kochgroup / neural_npfp

Repository to recreate experiments and calculate the Neural Natural Product Fingerprint.

Neural Natural Product Fingerprint

This is the accompanying Repository for our work on "Natural Product Scores and Fingerprints Extracted from Artificial Neural Networks." https://doi.org/10.1016/j.csbj.2021.07.032

The code can be used to reproduce our results, retrain the models and compute fingerprints for your own SMILES-string.

Installation

1. Download the Repository

2. Download the data here

   1. extract the data.zip file into the same repository

   |- data
   |- neural_npfp
   |- results
   |- settings
   

3. Create and Activate Conda Environment: Navigate to the Folder containing the environment.yml

   conda env create -f environment.yml
   conda activate neural_npfp_env
   

   Please install Pyorch with you appropriate choice of cuda.

   conda install pytorch==1.7.0 cudatoolkit=*Your Version* -c pytorch
   

Data

The data folder contains multiple datasets in case you are interested in just the dataset consisting of the NP and synthetic compounds from ZINC please use the data/coconut_synthetic.csv The validation data collected by us can be found in data/validation_sets/np_target_identification/

• fps_targets contain the precomputed ECFP for each target
• smiles_targets contains the SMILES.

The clean_task1.csv and clean_task2.csv were not created by us, we did not include the SMILES for those compounds. If you are interested in the actual compounds we refer to the original publication.

Seo, M.; Shin, H.K.; Myung, Y. et al. Development of Natural Compound Molecular Fingerprint (NC-MFP) with the Dictionary of Natural Products (DNP) for natural product-based drug development. Journal of Cheminformatics. 2020, 12(6) https://doi.org/10.1186/s13321-020-0410-3

The data was collected with the help of:

• Sterling, T.; Irwin, J. J. ZINC 15–ligand Discovery for Everyone. Journal of Chemical Information and Modeling. 2015, 55, 2324–2337.

• Sorokina, M.; Merseburger, P.; Rajan, K.; Yirik, M. A.; Stein-beck, C. COCONUT online: Collection of Open Natural Products Database. Journal of Cheminformatics 2021, 13, 1–13.

• Mendez, D. et al. ChEMBL: Towards Direct Deposition of Bioassay Data. Nucleic Acids Research. 2018, 47, D930–D940.

• Zeng, X.; Zhang, P.; He, W.; Qin, C.; Chen, S.; Tao, L.;Wang, Y.; Tan, Y.; Gao, D.; Wang, B.; Chen, Z.; Chen, W.;Jiang, Y. Y.; Chen, Y. Z. NPASS: Natural Product Activity and Species Source Database for Natural Product Research, Discovery and Tool Development. Nucleic Acids Research. 2018, 46, D1217–D1222.

Experiments

To reproduce the results, please run:

python experiment.py

This will train the models again. The hyperparameters used can be found in settings/settings.yml You can also change the hyperparameters. We recommend creating a new settings.yml If you want to use your own settings.yml please run. The trained models will be saved in data/trained_models/*folder name specified in settings.yml*

python experiment.py --input *path to your settings.yml*

After training the results script will perfrom the similarity search and reproduce some of the graphics used.

python results.py

In results/plots the plots and some addtionall files will be saved. In case you want to evaluate models that you trained with experiment.py but a different settings.yml Run:

python results --input *path to the folder containing the models*

Generate Fingerprints

You can use a csv file containing a column with SMILES strings as input to our model. Naviagte to *your path*/neural_npfp/neural_npfp and run:

python get_fp.py ../data/testdata.csv -s smiles

smiles refers to the name of the column containing the SMILES strings.

By default the fingerprints using the NP_AUX model. If you want to use a different model use the flag -m followed by either ae, aux or base.

python get_fp.py ../data/testdata.csv -s smiles

You can also provide the Index of the column containing the SMILES.

python get_fp.py ../data/testdata.csv -s 0

If you do not have a header add the -n flag

python get_fp.py ../data/testdata.csv -s 0 -n

Perform a Similarity Seach based on the produced Fingerprint

If you want to use the NNFPs for a similarity search make sure that the query is in the same file as the molecules you want to screen before you generate the Fingerprints.

With python simsearch.py *path of fingerprintfile* -q *index of the query* the similairty search can be performed

Given you generated fingerprints for the testdata.csv. The following code will perform a similarity search for the query with index 0

python simsearch.py ../data/aux_npfp_*date*.csv -q 0

You can also perform a similarity search for multiple queries by adding addtional indices.

python simsearch.py ../data/aux_npfp.csv -q 0 15 8 1 84

A new folder will be generated containing the results of the similairty search Like in the original paper, the cosine similarity is used for the search.

You can add the -d to not include the other queries in the similarity search.

neural_npfp's Issues

About data and pre training model

Hello, can you upload the supporting data and pre training model, otherwise the experiment will be difficult to reproduce