Dear community,
I am facing some issues and scientific questionings regarding the dose-response analysis using the drc package from R.
Context :
I want to know if two strains have different responses to three drugs. To do so, I am using the dcr package from R. I then determine the EC50 for each strain regarding each drugs. I later plot my EC50 and use the estimate and standard error to determine if the EC50 is statistically different between strains. For each strain and drug I have four technical replicates and I will have three biological replicates. Visually, the model produced by the package matches my experimental data. However, I am looking for a statistical approach to determine if the model given by drc is not too far from my experimental data. How to know if I can be confident in the model ?
My approach :
I am using mselect() to determine which drc model is the most accurate with my data. However, I do not know how to interpret the results. I read that the higher the logLik is, the best the model describes the data provided. But do you know if a threshold does exist?
For example I have from the mselect() :
> mselect(KCl96WT.LL.4, list(LL.3(), LL.5(), W1.3(), W1.4(), W2.4(), baro5()), linreg = TRUE)
logLik IC Lack of fit Res var
LL.3 101.90101 -195.8020 0 0.0003878212
LL.3 101.90101 -195.8020 0 0.0003878212
W1.3 101.53204 -195.0641 0 0.0003950424
W2.4 102.48671 -194.9734 0 0.0003870905
LL.5 103.05880 -194.1176 0 0.0003869226
W1.4 101.52267 -193.0453 0 0.0004062060
Cubic 101.42931 -192.8586 NA 0.0004081066
baro5 101.98930 -191.9786 0 0.0004081766
Lin 96.45402 -186.9080 NA 0.0004958264
Quad 96.64263 -185.2853 NA 0.0005044474
I also used the glht() element and coeftest(KCl96WT.LL.4, vcov= sandwich). But I am facing the same issue.
> coeftest(KCl96WT.LL.4, vcov= sandwich)
t test of coefficients:
Estimate Std. Error t value Pr(>|t|)
b:(Intercept) 4.3179185 0.6187043 6.979 3.024e-08 ***
d:(Intercept) 0.0907908 0.0080186 11.323 1.397e-13 ***
e:(Intercept) 0.9809981 0.0686580 14.288 < 2.2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Do you know what approach could indicate if I can be statistically confident regarding my model? Can I be mathematically confident in the EC50 given by the package?
Thanks for your time! I am looking forward to discover new ways to be more critical regarding my data analysis. If you have any questions or comment regarding my approach, feel free to ask me !
See the link: https://www.google.com/search?q=Dose+Response+curves+R&rlz=1C1CHBF_enUS917US917&oq=Dose+Response+curves+R&aqs=chrome..69i57j0i22i30l9.19372j0j4&sourceid=chrome&ie=UTF-8
Best wishes, D. Booth
- Badges
- Science topic
- Similar topics
- Biological Science
- Immunology