Version 0.1.0

v0.1.0 — Initial open-source release

GRACE (Geometric Residual Adduct Conditioning via Early-fusion) is a 3D
collision cross section (CCS) predictor for ion mobility mass spectrometry.

For full details see our manuscript: arXiv:2609.12223

What's included

Source code

• Full model and training code under src/ccs3d/
• scripts/fetch_sample_data.py — downloads a 185-compound sample dataset
• scripts/run_external_eval.py — evaluates a trained checkpoint on an external molecule set

Attached model files

• random_single_residual.ckpt — best-validation checkpoint trained on the full dataset (random split, single conformer, residual target)
• ridge_model_random.pkl — paired Ridge regression baseline required when evaluating a residual-target checkpoint

See the README for install instructions, training details, and evaluation usage.