Workflow Example: End-to-End Design Loop¶

This workflow combines generation with downstream analysis and prioritization. It is the most complete pattern in the framework and is the closest to how candidates are handled in a real design campaign.

Conceptual Flow¶

[Generate molecules]
        |
        v
[Analyze outputs]
        |
        v
[Filter valid molecules]
        |
        v
[Optimize structures]
        |
        v
[Evaluate with xTB or predictors]
        |
        v
[Rank candidates]

What This Adds Beyond Generation¶

Generation alone is usually not enough. This workflow separates:

molecules that are merely generated,
molecules that are chemically reasonable,
molecules that remain promising after optimization and evaluation.

That distinction is what turns a raw sample set into a candidate-selection pipeline.

Example Decision Table¶

Stage	Main question	Typical outcome
Generate	What can the model propose?	Raw molecule pool
Analyze	Are the structures valid and diverse?	Cleaned subset
Optimize	Do the structures remain sensible after relaxation?	More reliable geometries
Evaluate	Do they satisfy the target property or descriptor?	Scored candidates
Rank	Which molecules deserve follow-up?	Final shortlist

When to Use This¶

Use this workflow when:

you need a ranked candidate list rather than a raw sample set,
downstream physics- or chemistry-based checks matter,
you want to compare candidates consistently across multiple filters.

Where to Go Next¶