Resazurin stability readings analysis

Analyzing stability readings from Naomi’s resazurin experiment

No AI was used in this analysis or process.

Overview

Namoi previously conducted resazurin stability experiments to test whether resazurin readings stay stable once sampled from the oyster cup and left in the fridge for up to 5 weeks. This analysis looks at the data to see if readings are stable. Data are available on GitHub here with the code to analyze here.

Analysis and results

Resazurin was measured on 12 oysters with samples taken at 0, 0.5, 1, 1.5, and 2 hours during the initial incubation. These samples were re-read at each time point and again at 2, 3, 4, and 5 weeks to test stability in measurements.

Since I am only interested in stability of the readings over time, I am not doing normalization for this analysis - I am analyzing the raw values and how they change over time.

Load data

data<-read_excel("data/fluorescence_stability/20260924/stability_data.xlsx")
head(data)

plot<-data%>%
  mutate(unique=paste0(original_time, reading_time, well))%>%
  filter(reading_time>2)%>%
  
  ggplot(aes(x=reading_time, y=value, group=well))+
  facet_wrap(~original_time)+
  geom_point()+
  geom_line()+
  theme_classic()

plot

This plot shows the resazurin value (y-axis) across the repeat time points at 2, 3, 4, and 5 weeks (shown in hours). Each facet shows the original time point the samples were taken during the 2 hr assay (taken every 30 minutes).

Here is what I noticed:

• Values increased over the incubation period (0-2 hours) as expected.
• Values appear to increase after hour 2 (the time point at which we care about the value relevant to the assay) peaking at 3 weeks after the assay and decreasing 4-5 weeks after the assay.
  
• Except for one case in the 2 hour group, the rankings of samples does not appear to change even though the values change, suggesting relative differences between samples are not affected

Run a linear mixed effect model to see if time is a significant factor.

model<-data%>%
  filter(reading_time>=2)%>%
  
  lmer(value~as.factor(reading_time) + (1|well) + (1|original_time), data=.)

anova(model, type="II")
summary(model)

Type II Analysis of Variance Table with Satterthwaite's method
                        Sum Sq Mean Sq NumDF DenDF F value    Pr(>F)    
as.factor(reading_time) 147936   36984     4   236  82.608 < 2.2e-16 ***

Reading time is significant.

See what the posthoc comparisons are.

emm<-emmeans(model, ~reading_time)
pairs(emm)

I am most interested in the time 2 measurements vs subsequent time points. All of the subsequent time points are different from the original readings with the greatest difference at weeks 2-3 (3-4% difference from original values) and smaller differences at weeks 4-5 (1.5-3% differences from original reading).

contrast                          estimate   SE  df t.ratio p.value
 reading_time2 - reading_time336     -55.47 3.86 236 -14.358 <0.0001
 reading_time2 - reading_time504     -63.58 3.86 236 -16.459 <0.0001
 reading_time2 - reading_time672     -35.97 3.86 236  -9.310 <0.0001
 reading_time2 - reading_time840     -30.08 3.86 236  -7.787 <0.0001
 reading_time336 - reading_time504    -8.12 3.86 236  -2.101  0.2230
 reading_time336 - reading_time672    19.50 3.86 236   5.048 <0.0001
 reading_time336 - reading_time840    25.38 3.86 236   6.571 <0.0001
 reading_time504 - reading_time672    27.62 3.86 236   7.149 <0.0001
 reading_time504 - reading_time840    33.50 3.86 236   8.672 <0.0001
 reading_time672 - reading_time840     5.88 3.86 236   1.523  0.5486

Overall conclusions

The measurements are largely stable over time, but do deviate from original readings by 1.5-4% depending on time point. The samples behave uniformely, so relative differences between samples are unlikely to be affected. I think we can proceed with using samples at later time points for field work or places where having a plate reader are not possible.