GRIP: physics-informed neural network for gradient retention time prediction in liquid chromatography

GRIP is a physics-informed deep learning model for retention time prediction in reversed-phase gradient liquid chromatography (RP-HPLC). The model combines molecular representations learned from chemical structures with explicit chromatographic system parameters and the fundamental equation of gradient elution, enabling retention time prediction across diverse chromatographic conditions. You can read more about the model in our preprint.

Installation

git clone https://github.com/gurevichlab/grip.git
cd grip

conda env create -f environment.yml
conda activate grip

pip install -e .

Run inference on test data

The repository includes a small example input in test_data/ and a pretrained model checkpoint in model/.

python grip_inference.py \
  --input test_data/smiles.tsv \
  --id-col id \
  --smiles-col smiles \
  --system-cfg test_data/system_100mm_18min.yaml \
  --ckpt model/model-i6kwrq9u.ckpt \
  --out predictions.tsv \
  --device cpu

The output file predictions.tsv contains predicted retention times for all compounds in the input file.