My pharmacophore modeling experiment in Accelrys Discovery Studio constitute 36 actives and 43 inactives as validation set towards one pharmacophore hypothesis (2 HBA, 1HBD and 1 hydrophobe). I got predictions in 2 models:
Model 1: True Positives=34 True Negatives=11 False Positives=32 False negatives=2 Sensitivity=0.94 Specificity=0.25
Model 2: True Positives=36 True Negatives=2 False Positives=41 False negatives=0 Sensitivity=1.0 Specificity=0.04
Which one Pharmacophore model you will recommend to select me for further analysis? Specificity is more important than sensitivity or vice versa. Please give valuable suggestions.
May be you can use model 1 but again this is also not a good model. Since ratio of true positive and false positive is lower. You can use Güner–Henry scoring equation to measure the goodness of fit.
Your test study is too far from a real-world setting. In a typical screening set, the ratio of actives to inactives is about 1 to 1,000. Hence, with both models, you will have way too many false positives.
both of them are not preddy good, yet out of them 2 model seems to be comparable,as having high rate of positives.
u can refer the article
http://jbiomedsci.biomedcentral.com/articles/10.1186/1423-0127-18-8
This case study is true when we employ structure-based pharmacophore model to study enrichment. The single-ligand:receptor interaction details cannot be seen over molecules that have been synthesized (for which a docking study may have been accompanied in publication). This observation stems from the fact that molecules may have diversity in the pursuit of optimization, second, a representative crystallographic ligand may not have all the features shared with the synthesized molecules (unless congeneric series). Third, if we attempt to identify features from a single ligand and the match-able validation set, we are compromising with the receptor-ligand interactions of the reference structure only to enhance goodness of hits? I agree with Dmitri Kireev that this case study doesn't fit into the real-world setting, but given a condition such as the availability of one ligand-receptor complex, the synthesized molecules as validation set (79 molecules) and the need to prioritize selected features from its hypothesis encoded by the crystal ligand so as to recognize potent molecules from VS, a validation is still necessary as we don't have exact binding conformations of the synthesized 'validation' set? Please limit your answers keeping these keywords in mind: structure-based modeling of one protein-ligand system, validation of synthesized molecules,
I am thankful to Ashish and Mohd. Athar for pointing out the Güner-Henry scoring method for evaluation of VS protocol. The paper suggested that they have used 66 active molecules to make a database of decoys comprising 3606. It is acceptable that what will be the condition if you append Inactives of acetylcholinesterase inhibitors in place, whether the VS protocol may have not recognized inactives as hits since it is chemically similar to actives. Give your take.
- Badges
- Science topic
- Similar topics
- Philosophy Of Science
- Models