# README instructions for chemical-biological read across (CBRA)
#
# Also in supplemental material of publication:
#    Integrative Chemical–Biological Read-Across Approach for Chemical Hazard Classification. Y
#    Yen Low, Alexander Sedykh, Denis Fourches, Alexander Golbraikh, Maurice Whelan, Ivan Rusyn, and Alexander Tropsha. Chemical Research in Toxicology 2013 26 (8), 1199-1208.
#    DOI: 10.1021/tx400110   http://pubs.acs.org/doi/suppl/10.1021/tx400110f 
#
# Yen Low ([email protected])
# 05 June 2013 version 1.0
##########################################################

################ OBJECTIVES OF PROGRAM ###################
1. Builds 4 models for comparison
   a) dual-space kNN of chemical and biological neighbors
   b) single space kNN of chemical neighbors
   c) single space kNN of biological neighbors
   d) single space kNN of hybrid neighbors (i.e. chemical+biological spaces)

2. Generates radial plots shown in desired_output.pdf
In each radial plot, 
- nodes represent compounds 
- central node represents the target compound to be predicted
- nearest kbio biological neighbors are positioned left of vertical axis 
- nearest kchem chemical neighbors  are positioned right of vertical axis 
- edge length is proportional to the Jaccard distance between target compound and its neighbor
- nearest neighbors are positioned closest to the 12-o'clock position
- colors denote the observed class of the compound (black=nontoxic,-1; red=toxic,+1)


################ DATA FILES PROVIDED ###################
Data set: Rat acute toxicity (Oral LD50)
   a) ld50_drg_n.xa (chemical descriptors)
   b) ld50_atp_csp_n.xa (biological descriptors)


At least R 2.14 is recommended.
Tested on R 2.14, 3.0, Windows 7-8 and Ubuntu 12.10

The following R packages are required:
(script will automatically install them if necessary)
1. boot
2. caret
3. class
4. e1071
5. plotrix
6. ROCR
7. vegan


###### INSTRUCTIONS #############################
STEP 1: Unzip .zip package. Check that the following files are in the same folder:
1.  ld50_drg_n.xa      (chemical descriptors)
2.  ld50_atp_csp_n.xa  (biological descriptors)
3.  master_script.R
4.  multispaceNNobj.R
5.  multispace_functions_AD_scaled.R
6.  readXAfile.R
7.  sampling.R
8.  validationstats_BIN.R
9.  bootstrapSD_BIN.R
10. desired_output.pdf
11. README.txt


STEP 2: Run master_script.R using the following command
If running R in terminal MODE (e.g. in linux): 
	At the command prompt, enter: Rscript master_script.R
OR
If running R GUI (e.g. in Windows): 
	Step 2.1: Start R GUI.
	Step 2.2: Set working directory to where CBRA is unzipped to
	  	  Within R GUI, go to File -> Change dir... and enter the file path of the CBRA folder
	          OR enter: setwd("[file path of CBRA]")  - Use foward slash "/" instead of backslash "\"
	Step 2.2: Open master_script.R by File -> Open
	Step 2.3: Run master_script.R. 
		  Within R Editor, go to Edit -> Run all

PROCESS TAKES 3 MINUTES ON A QUAD-CORE COMPUTER. PLEASE BE PATIENT.
(Each model consists of multiple models generated by 5-fold external cross-validation and 10-fold internal cross-validation)


###################################################
Output files generated:
1. .pdf (figures showing dual-space kNN)
   a) singlecpd.pdf      (Compound #20's nearest biological and chemical neighbors)
   a) 6cpds_2by3grid.pdf (6 compounds' nearest biological and chemical neighbors in 2 by 3 grid)
   a) 4cpds_2by2grid.pdf (4 compounds' nearest biological and chemical neighbors in 2 by 2 grid)
2. .pred files (tables containing observed and predicted values of each compound)
   a) chem.pred
   b) gene.pred
   c) hybrid.pred
   d) dual.pred
3. validationstats_xxx.txt files (prediction performance of models, e.g. specificity, sensitivity, AUC)
   a) validationstats_chem.txt
   b) validationstats_gene.txt
   c) validationstats_hybrid.txt
   d) validationstats_dual.txt   
4. dualspace.RData (.RData object containing data, models, objects for radial plots)
5. shuffleID.RData (.RData object containing randomizer seed used for cross validation)


######## NOTES #########################
If dependencies cannot be loaded, manually install the following R packages using:
install.packages("[Rpackage]")
For example, to install the R package, boot, enter: install.packages("boot")
Required R packages:
1. boot
2. caret
3. class
4. e1071
5. plotrix
6. ROCR
7. vegan


Please contact Yen Low at [email protected] to report bugs.