Have you tried to fit the Fa-CI plot after removing the last data point?

It is possible that the extreme data points at Fa close to 1 cause a skewing of the fitting, since the software takes into account each data point entered equally importantly. Extremely low or extremely high effects that are beyond the accuracy of assay should not be entered.

Luca,

Thanks for the input, I thought about whether I should remove data point at the end, but in the end I didn't with the following reasons:

1) Since it is cancer drug research we are talking about, in my understanding the "meaningful" CI will be at Fa >0.9 (plotting effects, therefore indicate >90% cell death), as that's the dosage it will be biologically relevant while you use that to treat cancer. It doesn't seems logical to remove that last data point which actually achieve the needed effect.

2) Pardon my poor maths, but in my understanding the simulation line doesn't fit the data points at all (for which I mean the simulation is "sigmodal(!?)" while my data points are "parabola(!?)"),  then either my data points are wrong (which I doubt as in another independent experiment for these combination, the shape of my data points are still parabola),  or the logarithm doesn't apply as the compound does not fit the model or certain assumptions?

I am new to drug combination but I try to read a lot literature and understand different models and definition the best I can with my limited maths.

Thanks for your input.

Regards,

Nick

Conceptually, the interaction of two or more agents is described to be ‘synergistic’,

when it produces a combined effect greater than the sum of their individual effects.

However, such a mere concept is not good enough to precisely and scientifically evaluate any data of drug combination/s, especially, in the context of combination chemotherapy, which may have clinical implications.  This software was

specifically developed to analyze drug combination effect based on the median effect equation of drug dose and the response and the algorithms developed using several aspects viz., law of mass action, biochemical kinetic phenomena such as association, dissociation, co-operation of molecules that participate in dynamic reactions. It basically needs the data of dose response of the individual drugs, and the different combinations of them. When the data are analyzed using this software, it generates a plot of the combination index (CI) against the fraction affected (Fa) with reference to the each drug combination tested in the experiment. For each of the specific drug combination tested, if the generated CI is 1, it is ‘antagonistic’.

Dear Nick,

1) you're right in principle: cancer therapy should aim at >90% effect. However, here you are trying to establish how two drugs interact in vitro. Therefore, it is more accurate to determine the effects at Fa between 0.1 and 0.9, which are usually more reproducible than the extremes. You will then possibly validate the findings in vivo, using efficacious doses.

2) I am not strong in maths either, and I do not know exactly how the software fits the data. Your points indeed look like a parabola, but it is possible that the system forces the simulation curve into a sigmoid to account for the last point. That's why I suggested to remove it, and see how the curve changes.

Then again, it may also be the case that your combination is only synergistic at high doses. In such a case, if your experimental data are confirmed, I would not consider the fitting. The plot is not a dogma. There are examples of drugs that are synergic at some values and neutral or even antagonist at other concentrations. Understanding why, that's the challenge!

Good luck,

Luca

Luca,

Thank you so much for your response, I think that's all the advice I need for now. I guess in-vitro assay doesn't mean much but a suggestion for drug combination afterall.

I will keep my data points and look into the CI between Fa=0.5 - 0.9 without the simulation lines, as an indication on whether it's worth going further into in-vivo experiments.

Nonetheless, these data still allow me to choose the ratio between compounds to be used in in-vivo assays.

Once again, appreciate your input into this.

Hi Nick, I had the same issue with this software. I only take into account only 0.1

Reagan,

Thank you for your input.

Yes, the CI value is based on the IC50 (Dm) determined (however it is not as you stated that because they are in nM vs uM range) and therefore if your medium effect curve does not have a good fit (R>0.95 in my opinion), your CI value will not be reliable at all. I recommend you to read more about how CI is derived and how to read the isobologram.

And because it is a linear fit in log axis, in order for IC50 determined by the program to be reliable, it is essential to discard extreme data point from your single drug dose-response arm (any data point having response >0.9 or

Hi Nick, I do agree discarding single drug Fa0.9. I do admit that I still need to read further into the software to fully understand the calculation involved in the software itself.  I used this software for multiple anticancer drugs screening against various cell lines. I got the same graph FA-CI like yours with one or two points in the middle combination concentration becomes antagonism - like a sudden hump (CI >1, some people used 1.2) and becomes synergistic at low and highest (well it could be it became off target at higher concentration? should that drug be discarded?). I narrow down my drug screening based on drugs that are synergistic at low to middle combination concentration and number of cell lines that the combination is active at. And also normal cell lines (hoping to get a good antagonism here or as little effect as possible)

That is the thing about T-test, some paper do still use them but I would not dare to say any significant difference = synergy. I used it to help me screen my drugs (used during my MSc because I did not have the combusyn then).