Dear all, I want to predict the IC50 values of my new synthesized compounds verses the MCF-7 cell line (anticancer activity).In fact; I tried to find a good QSAR model using the MOE (Molecular Operating Environment) program. Unfortunately, I have not succeeded in that, because I did not find any experimental data gave the IC50 of compounds similar to mine. Therefore, I want to know if I can found this model without basing on previous data of other compounds, and if I found some experimental dada concerning the IC50 in the literature ,is the number of compounds used affected the quality of my model ?
You cannot build a QSAR model without experimental data. The number of compounds used for building the model has influence on the robustness of the model. The more the number of compounds, the more robust will be the model.
For the development of QSAR model, you need to have well defined experimental data. It is good to have maximum number of diverse compounds for the development of QSAR model.
Ideally, you would need data:
-from compounds that are comparable to yours
-in an assay that is comparable to yours (or, if not possible, in a consistent assay format).
The number of compounds is important because the chemical space covered by the compounds more or less defines the chemical space wherein your model is applicable.
However, if you have some very dissimilar compounds working in very different ways, assembling data collected from different compound families can be a problem. Here, your readout seems to be related to viability / proliferation of a cell line. Chemical compounds can affect this in a variety of ways. In the absence of any evidence that your compounds and the compounds of your training set work through the same mechanism, building a QSAR model can be a tricky thing, anyway!
If you have tested enough compounds one approach would be to take half of your compounds to build a model and predict the other half. Either you could do a random split or select a number of compounds representing different structural features or mechanisms of action.
This approach is advantages since:
- all data comes from the same lab and the same assay
- compounds are (normally) related since they come from a certain series
- if you keep testing you could validate you predictions and your model to increase confidence in the model
The remaining question would be the number of compounds you tested.
There seem to be a possibility that you could, if you know the target of your compounds. i.e. where it binds to the receptor surfaces. A simple similarity search can be used to find the target followed by receptor site docking. A QSAR by docking strategy can potentially be employed where one can correlate MOE descriptors with docking scores of the new analogs you synthesized, which in turn can be used to predict the IC50 of similar analogs.
What Ben suggests could be feasible IF the target of the compounds is formally identified....
However
I do not like working that way.
Indeed, by doing so, you train your model against your scoring function, and not against real activities. In the absence of any evidence that your scoring function reasonably correlates with biological activity, this is a very risky move. And, if there was good evidence that docking score correlates well with biological activity, why go for a QSAR model? (The only reason I'd see would be as an intermediate filter during virtual screening, as descriptor calculation would be less demanding than docking... but this is another discussion).
Hi, it is not possible build a QSAR model without experimental data. Moreover, the activity data should be related to the ligand-receptor interactions. If you use the ic50 against a cell line it is hard to find a correlation between the predicted activity and the experimental. I have use the mcf7 data to derive some QSAR and pharmacophore modeling but the model is quantitative
http://pubs.acs.org/doi/abs/10.1021/ci900254b
In the last paper in which I have worked with MCF7 cells there is a discussion on the activity data to use. maybe could be interesting for you. you can find it at this link:
http://www.sciencedirect.com/science/article/pii/S0223523413004169
Sincerely
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Medical Pharmacology