OpenSciFlow Workflow Templates

Reusable workflow templates for AI for Science tasks, starting with protein-computing workflows.

What is a workflow template?

A workflow template describes a scientific task as ordered steps or a DAG. It references plugins, declares fallback tools, estimated runtime, hardware requirements, reproducibility metadata, and report templates.

Plugin manifests describe how one tool/model runs. Workflow templates describe how tools are composed into a scientific task.

Status

Draft v0.1. These templates are intended for review and iteration.

Current count: 7 draft workflow templates across protein computing and materials/atomistic modeling.

Required fields

Field	Purpose
schema_version	Template schema version
workflow_name	Stable workflow identifier
domain	Scientific domain tags
task_description	Human-readable task
inputs	Required/optional inputs
outputs	Expected artifacts
steps	Ordered task steps
dag	Step dependencies
plugins	Required and optional plugins
fallbacks	Alternative execution paths
estimated_runtime	Runtime estimates
hardware	CPU/memory/GPU requirements
reproducibility	Metadata that must be recorded
report_template	Report template path
example_dataset	Example data information
limitations	Scientific and technical caveats

Initial protein templates

• protein/protein-ligand-stability.yaml
• protein/md-stability-analysis.yaml
• protein/mutation-impact-analysis.yaml
• protein/ptm-dynamics-analysis.yaml
• protein/conformational-ensemble-generation.yaml
• protein/gromacs-rmsd-slurm.yaml

Initial materials templates

• materials/mace-evaluation-slurm.yaml

Review checklist

Short version before merging a template:

• Is the scientific task boundary clear?
• Are inputs and outputs concrete?
• Are required and optional plugins declared?
• Is there a fallback path?
• Is the DAG acyclic?
• Are runtime and hardware estimates included?
• Does the template record reproducibility metadata?
• Is there a report template path?
• Are limitations explicit?
• Does the template avoid overstating computational results as experimental proof?

Full checklist:

• docs/workflow-review-checklist.md

Related conventions:

• docs/report-template-convention.md
• docs/example-dataset-policy.md
• docs/dag-validation.md
• docs/artifact-handoff-validation.md
• docs/reproducibility-validation.md
• docs/protein-template-review-matrix.md

HPC-specific draft:

• protein/gromacs-rmsd-slurm.yaml
• materials/mace-evaluation-slurm.yaml

First reference workflow

The first recommended demo workflow is:

molecular-dynamics-stability-analysis

It analyzes an existing protein MD trajectory and produces RMSD, RMSF, radius of gyration, plots, a report, and a run manifest.