Building a scalable genomic matching algorithm
summary
A case study of building Atlas reveals concrete claims and tradeoffs: a reproducible Atlas open-source cloud-hosted algorithm that compares patient and donor HLA to return ranked donors, implemented with Azure Functions and serverless SQL, and stress‑tested against global WMDA data. Engineers hit a memory wall from an estimated 10^26 possible genotypes, resolve it by reversing lookups into reference haplotype data and cutting heavy‑tail probability mass, and achieve live searches in 2023 that complete billions of comparisons in minutes.
AI-written summary · about
chapters (8)
- 0:00Genomic matching intro
- 5:15Matching algorithm challenges
- 8:04Building Atlas algorithm
- 11:41Memory crisis in matching
- 16:58Performance bottleneck
- 21:33Atlas live and takeaways
- 22:37Domain beyond tech
- 25:01Call to action
related talks
26mBuilding generative AI–enriched applications with AWS & MongoDB Atlas
50mCancer Genomics - a biologist and a developer
35mHow we've made a global search engine for genetic data
46mScalable Genomics with ADAM
42mHow Novel Compute Technology Transforms Life Science Research
46mKeynote: Life Science Research at Scale
23mGenomics and AI on Cloud Foundry
10mGenesis Final Round - Drylab AI