We have implemented a kernel partial least squares (KPLS) algorithm that operates on molecular fingerprints. We show that this method is capable of building robust models for understanding and predicting SAR. Furthermore, we present a method to assess the atomic contributions of the KPLS coefficients, thereby facilitating the visual representation of SAR results on 2D molecular representations. Finally, we develop a scoring function that takes account of the overall prediction, atomic contributions, and confidence in the predictions. We show on an independent test set that this scoring function is capable of predicting compounds that improve affinity with a high level of success.

Kernel partial least squares (KPLS) applications to understanding SAR and improving confidence in predictions

244th National Meeting (2012)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.