Hello everyone. I work with drug development, and routinely conduct IC50 measurements. Usually I report IC50 as sample mean +- sample standard deviation. Now I have been looking on the correct way to report this data as a Confidence Interval (95%). Since my sample sizes are very small (mostly 2), I have noticed that my confidence intervals are far from great. Do anyone has some observation on this matter?
Besides, usually I extract the IC50 value from individual experiments and calculate mean and standard deviation from these single numbers. However, I have seen works where the points in the concentration-response curve have error bars, so IC50 statistics will come from a single fitting of points with error considerations. What would be the correct way to do this?
Thanks in advance
If you perform a regression analysis, such as the nonlinear regression used to calculate the IC50 from a concentration response, the software will also calculate the standard error of the parameter estimate. You can report this standard error as an estimate of the precision of the single IC50 measurement, if that is the only IC50 measurement you have.
If you make multiple IC50 measurements (preferably 3, but you can do the calculation with 2), you can calculate the average and standard deviation. This assumes that the measurements follow a normal distribution. The 95% confidence interval of a normal distribution is the average +/- 1.96 times the standard deviation.
Adam B Shapiro Thanks for your quick response. I have read that for small sample sizes (less than 30), the correct procedure would be to use the t-distribution (instead of the normal distribution), so my factor would be different from 1.96 ("worse" as my data is smaller). Aside from this factor consideration, I have been doing this analysis as you said, that is, taking individual IC50 values to perform the statistics, instead of combining the concentration-response curves.
A useful guide for calculating confidence intervals:
https://www.statisticshowto.com/probability-and-statistics/confidence-interval/#CISample
My observation is that in reporting measurements such as IC50s in publications from drug discovery projects, wherein there is usually only a single measurement made for most compounds in a structure-activity relationship study, the authors often just give the single measured value with no error estimate. When a small number of well-studied compounds are being compared, it is usual to report the average and standard deviation of 3 measurements. I rarely see confidence intervals reported. Of course, as long as the number of measurements is given, the reader can calculate the confidence intervals if he or she wishes to do so.
The confidence interval for a small number of measurements is quite wide, as you pointed out, which is probably why we don't tend to show it.
That sums up what I was thinking. Now I'll be more reassured about what I'm doing. Thanks for the discussion, Adam!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression