After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Using GPUs to Generate Clusters and Consensus from Variable Time Series Data: Applications from Viral Genomics to English Accents
Paul Gordon, Cumming School of Medicine's Centre for Health Genomics
Learn about "OpenDBA," GPU-accelerated code we've created to analyze data from handheld nanopore genome sequencing devices. Electrical current measurements from these emerging devices are typically run through neural networks, transcoding to DNA/RNA sequence for biological interpretation (for example, disease agent detection and tracking). We ran thousands of these data series through the canonical DTW Barycenter Averaging (DBA) to identify sequence clusters and to generate high-quality consensus for each cluster. DBA clustering identifies many more viral RNA matches than extant neural network approaches. To show general utility and scalability, we also built consensus audio from 624 native English voice samples (~1M points each) in the Speech Accent Archive.