Skip to content

Protein Diffusion Example¤

Status: Exploratory workflow Device: CPU-compatible

This walkthrough uses real lower-level Artifex protein structure owners around synthetic data and padded batches, but it does not demonstrate a shipped high-level Artifex protein diffusion API. The historical filename is retained for continuity with the existing example tree, while the published contract is now explicit: this is an exploratory direct-owner workflow rather than a canonical protein-diffusion tutorial.

Files¤

Run It¤

python examples/generative_models/protein/protein_diffusion_example.py
jupyter lab examples/generative_models/protein/protein_diffusion_example.ipynb

What This Workflow Actually Uses¤

  • ProteinPointCloudModel and ProteinGraphModel for direct model construction
  • ProteinDataset plus create_synthetic_protein_dataset for retained synthetic protein data
  • protein_collate_fn for padded batch creation
  • create_protein_structure_loss for protein-specific loss evaluation
  • optional visualization helpers from artifex.visualization.protein_viz

Current Data And Extension Contract¤

The retained data and extension path is still explicit in the paired source:

from artifex.data.protein import ProteinDataset, ProteinDatasetConfig
from artifex.generative_models.core.configuration import (
    ProteinDihedralConfig,
    ProteinExtensionConfig,
    ProteinExtensionsConfig,
)

dataset_config = ProteinDatasetConfig(max_seq_length=64)
dataset = ProteinDataset(dataset_config, data_dir="data/proteins")

extensions = ProteinExtensionsConfig(
    name="protein_extensions",
    backbone=ProteinExtensionConfig(name="backbone", weight=1.0),
    dihedral=ProteinDihedralConfig(name="dihedral", weight=0.3),
)

config = ProteinPointCloudConfig(
    name="protein_point_cloud_exploratory",
    extensions=extensions,
)

Why It Is Exploratory¤

  • it does not demonstrate a shipped high-level Artifex protein diffusion API
  • it focuses on lower-level direct-owner surfaces and synthetic data instead of a polished public protein-diffusion facade
  • it is useful for contributors inspecting the retained protein model and batch contracts, but it should not be read as the canonical end-to-end protein example for the library

Use This When¤

Use this pair if you want to explore the retained protein point-cloud and graph owners, inspect the current padded-batch contract, or prototype protein loss and visualization work on top of synthetic inputs. For retained protein tutorials with a narrower public claim, use the other protein pages in the main catalog.