Hummingbird

Hummingbird predicts cloud instance performance for genomic workflows to select optimal CPU and memory configurations that minimize runtime and cost for pipelines such as GATK HaplotypeCaller, GATK Mutect2, and ENCODE ATAC-seq.

Key Features:

• Performance prediction: Predicts runtime performance of computing instances across multiple cloud platforms given memory and CPU specifications.
• Cost-efficiency forecasting: Forecasts fastest, cheapest, and cost–performance tradeoffs among instance types to inform cost-minimizing choices.
• Pipeline validation: Validated on genomic workflows including GATK HaplotypeCaller, GATK Mutect2, and ENCODE ATAC-seq.
• Workflow format support: Parses applications specified in command-line JSON and Workflow Description Language (WDL) formats.
• Implementation and benchmarking: Implemented in Python and performs experimental benchmarking across various cloud platforms and instance types.

Scientific Applications:

• Genomic resource optimization: Selects cloud instance configurations to reduce trial-and-error and optimize resource allocation for genomic analyses.
• Cost and runtime optimization for specific pipelines: Enables selection of instances that minimize runtime or cost for GATK HaplotypeCaller, GATK Mutect2, and ENCODE ATAC-seq workflows.

Methodology:

Conducts experiments across various cloud platforms evaluating instance types by memory and CPU configurations and uses the measured performance and cost data to predict optimal instances for specific genomic workflows.