Scripts for assisting in modeling quantitative structure activity relationships from 2D chemical data
Given a training set, compute a set of smarts descriptors unique to that set. Can choose between path-based descriptors and circular descriptors.
Given a SMILES file, outputs a space delimited file of the fingerprint descriptors for each molecule for use with training a model
Given a fingerprint file train a linear model to predict a numerical quantity of interest. This quantity is assumed to be the second column of the file. Outputs model along with cross-validation statistics.
Given a model and a set of compounds, uses the model to predict the quantity of interest.
Given a fingerprint file train a classifier model to predict a the class of interest. Does NOT support multiple classes. This quantity is assumed to be the second column of the file. Outputs model along with cross-validation statistics.
Given a model and a set of compounds, uses the model to predict the class of interest.
Given a model trained using smarts descriptors and a compound, map the model onto the compound and produce an image.
data/solubility.smi contains SMILES where the molecule name is the desired numerical label (logS in this case).
$ time ./createsmartsdescriptors.py --subgraph -c 3 data/solubility.smi -o solubility_smarts.txt
real 0m48.485s
user 0m48.611s
sys 0m0.370s
$ wc solubility_smarts.txt
8947 17894 161131 solubility_smarts.txt
$ time ./outputfingerprints.py --smarts --smartsfile solubility_smarts.txt data/solubility.smi -o solubility.fp.gz
real 0m46.029s
user 0m46.247s
sys 0m0.589s
$ time ./trainlinearmodel.py --elastic solubility.fp.gz -o solubility.model
CV: R^2=0.7765, RMS=0.9836, NullRMS=2.0738 (stds 0.0081, 0.0188, 0.0070)
Gap: R^2=0.1205, RMS=-0.3087, NullRMS=0.0000
real 89m29.045s
user 1504m2.279s
sys 515m52.762s
# the following will make images named using the molecule names in one.smi
$ ./visualizesmartsmodel.py solubility.model solubility_smarts.txt one.smi