From 20 Years to a Diagnosis: Building LATAM’s Largest Rare-Disease Knowledge Graph

Session track: Data Intelligence

Session time: 11:30 pm - 12:00 am UTC November 13

Session description:

Rare diseases aren't rare in aggregate: they affect over 300 million people, yet the average patient waits years for a diagnosis — Dimas himself waited 20. The data needed to shorten that journey exists, but it's fragmented across genetic databases, clinical trials, drug registries, public-health systems, and scientific literature that never talk to each other. In this session, Dimas will show how Raras built RarasNet, a Neo4j knowledge graph that unifies 8,000+ rare diseases with their genes, phenotypes, drugs, repurposing candidates, clinical trials, papers, and Brazilian public-health (SUS) coverage data. He will walk through the modeling decisions behind connecting heterogeneous biomedical sources, the Cypher patterns that power patient-similarity and treatment-discovery queries, and how the graph feeds an AI copilot that helps patients and clinicians navigate care. You will learn how to model multi-source biomedical data without losing provenance, how to design relationships that surface non-obvious connections (a drug approved for disease A that may help disease B), and how a knowledge graph becomes the backbone of a real-world health platform serving thousands of patients across Latin America. Whether you work in healthtech or any domain drowning in disconnected data, you will leave with concrete patterns for turning fragmentation into discovery.