Nuclear magnetic resonance (NMR) spectroscopy provides us a powerful tool to analyze mixtures consisting of small molecules but is difficult to identify compounds in mixtures because of chemical shift variation of the same compound in different mixtures and peak overlapping among molecules. We presented a pseudo Siamese convolutional neural network method (pSCNN) to solve the problems of compound identification in NMR spectra of mixtures. This is the code repo for the paper Deep Learning-based Method for Compound Identification in NMR Spectra of Mixtures.
Python and TensorFlow:
Python 3.8.13 and TensorFlow (version 2.5.0-GPU)
The main packages can be seen in requirements.txt
-
Install Anaconda https://www.anaconda.com/
-
Install main packages in requirements.txt with following commands
conda create --name pSCNN python=3.8.13 conda activate pSCNN python -m pip install -r requirements.txt pip install tqdm
git clone at:https://github.com/yuxuanliao/pSCNN.git
Since the model exceeded the limit, we have uploaded the model and all the NMR data to Zenodo.
Training your model and predict mixture spectra data
Run the file 'pSCNN.py'. The model and these data have been uploaded at Zenodo. Download the model and these example data, pSCNN can be reload and predict easily.
Yuxuan Liao: [email protected]