Open source and reproducible and inexpensive infrastructure for data challenges and education

Area of Science:

• Biomedical research
• Data science
• Computational biology

Background:

• Data sharing is crucial for maximizing research knowledge and enabling secondary analyses.
• Current biomedical data challenges often require expensive cloud computing and industry partnerships, creating financial barriers.
• There is a need for cost-effective and computationally accessible methods for data sharing and hosting data challenges.

Purpose of the Study:

• To develop an inexpensive and computationally lightweight workflow for reproducible model training, testing, and evaluation.
• To demonstrate the utility of this workflow by conducting a data challenge.
• To provide a resource for researchers seeking accessible infrastructure for data sharing and data challenges.

Main Methods:

• Leveraged public GitHub repositories for code sharing and version control.
• Utilized open-source computational languages for model development and analysis.
• Employed Docker technology for containerization, ensuring reproducibility and portability.
• Developed a workflow for reproducible model training, testing, and evaluation.

Main Results:

• Successfully developed and implemented a cost-effective infrastructure for hosting a data challenge.
• The workflow enabled reproducible model training, testing, and evaluation.
• The developed infrastructure and workflow proved effective for the conducted data challenge.

Conclusions:

• The developed workflow and infrastructure offer an accessible alternative to expensive cloud-based solutions.
• This approach facilitates data sharing and the hosting of data challenges in biomedical research.
• The methodology is likely to be valuable for both data challenges and educational purposes in science.