This scripts tries to predict the bioactivity of 131 small molecule compounds (taken from ChEMBL) related to Aspartate Racemase enzyme with the aid of MACCS166 Chemical Fingerprints (from Rcpi package), the Decision Tree algorithm (from rpart package), and Support Vector Machines (from e1071 package).
Make sure to have installed the following Bioconductor/CRAN/GitHub packages Rcpi, rpart, dplyr, e1071, stringr in your Rstudio IDE.
The result of a simple search in ChEMBL (about Aspartate Racemase, small molecules filtered) were downloaded as SDF and TSV files. The bioactivities were extracted from the TSV file with the following criteria : Bioactivity < 10 ==> High (H) , Bioactivity >= 10 ==> Low (L) and later stored as Bio2.csv file. The SDF file was passed to galaxy server and via open babel, the corresponding smi files were retrieved and saved as Smile_Files.smi.
Note: As one can observe, since the dataset is relatively small for such predictions, the change of seed or # of train data would significantly change the accuracy.