DeepVariant-on-Spark

DeepVariant-on-Spark performs scalable germline short variant calling from whole-genome sequencing data by integrating Google's DeepVariant image-classification deep neural network with an Apache Spark parallelization framework to accelerate and parallelize variant calling.

Key Features:

• Deep Neural Network Integration: Integrates DeepVariant's image-based deep neural network for germline short variant calling and reports higher precision relative to GATK.
• Apache Spark Parallelization: Uses Apache Spark to parallelize computation and optimize resource allocation for processing whole-genome sequencing data.
• Multi-GPU Support: Supports multi-GPU configurations to accelerate DeepVariant model inference.
• Cloud Deployment: Deployable on Google Cloud Platform (GCP) for cloud-based execution of the Spark-enabled pipeline.
• Scalability and Flexibility: Provides scalability for both small and large sequencing projects through Spark parallelization and cloud deployment.

Scientific Applications:

• Genome projects: High-accuracy germline short variant calling from whole-genome sequencing data for individual genome variant discovery.
• Population genetics: Processing cohort-scale whole-genome datasets to support population genetic studies that require precise variant calls.

Methodology:

Integrates Google's DeepVariant image-classification deep neural network into an Apache Spark execution framework with multi-GPU support and optional deployment on Google Cloud Platform (GCP).