This tutorial describes in detail how to implement the analytical methods described in "A statistical method for comparing viral growth curves" by Wang and Bushman. It would be helpful to read this paper before going through the tutorial. In addition, it would be useful to read through some of the introductory material in GraphPad Prism.
Before beginning the experiment, it is helpful to consider how to maximize the sensitivity of the final statistical tests. The recommended method involves generating a summary statistic for each culture--the growth rate--and then comparing the collection of growth rates among drug concentrations, mutants, etc. The statistical power is governed by the number of growth rate measurements in the comparison. For this reason, maximizing the sensitivity involves carrying out the maximum reasonable number of replicates. If replicates are expensive, it is wise to carry out the minimal number of measurements per independent culture, and analyze the maximum feasible number of cultures. It may be helpful to carry out a pilot study to find the range of exponential growth, then analyze only this part of the curve in a more extensive series of replicates. In the analysis below we have sought to keep the methods as simple as possible. The data is first transformed in Excel, which will be familiar to most virologists. We then analyze the data using GraphPad Prism (www.graphpad.com). This statistical package has the advantage of resembling Excel in its format for data tables and general character. Other statistical packages could of course be used, but many of these have much steeper learning curves.
Data was tabulated in Excel as shown. Data points below the level of detection were given a value of 1. Note that designations such as ND for "not detected" will cause problems in the downstream analysis. Similarly, zeros are a problem since the logarithm is undefined.
Data was then smoothed by multiplying data from days 5-8 by 4, days 9-12 by 16, days 13-16 by 64, and days 17-20 by 256, to account for splitting of cells during the culture period. This yielded the data matrix shown.
Data for the tests of integrase inhibitors described in this study can be found here: Sample data
After these transformations, the data was taken into GraphPad Prism. First, the nature of the data table needs to be specified as shown. Data from each culture dish is treated as a separate column in the data table. Choose New Project under the File Menu, and configure "Y columns" with "A single column of values". Press OK.
The exponential phase is then selected for study. The times that we plan to exclude are removed using the Exclude Values function in the Edit menu. In this example, we block out days 1-5 and 15-20 and analyze only days 6-14. Note that this choice was made by inspecting graphs of the time versus log (p24) values, averaged over each treatment, as shown at left. These graphs can be generated in either Excel or GraphPad Prism.
We now need to construct a second data table containing the growth rates. For the statistical analysis, we need to make the four rates for each replicate a separate column. Thus we need four columns, one for each condition, each containing four values. Choose New Project under the File Menu, and configure "Y columns" with "A single column of values". Press OK.
To fill in the new data table, copy the first four SLOPE values, which correspond to the no-inhibitor control. Then bring up the new table, and click on the box in the upper left of the the Y-value field (cell A1). We now want to paste the row of four SLOPE values into a single column with four entries (i. e. transpose the values). Go to the Edit menu and select Paste Special. In the Format Data Object dialog box, select Placement, and in the new dialog box select "Transpose". Then press the "Paste" button in the bottom right corner. This places the four rates in a single column. Enter the rates for the three other conditions in the next three columns.
We now must set up the control for multiple comparisons. Under Post Test, select Dunn's: Compare selected pairs of columns. Press Select.
Use the tool at the top to specify comparison of A to B, A to C, and A to D. Press OK.
This yields our final answer. The overall P value (line 4) is significant at P=0.0041, so we have a basis for asking where the differences are present in the data. Under Dunn's Multiple Comparison Test, we can see that the no-inhibitor value is significantly different from the 20 uM value, but the other two concentrations (100 nM and 2 uM) were not significantly different. If we wished to try to establish that there was a significant difference between the no-inhibitor values and the 2 uM values, we would need to carry out another experiment with more independent rate measurements to increase the statistical power.