Tackling analytical variability with Workflomics

Data analysis workflows are characterized by combining multiple software tools. Their selection and composition into workflows is both complex and critical for scientific quality, yet workflow design is typically approached based on intuition, experience and subjective decision-making, rather than taking a systematic approach that takes into account the enormous analytical variability enabled by today’s research software ecosystem. This paper introduces Workflomics, a novel framework that addresses this issue by enabling systematic generation and evaluation of workflows, thereby improving the quality of data analysis workflows and the robustness of research results. Workflomics automates the process of creating multiple candidate workflows, supports their evaluation using different design-time and runtime metrics, and helps to select the most optimal ones based on scientific benchmarking. We demonstrate the practical utility of Workflomics through a reference implementation focused on computational proteomics, featuring tools from bio.tools, the EDAM ontology, automated workflow generation via APE, and container-based CWL workflows.