The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Predictive Disease Risk Modeling at 23andMe with Subarna Sinha - #436
Dec 11, 2020
Subarna Sinha, a leader in Machine Learning Engineering at 23andMe, dives into the world of genomic data and disease prediction. She discusses the development of polygenic risk scores and the complexities involved in predicting disease likelihood from genetic variations. The conversation highlights the technological innovations behind their ML platform, including tools like AWS and Jenkins. Subarna also addresses the challenges of data drift and the importance of team dynamics in refining predictive models, ensuring more accurate health assessments.
AI Snips
Chapters
Transcript
Episode notes
Subarna's Career Path
- Subarna Sinha's background wasn't directly in machine learning and biology.
- She transitioned from electronic design automation to genomics by working at Stanford and SRI International, eventually landing at 23andMe.
Bridging Engineering and Data Science
- Subarna Sinha's team at 23andMe bridges the gap between engineering and data science.
- They focus on building and deploying machine learning models, primarily for health-related applications, collaborating closely with data scientists.
Polygenic Risk Scores
- Polygenic risk scores combine multiple genetic variants to assess disease risk.
- 23andMe uses machine learning models trained on genetic data and survey responses, considering ethnicity for accurate predictions.