Overview

Scripts used to run the experiments presented in the paper:

"Liquid-Chromatography Retention Order Prediction for Metabolite Identification",

Eric Bach, Sandor Szedmak, Celine Brouard, Sebastian Böcker and Juho Rousu, 2018

Summary of the results shown in the paper (File needs to be downloaded and opened with a web-browser.).

Installation

There is no further installation required. The scripts run out of the box, if all the package dependencies are sattisfied. All the source code in this repository is under the MIT License.

Order prediction and evaluation code

The order predictor, e.g. RankSVM, and evaluation scripts are implemented in Python. The code has been tested with Python 3.5 and 3.6. The following packages are required:

• scipy >= 0.19.1
• json >= 2.0.9
• numpy >= 1.13.1
• joblib >= 0.11
• pandas >= 0.20.3
• sklearn >= 0.19.0
• networkx >= 2.0
• matplotlib >= 2.1 (optional)

Data pre-processing and evaluation report creation

The data pre-processing scripts as well as the script to reproduce the results shown in the paper are written in R. For the development R version 3.4 was used. The following packages are required:

• Reproduction of results: ECCB2018.Rmd
  • data.table
  • ggplot2
  • knitr
• Reproduction of data pre-processing:
  • Matrix
  • obabel2R
  • rcdk (used for fingerprint calculation)
  • fingerprint

Furthermore, the OpenBabel (>= 2.3.2) command line tool obabel must be installed only if the data pre-processing needs to be repeated.

Calculation of MACCS counting fingerprints

The rcdkTools package allows the computation of several counting fingerprints through the Chemical Development Kit (CDK).

Usage

All experiments of the paper can be reproduced by using the evaluation_scenarios_main.py script with the proper parameters:

usage: evaluation_scenarios_main.py <ESTIMATOR> <SCENARIO> <SYSSET> <TSYSIDX> <PATH/TO/CONFIG.JSON> <NJOBS> <DEBUG>
  ESTIMATOR:           {'ranksvm', 'svr'}, which order predictor to use.
  SCENARIO:            {'baseline', 'baseline_single', 'baseline_single_perc', 'all_on_one', 'all_on_one_perc', 'met_ident_perf_GS_BS'}, which experiment to run.
  SYSSET:              {10, imp, 10_imp}, which set of systems to train on.
  TSYSIDX:             {-1, 0, ..., |sysset| - 1}, which target system to use for evaluation.
  PATH/TO/CONFIG.JSON: configuration file, e.g. PredRet/v2/config.json
  NJOBS:               How many jobs should run in parallel for hyper-parameter estimation?
  DEBUG:               {True, False}, should we run a smoke test.

Example: Reproducing results shown in Table 3:

The following function calls are need:

MACCS counting fingerprints:

python src/evaluation_scenarios_main.py ranksvm baseline_single 10 -1 results/raw/PredRet/v2/config.json 2 False

• baseline_single: Single system used for training and testing.
• 10: Use "Eawag_XBridgeC18", "FEM_long", "RIKEN", "UFZ_Phenomenex", "LIFE_old" for training and testing.
• -1: By setting TSYSIDX to -1, we run all target systems in a single job. This parameter can be used for parallelization.
• results/raw/PredRet/v2/config.json: Configuration of the experiment, e.g. molecular features and kernels.
• 2: Number of jobs/cpus used for the hyper-parameter search.
• False: Not running in debug-mode. Results will be stored in the final directory.

The results will be stored into:

results/PredRet/v2
                └── final
                    └── ranksvm_slacktype=on_pairs
                        └── allow_overlap=True_d_lower=0_d_upper=16_ireverse=False_type=order_graph
                            └── difference
                                └── maccsCount_f2dcf0b3
                                    └── minmax
                                        └── baseline_single

MACCS binary fingerprints:

Modify the results/raw/PredRet/v2/config.json configuration file:

"molecule_representation": {
  "kernel": "minmax",
  "predictor": ["maccsCount_f2dcf0b3"],
  "feature_scaler": "noscaling",
  "poly_feature_exp": false
}

becomes

"molecule_representation": {
  "kernel": "tanimoto",
  "predictor": ["maccs"],
  "feature_scaler": "noscaling",
  "poly_feature_exp": false
}

Then run:

python src/evaluation_scenarios_main.py ranksvm baseline_single 10 -1 results/raw/PredRet/v2/config.json 2 False

The results will be stored into:

results/PredRet/v2
                └── final
                    └── ranksvm_slacktype=on_pairs
                        └── allow_overlap=True_d_lower=0_d_upper=16_ireverse=False_type=order_graph
                            └── difference
                                └── maccs
                                    └── tanimoto
                                        └── baseline_single

How the results can be loaded and visualized is described here.

Citation

To refer the original publication please use:

@article{doi:10.1093/bioinformatics/bty590,
    author  = {Bach, Eric and Szedmak, Sandor and Brouard, Céline and Böcker, Sebastian and Rousu, Juho},
    title   = {Liquid-chromatography retention order prediction for metabolite identification},
    journal = {Bioinformatics},
    volume  = {34},
    number  = {17},
    pages   = {i875-i883},
    year    = {2018},
    doi     = {10.1093/bioinformatics/bty590},
    URL     = {http://dx.doi.org/10.1093/bioinformatics/bty590},
    eprint  = {/oup/backfile/content_public/journal/bioinformatics/34/17/10.1093_bioinformatics_bty590/2/bty590.pdf}
}