DeepSPInN

A framework that predicts the molecular structure when given Infrared and 13C Nuclear magnetic resonance spectra without referring to any pre-existing spectral databases or molecular fragment knowledge bases

Setting up dependencies

The dependencies required for running this code can be installed by setting up an Anaconda environment.

conda env create -f custom_environment.yml

If making an environment from the .yml file does not work, the script make_environment.sh contains the commands to set up a minimal environment that worked on an environment with CUDA 10.2. If it is being run on a HPC with a Slurm workload manager, the CUDA 10.2 toolkit can be loaded with the module load... command that is commented out in the script.

Data

The data required for training and testing the code can be downloaded by following the instructions in the data folder.

Training Code

cd train
ray start --head --num-cpus=39 --num-gpus=4 --object-store-memory 50000000000
python parallel_agent.py

Testing Code

cd test
ray start --head --num-cpus=40 --num-gpus=4 --object-store-memory 50000000000 
python parallel_agent.py -s MOL_START_INDEX -e MOL_END_INDEX -d test

MOL_START_INDEX and MOL_END_INDEX are the start and end indices in the test dataset.

For testing on a HPC with a Slurm workload manager, the script test/sbatch_test.sh is a good starting point for an SBATCH script that tests the code for 1000 molecules in the test set.

The script creates log files in the folder test/test_outputs/. These log files can be analysed with the script test/test_outputs/check_results.py. The log files for all the experiments and main results are in miscellaneous/results_deepspinn.

Miscellaneous

The code to create all the images and results presented in the mansucript is in miscellaneous/Illustrations.ipynb.