Evaluating 2D and 3D Graph Encoders for Infrared-Based Organic Structure Verification

Setup

make setup

This uses uv to create .venv and install dependencies. If you prefer a manual setup:

python3.9 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Repo layout

src/pasgen/          # core code
analysis/            # analysis outputs (generated)
manuscript/          # LaTeX sources + figures/tables
configs/             # pipeline configuration
scripts/             # workflow entry points
assets/              # static resources (embeddings)

Data sources

For full reproduction, you need two things:

1. NIST raw data (not bundled):

• jdx/ and mol/ directories (downloaded from NIST via the BeautifulSoup script linked in 10.1039/d3dd00254c)
• place them under pasgen_data/jdx and pasgen_data/mol

2. Paper artifacts (Zenodo, auto-downloaded):

• prediction archive and xTB/DFT calculation bundle from Zenodo record 18471841

Reproduce manuscript

Run:

make paper

make paper runs the full pipeline (./scripts/run_manuscript_pipeline.sh) and compiles manuscript/build/.

If prediction/calculation artifacts are missing, the script downloads them from Zenodo automatically.