Andrew leads product development for the genomics team in Google Health. The genomics team develops methods that improve real-world applications in clinical genomics, population-level sequencing, drug discovery, and the combination of genomic and clinical data. Prior to Google, Andrew was Chief Scientific Officer at DNAnexus, where he supported many of the first large-scale genomics projects, such as the CHARGE Consortium, Regeneron-Gesinger and Regeneron-UKBiobank cohorts, the 3000 Rice Genomes Project, PrecisionFDA, and the St. Jude Pediatric Cancer Cloud. Andrew holds a PhD in Molecular Biology from Stanford University and a Bachelor’s degree in Physics from the University of Virginia.
DeepConsensus: Sequence Transformers for Sequence Correction
DeepConsensus improves accuracy for Pacific Biosciences sequencing output using a ML method called a transformer. This talk details how we optimized the training of the transformer for biological sequence data. By designing the model architecture with domain-specific insights in mind, we walk through how we achieve substantial improvements to accuracy and analysis speed.