CCS prediction from SMILES using deep neural network.

For the impatients

After installation, go to the DeepCCS/interface/ directory.

DeepCCS predict -i INPUT

• INPUT is the input file with at least a “SMILES” and an “Adducts” column

The default model and encoders will be used. See the predict section bellow for more options.

Installation

DeepCCS was tested and works under Python 3.6. We recommend to use conda.

Package required:

• Numpy
• Pandas
• Scikit-learn
• Tensorflow
• Keras

To install, go to the core directory and simply perform the following command using a terminal:

python setup.py install

Functionalities

On Windows operating system, the symbolic link DeepCCS will not work and user should use the command_line_tool.py script in order to use DeepCCS

Predict

Predict CCS using a SMILES and an adduct.

DeepCCS predict -mp MODEL_DIR -ap ADDUCTS_ENCODER_DIR -sp SMILES_ENCODER_DIR -i INPUT_F -o OUTPUT_F

Required args :

• i : The input file, with at least a “SMILES” and a “Adducts” columns

Optionnal args :

• mp : Directory containing the model.h5 file (default="../saved_models/default/")
• ap : Directory containing the adducts_encoder.json file (default="../saved_models/default/")
• sp : Directory containing the smiles_encoder.json file (default="../saved_models/default/")
• o : Desired name for the output file(ex: MyFile.csv), if none stdout will be used.

Compare

Compare provided CCS values to the ones contained in every dataset used to train and test DeepCCS (no predictions involved).

DeepCCS compare -f H5_F -i REFERENCE_F -d S_DATASET1,S_DATASET2,... -o OUTPUT_P

Required args :

• i : The reference file, with at least “SMILES”, “Adducts” and “CCS” columns
• f : The hdf5 file containing all the source datasets

Optionnal args :

• d : Names of the source datasets (as a list without “ ”) to use for comparison, if none they are all considered.
  • Choices are : MetCCS_pos, MetCCS_neg, Agilent_pos, Agilent_neg, Waters_pos, Waters_neg, PNL, McLean, CBM
• o : Desired prefix for the output files (ex: compare_to_MetCCS_), because there is one output file per compared source dataset

Evaluate

Perform CCS predictions and evaluate the model using measured values

DeepCCS evaluate -mp MODEL_DIR -ap ADDUCTS_ENCODER_DIR -sp SMILES_ENCODER_DIR -i REFERENCE_F -o OUTPUT_F

Required args :

• i : Input reference file. Must contain at least “SMILES”, “Adducts” and “CCS” columns

Optionnal args :

• mp : Directory containing the model.h5 file (default="../saved_models/default/")
• ap : Directory containing the adducts_encoder.json file (default="../saved_models/default/")
• sp : Directory containing the smiles_encoder.json file (default="../saved_models/default/")
• o : Desired name for the output file (ex: MyFile.csv), if not specified stdout will be used.

Train

Train a new model including your own measurements with or without the available datasets.

DeepCCS train -f H5_F -ap ADDUCTS_ENCODER_DIR -sp SMILES_ENCODER_DIR -mtrain -pnnl -cbm -mclean -o OUTPUT_DIR -nd NEW_D1 -nepochs 150

Required args :

• f : The hdf5 file containing all the source datasets

Optionnal args :

• ap : Directory containing the adducts_encoder.json file. "d" will make the model train with the default encoder. If argument is not used a new encoder will be created (default = None)
• sp : Directory containing the smiles_encoder.json file. "d" will make the model train with the default encoder. If argument is not used, a new encoder will be created (default = None)
• mtrain : Use the MetCCS_pos and MetCCS_neg datasets as training data (default = false)
• mtestA : Use the Agilent_pos and Agilent_neg test datasets from MetCCS as training data (default = false)
• mtestW : Use the Waters_pos and Waters_neg test datasets from MetCCS as training data (default = false)
• pnnl : Use the PNNL dataset as training data (default = false)
• cbm : Use the CBM2018 dataset as training data (default = false)
• mclean : Use the McLean Lab dataset as training data (default = false)
• nd : New datasets to create the model. If multiple files, as a list seperated by "," (default = None)
• test: Proportion of each dataset that must be kept in the testing set (default: 0.2)
• o : Existing directory to ouput model and mappers (default = current directory)
• nepochs : Number of epochs to use for the model’s training (default = 150)

At least one dataset is required to train a new model. Datasets selected will be splited between the training and testing set according to the test argument value except for mtrain which is always completly in the training set.

Additional notes

• The Adducts column of the input file must contain adducts as: M+H, M+Na, M-H and M-2H.
• The SMILES column accept any SMILES format but isomeric SMILES are recommended.
• The package includes a DeepCCSModel class that can be used directly in python without the command line tool.

References

DeepCCS relies heavily on datasets that were previously published by others:

• Zhou Z, Shen X, Tu J, Zhu ZJ. Large-Scale Prediction of Collision Cross-Section Values for Metabolites in Ion Mobility-Mass Spectrometry. Anal Chem. 2016 Nov 15;88(22):11084-11091. Epub 2016 Nov 1. PubMed PMID: 27768289.
• Zheng X, Aly NA, Zhou Y, Dupuis KT, Bilbao A, Paurus VL, Orton DJ, Wilson R, Payne SH, Smith RD, Baker ES. A structural examination and collision cross section database for over 500 metabolites and xenobiotics using drift tube ion mobility spectrometry. Chem Sci. 2017 Nov 1;8(11):7724-7736. doi: 10.1039/c7sc03464d. Epub 2017 Sep 28. PubMed PMID: 29568436; PubMed Central PMCID: PMC5853271.
• May JC, Goodwin CR, Lareau NM, Leaptrot KL, Morris CB, Kurulugama RT, Mordehai A, Klein C, Barry W, Darland E, Overney G, Imatani K, Stafford GC, Fjeldsted JC, McLean JA. Conformational ordering of biomolecules in the gas phase: nitrogen collision cross sections measured on a prototype high resolution drift tube ion mobility-mass spectrometer. Anal Chem. 2014 Feb 18;86(4):2107-16. doi: 10.1021/ac4038448. Epub 2014 Feb 4. PubMed PMID: 24446877; PubMed Central PMCID: PMC3931330.
• Mollerup CB, Mardal M, Dalsgaard PW, Linnet K, Barron LP. Prediction of collision cross section and retention time for broad scope screening in gradient reversed-phase liquid chromatography-ion mobility-high resolution accurate mass spectrometry. J Chromatogr A. 2018 Mar 23;1542:82-88. doi: 10.1016/j.chroma.2018.02.025. Epub 2018 Feb 15. PubMed PMID: 29472071.