Curated workflow catalogs are central to modern scientific computing, yet the reuse of existing components remains surprisingly limited. This challenge is further amplified by AI coding assistants, which frequently generate custom logic from scratch because their optimization objectives do not explicitly reward component reuse. This work introduces a unified framework for evaluating, recommending, and adapting reusable workflow components for both human developers and AI agents. At its core is the Missed Reuse Score (MRS), a novel metric that estimates the extent to which a workflow step represents a missed opportunity to reuse an existing catalog component. An empirical analysis of community Nextflow pipelines shows that 76% of process steps are custom-built. However, 29% of all process steps have MRS values exceeding 0.5, indicating a greater-than-even likelihood of missed reuse opportunities. We also propose MRS-centered AI-assisted workflow composition tools, graph-based recommenders, and automated adapters to help users realize this reuse potential in practice.
Quantifying reusability in scientific workflows
Mulone Alberto
First
;Colonnelli Iacopo;Santimaria Marco Edoardo;Aldinucci Marco;
2026-01-01
Abstract
Curated workflow catalogs are central to modern scientific computing, yet the reuse of existing components remains surprisingly limited. This challenge is further amplified by AI coding assistants, which frequently generate custom logic from scratch because their optimization objectives do not explicitly reward component reuse. This work introduces a unified framework for evaluating, recommending, and adapting reusable workflow components for both human developers and AI agents. At its core is the Missed Reuse Score (MRS), a novel metric that estimates the extent to which a workflow step represents a missed opportunity to reuse an existing catalog component. An empirical analysis of community Nextflow pipelines shows that 76% of process steps are custom-built. However, 29% of all process steps have MRS values exceeding 0.5, indicating a greater-than-even likelihood of missed reuse opportunities. We also propose MRS-centered AI-assisted workflow composition tools, graph-based recommenders, and automated adapters to help users realize this reuse potential in practice.| File | Dimensione | Formato | |
|---|---|---|---|
|
escience_2026___poster_A.pdf
Accesso aperto
Tipo di file:
PREPRINT (PRIMA BOZZA)
Dimensione
222.64 kB
Formato
Adobe PDF
|
222.64 kB | Adobe PDF | Visualizza/Apri |
|
Quantifying_reusability_in_scientific_workflows.pdf
Accesso riservato
Tipo di file:
PDF EDITORIALE
Dimensione
287.89 kB
Formato
Adobe PDF
|
287.89 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



