Application: Library Design with Structure-Directed Generation¶

This workflow uses structure-directed generation to build a focused virtual library from a predefined scaffold. In practice, this usually means outpainting from a fixed core, optionally combined with filtering and descriptor calculations to cover a desired design space.

When to Use This¶

Tip

Use library design when you want to:

keep a core scaffold fixed across all generated candidates,
enumerate chemically diverse decorations around selected growth sites,
build a screening library with broad steric or physicochemical coverage.

The BINOL-Cp example from the manuscript follows this pattern: a fixed ligand scaffold is expanded into a broader virtual library, then analyzed with downstream steric descriptors.

Library design workflow

Conceptual view of the library-design workflow from the paper: scaffold-based generation is used to create Cp ligand variants, which are then optimized and analyzed in descriptor space.

Conceptual Workflow¶

  [Choose scaffold XYZ file]
            |
            v
[Define connector / growth atoms]
            |
            v
 [Outpainting: fix core scaffold,
  generate decorations]
            |
            v
  [Filter + analyze: steric,
   physicochemical coverage]
            |
            v
    [Virtual screening library]

The point of this workflow is to balance control and diversity. You keep the part of the molecule you care about, but allow the model to explore decorations around it.

Why This Matters¶

Note

This is a useful mode when you want to move from a single scaffold to a screening set. It helps answer questions such as:

How can I expand a known core into a more diverse virtual library?
Which substitutions give broad steric or property coverage?
How can I generate candidates that are related enough to compare, but diverse enough to screen?

Where to Go Next¶

See also

For the actual workflow, use:

Tutorial 6: Structure-Guided Generation — outpainting
Tutorial 9: Analysis — filtering and evaluating the generated library

Starting template: docs/cfg_examples/gen_outpaint.yaml

This repository also includes application-specific helper scripts for the virtual-library example:

Script	Purpose
`scripts/applications/virtual_lib_cp/filter_cp.py`	Filter generated Cp ligand candidates
`scripts/applications/virtual_lib_cp/compute_steric_descriptor.py`	Compute steric descriptors
`scripts/applications/virtual_lib_cp/process_dir.py`	Batch-process a directory of structures