Pharmacophore modeling

Data from pharmacophore modeling based on biochemical assays provide an important tool for preselection of the synthesizable molecules in combination with 3D data and especially, if co-crystallization data are not available. If the 3D structure of a target kinase is available, structure-guided drug discovery software tools might be used for the identification of potential ligands for a given target.

To enable pharmacophore modeling, the selected kinase targets shall be tested against the NCL and the CVL. The resulting hits will provide sufficient input for a successful virtual screening and rational design of new analogues based on a Prediction Oriented QSPAR model approach using Vichem’s in house developed pharmacophore modeling software. Thousands of molecular descriptors are calculated for hit compound structures and fed into the QSPAR program system and models are optimized by selecting predictive descriptors. The optimization cycle consists of four iterative steps:

1. training set/evaluation set splitting
2. descriptor selection based on sequential or genetic algorithms
3. applying chosen statistical method to achieve the best fitting between the training set structures represented by the selected molecular descriptors and the biological data
4. internal validation of fitted model with evaluation set

Enhanced ANN software (Artificial Neural Network) and AdaBoost methods can be used to establish relationship between activity and descriptors generating a pharmacophore model.

We have developed more than one hundred Prediction Oriented QSPAR and pharmacophore models for kinases and ADMET parameters.