level data) were stored in a MySQL database. The MySQL database was disconnected from the institutional internal network before it was served to the public via the RESTful web API (https://rare.cohd.io/). The association analysis is performed per query, and the results are returned on the fly.