I've been working on generation a QSAR model for a drug that binds in a promiscuous lipid transporter. Binding energies using MM models are generally unreliable and cannot produce binding energies that are in an order correspond with the IC50 values I have derived experimentally . I was wondering if anyones tried using ONIOM QM-MM or QM models for improve performance in screening drugs where van der walls a forces play an important role. How robust these methods are and if there is a reference to a paper you could recommend
As always, there are trade-offs between quantity and quality. QM might give you good results when there's only one binding pose at play and you already know it, but typically the description of vdW isn't that good either. MM-PBSA will probably give qualitatively good relative values, but no more than that. Free energy MD methods (alchemical transformations, umbrella sampling) will in many cases yield the best estimates due to improved sampling of the conformational space (probably the only approach that can consistently reproduce values like IC50 or at least the binding constant), but will also require extensive work and lots of computational resources, so they might be the method of choice for more extensive validation of selected compounds.
In the paper linked below you can see that this approach you suggest (AutoDock + 3-layer ONIOM) might predict the same interaction energy for compounds with IC50 three orders of magnitude apart, but apparently works well for strongly binding compounds (that is, where it should work well).
http://pubs.acs.org/doi/abs/10.1021/ci8004034
The Linear Interaction Energy (LIE) method.
Hansson, Marelius, Aqvist (1998) J Comput Aided Mol Des 12: 27-35.
https://www.ncbi.nlm.nih.gov/pubmed/9570087
---
Simple & robust.
Hello Martin, can you attach the manuscript. I cannot seem to access this publication through the ETH Proxy
- Badges
- Science topic
- Similar topics
- Computer Science
- Program