I have an ongoing gene expression study and have had difficulty finding an appropriate way to analyze the data. For context: I want to observe the relative fold change in expression over time for several genes (normalized with a housekeeping gene) in my samples. Now I understand the 2^-(ddct) method is commonly used to measure the difference in expression between one treatment and a control which can be analyzed using a T test. I have calculated the relative fold change at each time point relative to the same control. I know I can perform a T test making individual comparisons between each treatment and the control but my main objective is to see if there are statistical differences in the expression changes over time. Given that my only treatment is time, can you make statistical comparisons between the relative fold change differences? From a design standpoint, I am using the same biological replicates at each time point so theoretically I feel there should be a simple way to make the comparisons. However, I am even struggling to find references to literature with a similar design where time is the treatment.
I'm assuming that since no treatment is applied, you already expect the expression to change over time. The way to statistically test this change depends on the pattern of expression changes you are observing. If expression is consistently creeping up or down over the time course, you can use a simple linear model with the gene's expression as the outcome variable and time as the explanatory variable. If you run this regression analysis in R, you can look at the summary for the linear model to get the slope (i.e., effect size) and P-value for the time term. If the change is all in the same direction but not linear (e.g., logarithmic growth), you can still use a generalized linear model but you'll need to be more careful about how run it (e.g., should you transform the data, or use a certain link function, etc.). However, if the expression change pattern is more complicated (i.e., clearly non-linear), you need to approach it differently. Maybe consider something like a random forest regression model and test its accuracy on a separate dataset.
Jason W Hoskins Thank you for the insight! You perfectly described the challenge I was facing. The data has upregulation and downregulation at different time points making it abnormally distributed. Using a log10 transformation cleared up the issue. Thank you again.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude