About
--------------------
Author:  Martin Gütlein
Contact: guetlein@informatik.uni-freiburg.de
Date:    17.7.2014
Content: This file is additional material to the publication "CheS-Mapper 2.0 for Visual Validation of (Q)SAR models" submitted to the Journal of Cheminformatics. It contains a description of the data used for the use-cases from the article. I do not own the provided datasets, please follow the provided links.

Usage in CheS-Mapper
--------------------
This folder contains two file types. The data files (.csv and .sdf) include the actual compound data and properties. The workflow files (.ches) can be imported with CheS-Mapper and include the wizard settings for chemical-space mapping ("Import"-Button at the top left of the wizard).

Datasets
--------------------
File:               hamster_knn.csv
Figures in article: 1
Source:             http://www.epa.gov/ncct/dsstox/sdf_cpdbas.html
Processing:         All compounds without hamster endpoint information have been removed; all fields apart form smiles, id and activity have been removed; pc descriptors with open-babel have been computed; prediction has been done with 3-nearest neighbor (loo-cv); field 'weka-prediction-prop' is the probability form the classifier (in this case: the ratio of active neighbors, e.g., 0.66 corresponds to 2 active and 1 inactive neighbors and yields a active prediction).

File:               caco2.sdf                       
Figures in article: 2,3,4,7,9                                       
Source:             http://pubs.acs.org/doi/suppl/10.1021/ci049884m
Processing:         Training and test-data have been joined.

File:               caco2_predicted.csv             
Figures in article: 10,11          
Processing:         This is the CheS-Mapper input file produced by the KNIME workflow (see Additional File 3); LOO-CV: 'SimpleLinear' and 'SMOreg' are the cross-validated predictions of simple-linear-regression and support-vector-regression, 'diff-caco2-simple' and 'diff-caco2-smo' are the errors of the corresponding predictors, 'error-diff' is the difference between both errors. Test-Set-Validation: 'dataset' specifies if this compound was part of the original training or test-dataset, 'smo-test' is the prediction of the support vector machine trained on the training data and 'diff-caco2-smo-test' is the corresponding error.

File:               cox2.sdf                        
Figures in article: 5,6,12,13                                       
Source:             http://pubs.acs.org/doi/suppl/10.1021/ci034143r
Processing:	    none

File:               epafhm_ad.sdf                   
Figures in article: 14,15,16                                        
Source:             http://www.epa.gov/ncct/dsstox/sdf_epafhm.html
Processing:	    We have computed PC-descriptors with OpenBabel (applicability domain computation is based on only 5 PC-descriptors, see article), we created 3D structures with OpenBabel, removed 37 (of 617) compounds that have no measured value for LC50_mmol, removed a single vast outlier compound (strychnine hemisulphate salt) to make the visualization more compact, and computed the LOG value for the endpoint LC50mm_mol; the result of the four applicability domain methods (knn,centroid,leverage,prop) is stored with 2 features each: a binary "<ad>-inside" with values true and false, and a probability value: "<ad>-probability".

Files:              hamster_pc.csv and hamster_pcfp.csv
Figures:            Additional file 2, Figures 1 and 2
Source:             http://www.epa.gov/ncct/dsstox/sdf_cpdbas.html
Processing:         All compounds without hamster endpoint information have been removed; removed all fields apart form smiles, id and activity; descriptors have been computed as described in Additional File 2; 5-fold cross-validation results are stored in 'activityoutcome-cpdbas-hamster_classified' (ratio that describes how often this compound was classified as active) and 'activityoutcome-cpdbas-hamster_missclassified' (ratio of how often this compound was missclassified).
