This is the code used in the paper https://pubs.rsc.org/en/content/articlelanding/2019/fd/c8fd00227d. The data in the pboldus project are the example for P. boldus. Running the script "nmrfilter pboldus" will produce the list of hits for P. boldus.

Requirements are Java and Python. For Java, version 1.8 or higher is needed. A JRE (Java Runtime Environment) is enough, a JDK is not required. The default of any operating system should do.

Python version must be 3 (3.66 was tested). You need numpy, scipy, louvain, and python-igraph libraires. The following alternatives exist

Using pip, you can install them by doing

pip3 install numpy

pip3 install scipy

pip3 install python-igraph

pip3 install louvain

Notice igraph is a different library. If install python-igraph gives an error about missing C libraries, try using a wheel, following https://stackoverflow.com/questions/34113151/how-to-install-igraph-for-python-on-windows

There is an environment file for anaconda. See next section for an explanation. The environment can also be used if respredict is not to be used.

On linux, it might be necessary to change the nmrfilter.sh file to executable with "chmod a+x nmrfilter.sh".

If you want to run the jupyter notebook code, you need to run

jupyter nbextension enable --py widgetsnbextension

before starting jupyter.

If you want to use the respredict prediction (https://jcheminf.biomedcentral.com/articles/10.1186/s13321-019-0374-3) giving better results, you need to install more packages. The easiest way is to use the two yaml files environment-cpu.yml (for usage of the CPU only) or environment.yml (for using the GPU). They use Anaconda to install an environment. The command is conda env create -f environment-cpu.yml respectively conda env create -f environment.yml. You can then activate the environment with conda activate nmrfilter.

The program works on projects, where each project is a folder. It must contain the required files (see below) and any results will be written to it. If you checkout the repository https://github.com/stefhk3/nmrfilterprojects you can use this as example projects.

The settings for a run are contained in the nmrproc.properties file. It also gives the names of the input/output files. You can change these, but you do not need to do so. The property datadir is where the projects/folders are searched for. If you have downloaded the nmrfilterprojects examples, set this to the nmrfilterprojects directory. directory. The following data/files need to be supplied to run an anlysis in the project folder you want to work on:

• A list of candidate SMILES. This is in the files specified by the msmsinput property (default testall.smi). This must have one structure per line.
• The measured spectra in spectruminput (default realspectrum.csv). This must be a list of shifts, separated by tab. One row is one shift. As a standard, HMBC and HSQC shifts are included here.
• Set solvent property to the solvent used if it is Methanol-D4 (CD3OD) or Chloroform-D1 (CDCl3). Otherwise, use Unreported.

Once these files are in place, run nmrfilter.sh <projectname> (linux) or nmrfilter.bat <projectname> (windows). This should produce the result list. Replace by the name of the project/folder you want to work on.

The following features are optional:

• You can include HSQCTOCSY shifts. For this, set usehsqctocsy=true and include the HSQCTOSY shifts in the spectruminput file.
• You can produce some debug output by setting debug=true. You need a file called testallnames.txt for this, which has the names of the compounds in the same order as in the msmsinput file.
• You can set paraemteres for tolerances and resolutions. Normally these do not need to be modified.
• With usedeeplearning=true you can activate the respredict prediction (https://jcheminf.biomedcentral.com/articles/10.1186/s13321-019-0374-3) instead of the HOSE code based one. This gives better results, but requires the installation as described above.