Bypassing the Recruitment Bottleneck: Mapping Autism Population Data with Open-Source Data Science
A data-driven exploration using UMAP, HDBSCAN, and Cosine Similarity to uncover the socio-demographic realities of families raising autistic children.
10/3/20261 min read
Rethinking Autism Research: The Public Data Advantage
Traditional participant recruitment in autism research presents a persistent bottleneck. Enrolling even a small clinical cohort can take upwards of 18 months and tens of thousands of dollars. Meanwhile, open-access national repositories—such as the National Survey of Children's Health (NSCH)—contain thousands of rigorously gathered, population-level family profiles waiting to be analyzed.
In my recent talk at FABA 2026 ("The Hidden Map: How Free Public Data Can Fuel Your Autism Research"), I shared how modern computational tools can turn these massive datasets into actionable clinical insights. By applying machine learning techniques like non-linear dimensionality reduction (UMAP), density clustering (HDBSCAN), and Cosine Similarity vector analysis, we can explore high-dimensional family profiles without incurring the costs or delays of primary field recruitment.