Sovereign 8s.is mirror of standardgalactic/secs
  • Jupyter Notebook 89.5%
  • Python 10.5%
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2026-02-12 10:31:35 +00:00
.github Create dependabot.yml 2024-09-04 09:06:51 +02:00
app feat: push 13C augment 2025-10-03 15:53:38 +00:00
configs feat: push 13C augment 2025-10-03 15:53:38 +00:00
data feat: analysis scripts, example of results (#91) 2025-08-27 14:40:11 +02:00
experiments feat: analysis scripts, example of results (#91) 2025-08-27 14:40:11 +02:00
notebooks feat: make plots for human eval results (#90) 2025-08-27 14:40:41 +02:00
src/molbind feat: add encoder 2025-10-03 19:06:45 +00:00
tests feat: add image encoder and dataset (#68) 2024-06-27 13:54:34 +02:00
.gitignore feat: add hsqc encoder and (chore): clean-up unused directories 2025-06-05 08:13:54 +00:00
.project-root Initial commit 2024-03-08 11:55:32 +01:00
open_source_licenses.txt feat (major update): removed geometric dependencies, simplified package installation 2025-05-18 15:47:55 +00:00
pyproject.toml feat: force transformer version 2026-02-12 10:31:35 +00:00
README.md feat: analysis scripts, example of results (#91) 2025-08-27 14:40:11 +02:00

Structure Elucidation from Chemical Spectra (SECS)

📜 Installation guide

It is recommmended to use uv for creating a virtual environment. The instructions to install uv can be found on uv's homepage.

uv venv --python 3.11 molbind
source molbind/bin/activate
uv pip install -e .

📁 Data availability

The simulated spectra data have been compiled from IBM's Multimodal Spectroscopic Dataset.

📋 Environment file

Your environment file should look like this:

WANDB_PROJECT="<your-wandb-project-name>"
WANDB_ENTITY="<your-wandb-account-name>"
TOKENIZERS_PARALLELISM=False

After you have defined your system variables in .env, it is read into the script as following:

load_dotenv("path/to/.env")

📉 Train models

The experiment configs can be found at config. For example, to run the train.py

python train.py 'experiment="train/ir_simulated_large_dataset"'

The training scripts outputs the checkpoints at experiments/checkpoints/<run-code-name>/<checkpoint-file-name>.ckpt To find all three checkpoints used in this work, please access the supplementary information on Zenodo.

To run the metrics on these experiments:

python retrieval.py 'experiment="metrics/ir_simulated_large_dataset"'

⚙️ System requirements

For the training script 4 NVIDIA A100-40GB GPUs have been used. For the retrieval script 1 NVIDIA A100-40GB GPUs has been used.

💰 Funding

This work was funded by the Carl-Zeiss Foundation. In addition, this work was partly funded by the SOL-AI project funded as part of the Helmholtz Foundation Model Initiative of the Helmholtz Association. Moreover, this work was supported by Helmholtz AI computing resources (HAICORE) of the Helmholtz Association’s Initiative and Networking Fund through Helmholtz AI.

Citation

@article{mirza2024elucidating,
  title={Elucidating structures from spectra using multimodal embeddings and discrete optimization},
  author={Mirza, Adrian and Jablonka, Kevin Maik},
  year={2024}
}
@inproceedings{mirza2024bridging,
  title={Bridging chemical modalities by aligning embeddings},
  author={Mirza, Adrian and Starke, Sebastian and Merdivan, Erinc and Jablonka, Kevin Maik},
  booktitle={AI for Accelerated Materials Design-Vienna 2024},
  year={2024}
}