Skip to content

Quick Start

This page walks from a fresh launch to a generated molecule and a simple visualization in ~15 minutes. It assumes Installation is complete.

1. Launch

conda activate molcraft
./dev.sh

Open http://localhost:8000. The app opens on the Management tab by default.

2. Generate a molecule

2a. Open the generation tab

Click 3D molecule generation in the tab bar.

MolCraftDiff not installed

If a yellow banner appears, the MolCraftDiff CLI is missing. Install it following Installation step 2 and restart the backend.

2b. Select a model

The Models panel (left column) lists every checkpoint folder found under MOLCRAFT_MODELS_DIR. Models labelled CFG support property-targeted generation; models labelled Unconditional generate freely. Click one to select it.

If the list is empty, check that your checkpoint folders each contain edm_chem.pkl and that MOLCRAFT_MODELS_DIR points to the right location (shown as grey text under the model list).

2c. Configure parameters

Keep defaults for the first run:

  • Total molecules: 1
  • Diffusion steps: 50
  • Size mode: random
  • CFG scale: 1 (if the model is conditional)

2d. Run

Click Generate (top-right of the centre panel) or press Shift+Enter.

The Results panel (right column) shows: - A status badge cycling queued → running → completed - A live log tail updating every 2 seconds

2e. Inspect the result

When the status shows completed, the molecule name appears as a pill in the results list. Click the pill to load it.

In the viewer:

  • Rotate: left-click drag
  • Zoom: scroll wheel
  • Download the structure: click XYZ or SVG below the viewer

To compare multiple molecules at once, change Split to 2–9 and click additional pills.

3. Load the result into Management

Click Use as ref (→ button) to send the molecule to the Structure-directed generation tab — or follow the steps below to load the whole job as a dataset for visualization.

  1. Click the Management tab.
  2. In Add generated molecules, click Refresh if your new job is not listed.
  3. Select the Model, Date, and Token / run that match the completed generation job.
  4. Keep ID prefix as molGen, or change it if you want a different row-id prefix.
  5. Set Data source label to a short name such as generated_1.
  6. Click Register generated molecules. The generated output is added as a staged source.
  7. Click Compile dataset.

A progress bar appears while the dataset is built. On completion, the header card shows the molecule and column count.

Loading non-generated data

For existing data, use the source registration panel instead. Choose CSV + XYZ folder or ASE database (.db), pick files or paste paths, then click Register source to stage and inspect columns or Register and compile for a one-step load.

4. Explore in Visualization

  1. Click the Visualization tab.
  2. In the action bar, click 2D Scatter. A panel appears with the first two numeric columns on the axes.
  3. Click the X or Y axis label to change the column binding.
  4. Left-click drag to draw a lasso and select a subset of molecules.
  5. The built-in 3D Structure Viewer panel updates to show the selected molecule.

Next steps

What you want to do Where to go
Tune generation parameters De-novo generation
Generate around a known scaffold Structure-guided generation
Compute xTB properties or UMAP layout Analysis tools
Build a dataset from multiple sources Data Manager
Full end-to-end example workflows Workflows