Developing and Delivering Personalized Polygenic Scores at Scale
summary
An end-to-end pipeline turns GWAS summary statistics from ~1.8M training samples into candidate polygenic scores evaluated on ~200k validation and similar test cohorts, shrinking development from months to days and enabling models with 25,000+ genetic features. The system called prs machine uses MLflow for model tracking and Metaflow for reproducible workflows, runs population-aware meta-analyses when >=5k cases exist, pairs each model with an interpreter to map raw scores to absolute risk, and serves real-time predictions to millions.
AI-written summary · about
chapters (10)
- 0:00Start
- 2:11Sample Health Predisposition Report
- 2:43The 23andMe Research platform
- 3:31More data: the virtuous cycle
- 7:25Disease prediction using polygenic risk scores
- 12:51How we build a polygenic risk score
- 16:12Genomic research needs more diversity
- 18:38Automated population optimization
- 20:27A new machine learning service: PRSmachine
- 26:37PRS endpoint producing customer results
related talks
17mBig data in personalized medicine
20mAdvancing Epigenetic Predictors with Scalable Machine Learning
45mUsing Support Vector Machines in Scikit Learn to discover genetic aetiologies
17mPrecision Medicine for All: Using Tidymodels to Validate PRS in Brazil
26mShiva Amiri, CEO, BioSymetrics at MLconf Seattle 2017
55mGPU-accelerated virtual drug screening
42mManualRivas
42mHow Novel Compute Technology Transforms Life Science Research