Reproducible Large-Scale Neuroimaging Studies with the OpenMOLE Workflow Management System

Area of Science:

• Scientific Computing
• Computational Science
• Software Engineering

Background:

• Scientific workflows require robust execution across diverse environments.
• Ensuring reproducibility and re-executability in distributed computing is a significant challenge.
• Existing workflow engines often lack seamless cross-platform compatibility.

Purpose of the Study:

• To present OpenMOLE's capabilities for workload distribution and experiment design.
• To detail improvements in workflow re-executability across Linux platforms.
• To evaluate the integration of OpenMOLE with CARE for reproducible scientific pipelines.

Main Methods:

• Utilized OpenMOLE's high-level Domain Specific Language (DSL) built on Scala for workflow design.
• Integrated OpenMOLE with CARE, a containerization solution, for cross-platform application execution.
• Evaluated the solution using a Python-based pipeline with scikit-learn and binary dependencies on High-Performance Computing (HPC) environments.

Main Results:

• Demonstrated successful packaging and re-execution of complex pipelines on various HPC environments.
• Achieved identical numerical results across different computing infrastructures.
• Showcased the flexibility with three neuroimaging pipelines on heterogeneous environments, including the European Grid Infrastructure (EGI).

Conclusions:

• OpenMOLE and CARE together provide a reliable solution for reproducible results and re-executable scientific workflows.
• The integrated solution simplifies scaling pipelines from desktops to large-scale distributed infrastructures.
• This approach enhances the sharing and replication of scientific studies across diverse computing contexts.