Workflow Example: Generate and Filter¶

This is the simplest end-to-end workflow. You start from a pre-trained diffusion model, generate a broad pool of molecules, and then filter the output based on validity, uniqueness, or downstream criteria.

When to Use This¶

Use this workflow when:

you want a fast baseline,
you do not yet have a task-specific fine-tuned model,
your first goal is to understand what the pre-trained model can already produce.

Conceptual Flow¶

[Pre-trained diffusion model]
            |
            v
   [Generate molecules]
            |
            v
   [Analyze and filter]
            |
            v
      [Candidate set]

What This Gives You¶

This workflow keeps the first pass simple. You use the pre-trained model to produce a broad set of structures, then use analysis and filtering to decide which molecules are worth keeping.

Typical Outcome¶

You usually obtain:

a sense of the chemical space already covered by the model,
a baseline validity and diversity profile,
an initial candidate pool for more focused follow-up.

Where to Go Next¶