Table 1:
Comparison of existing projects related to workflow execution of neuroimaging pipelines
| Feature | SABER | CWL-Airflow | TOIL | Galaxy | Air-tasks | Kubernetes |
|---|---|---|---|---|---|---|
| Purpose | Workflow management system | Workflow management system | Workflow management system | Workflow management system | Workflow management system | Distributed container orchestration |
| Container support | Yes | Yes | Yes | Yes | Yes | Yes |
| Workflow description | CWL | CWL | CWL or WDL | Custom (CWL beta) | Custom Python | NA |
| Computational background | Novice–expert | Novice–expert | Novice–expert | Novice–expert | Intermediate–expert | Expert |
| Installation | docker-compose | pip install | pip install | Install scripts | docker-compose | Cluster configuration |
| Cloud support | AWS | Planned | Multiple cloud providers | Multiple cloud providers | Docker Infrakit | Multiple cloud providers |
| Volumetric database | bossDB, DVID | None | None | None | cloud-volume | NA |
| Parallel processing model | Block-merge | None | None | None | Block-merge | NA |
| Workflow deployment for neuroimaging | Yes | No | No | No | Yes | NA |
| Workflow optimization for neuroimaging | Yes | No | No | No | No | NA |
| Tool benchmarking and datasets | Yes | No | No | No | No | NA |
| EM tool library | Yes | No | No | No | Yes | NA |
| Tools for other modalities | Yes | No | No | No | No | NA |
SABER delivers integrated containerized tools, a standardized workflow and tool description, and a volumetric database. It also provides tools for automating deployment over datasets by dividing into blocks (block-merge) and optimization of workflows. The most comparable tools are other workflow management systems such as CWL-Airflow, TOIL, Galaxy, and Air-tasks. Air-tasks provides similar capabilities but lacks support for common workflow descriptions and tool optimization and offers less flexibility for users. Similar projects such as TOIL, Galaxy, and CWL-Airflow lack neuroimaging-specific features to enable the use cases described in Methods. Scalable cluster systems, such as Kubernetes, provide essential functionality to deploy containers at scale but need capabilities built to manage workflows and data movement and are complementary to workflow management systems such as SABER. The SABER project adds critical features for neuroimaging by (i) interfacing with existing solutions; (ii) providing a library of portable, Dockerized neuroimaging tools; and (iii) providing scripting to analyze large-scale neuroimaging datasets. NA: not applicable; WDL: Workflow Definition Language.