Investigating Element Data with Google DeepVariant - Andrew Carroll, PhD

Andrew Carroll, PhD - Product Lead, Google Health Genomics

• Analyzed Element data through dual lenses: human-written heuristics and machine learning.
• Used DeepVariant as the open source method of variant detection.
• When compared with existing technologies Element performed as well or better in some cases:
  • Element matches existing technologies for typical WGS coverage
  • DeepTrio observed fewer false de novo calls with Element, facilitating rare disease classification.
  • A cleaner error profile at lower sequencing depths has the potential to unlock more applications.