Of the sixty enhancements listed in the Product Release Document for Discovery Studio 2020, users of the 3D QSAR Pharmacophore Generation protocol particularly appreciate DSC-29404 and DSC-29792.
Traditional pharmacophore models just classify compounds into those that are predicted active and those that are not. That protocol extends the approach and uses a set of molecules with known activity values to derive equations that allow a more quantitative prediction of bioactivity, much like in classical QSAR.
To guard against the risk of spurious correlations and to get an unbiased evaluation of the performance of such models, it is always recommended to validate them against an independent second set of compounds not used in training. Thanks to DSC-29404 and DSC-29792, this task has now become even easier.
Previously, the protocol reported the statistics for the validation set and a plot comparing predicted against actual activities, but did not provide the data these were based on. In the new release, the individual results from mapping each compound of the validation set onto the generated hypotheses are retained and available for detailed inspection. Also, the respective presentations of the data for the training and the test set have been harmonized to allow a direct comparison.
This makes it much easier to identify mis-predicted molecules, dig into the underlying reasons and improve the final model.
These two enhancements had been suggested by customers in India and Korea, respectively.
If you, too, have an idea how we could improve Discovery Studio and your experience using it, please don't hesitate to tell us.