BindingDB provides virtual screening tools to help identify the compounds in your own compound catalog that are most likely to be active against a desired target.

Three screening methods are provided:

  • Maximum Similaritya
  • Binary Kernel Discrimination (BKD)b
  • Support Vector Machine (SVM)c

These methods need to be trained with a set of compounds known to be active. Two of the methods, BKD and SVM, also require a set of compounds presumed to be inactive (decoys).

1. The actives can be up to 100 compounds from your BindingDB search
To create a set of actives from a BindingDB search, execute the search as usual, and then use the "Make Data Set" buttons at the top of the Query result table. You will then be provided with options for virtual screening.

2. The actives can be up to 100 compounds that you upload from your computer.
To use your own set of actives, use this file browser to upload an SDfile from your computer:

  1. Maximum similarity: Your compounds are ranked according to the maximum Tanimoto similarity of each compound to any of the actives, based upon JChem fingerprints
  2. Binary Kernel Discrimation (BKD): Divide up to the first 100 of these compounds ("actives") into a reference set and a test set of equal number, supplement each set with 500 other drug-like compounds presumed to be inactive, and Jchem binary fingerprints will be computed for each compound. Rank test-set compounds using BKD comparison to the reference set and report enrichment of actives at the top of the ranked list. Then combine reference and test sets into one large reference set and use it to rank your uploaded compounds. (Harper, G.; Bradshaw, J.; Gittins, J. C.; Green, D. V.; Leach, A. R. Prediction of biological activity for high-throughput screening using binary kernel discrimination. J. Chem. Inf. Comput. Sci. 2001, 41, 1295-1300)
  3. Support Vector Machine (SVM): Divide up to the first 100 of these compounds ("actives") into training and test sets of equal number, supplement each set with with 500 other drug-like compounds presumed to be inactive, and compute numerical descriptors for each compound. Train an SVM model with the training set, apply it to the test set, and report enrichment of actives at the top of the ranked list. Then compute descriptors for your uploaded compounds and apply the SVM model to rank them. (Jorissen, R.N. and Gilson, M.K. Virtual screening of molecular databases using a Support Vector Machine. J. Chem. Inf. Mod. 2005, 45, 549-561)