naex2-LigandScoringProtocol.txt

Name

lly_ec50_svm

Software

KNIME 2.12, [Lilly Internal QSAR scripts: svmfp_calibrate, svmfp_evaluate.v2, iwprismm]

Parameters

iwprismm -abr
iwprismm -ap
iwprismm -cmi
iwprismm -hpo
iwprismm -jwmc
iwprismm -w
iwprismm -jurs
iwprismm -morse
iwprismm -shd
SVM calibrate -TRpct 80
SVM calibrate -niter 20
SVM calibrate -arbor
SVM calibrate -cubist
SVM make -CATSP13-MAP7:HRY1-RS-CLOGP

Method

This method used QSAR methods predict the binding affinity of the test set compounds.
Activity data for compounds tested against FXR were obtained from ChEMBL (target ID: CHEMBL2047)
Inactive compounds were filtered and only compounds with EC50 measurements were retained (N=650).
A large set of molecular descriptors were generated using the [Lilly internal] iwprismm script.
The [Lilly internal] SVM calibrate script was used to test a large number of sets of fingerprints
and kernel methods. The CATSP13-MAP7:HRY1-RS-CLOGP fingerprint was chosen as it has one of the best average
cross validation accuracies at predicting pEC50s of the training set.
The entire EC50 training set was used to build a model with the chosen fingerprint and the test
set compounds were scored using it.