Presentations from the Cresset User Group Meeting 2019
Thank you to all attendees who contributed to the success of the Cresset User Group Meeting 2019. As I'm sure ...
Summarizing activity and selectivity in one picture was presented by Dr Mark Mackey, Cresset at the Cresset European User Group Meeting 2015.
3D-QSAR based on molecular interaction potentials can provide a wealth of information about the exact molecular characteristics required for activity. However, current techniques have a number of issues such as alignment noise, sampling errors and descriptor choice which can make it difficult to reliably produce effective models. We have presented in the past techniques for solving the sampling problem and shown that using accurate electrostatics combined with simple shape descriptors often gives meaningful models. However, there are still times when it is not possible to obtain a statistically valid linear regression model.
One useful qualitative data analysis method that is being increasingly used is activity cliffs analysis. In this technique, pairs of compounds are located that are similar (in some sense), but have different activities. Traditionally activity cliff analysis has used a 2D definition of similarity, but extension to 3D similarity metrics gives additional information that is very useful to locate the source of and reason for the activity differences.
An extension of 3D activity cliff analysis is to mine the entire data set for corresponding cliffs and use this to build a model for activity. Analysis of the data set to locate activity cliffs locates the pairs of molecules with the highest information content. However, this needs to be tempered with an analysis of how likely it is that the molecules are aligned correctly, as only properly-aligned molecules contain any information. We apply Bayesian corrections to the activity cliff data to obtain a map of the electrostatic and shape characteristics that seem to locally correlate with improving activity. The resulting model is semi-quantitative in that it attempts to describe the entire data set without building a linear regression model. This technique provides a valuable fall back to the computational chemist for information extraction from ligands in 3D.