I have data from the our experimental model - where we analyze the immune response following BCG vaccination, and then the responses and clinical outcome following Mtb infection of our vaccinated models. Because we cannot experimentally follow the very same entity after evaulating the post-vaccination response also for the post vaccination plus post infection studies - we have such data from different batches. Is it possible to do correlation here between post vaccination responses of 5 replicates in one batch (in different vaccine candidates) versus 4-5 replicates in vaccination & infection from another batch? I ask this because we are not following up the same replicates for post vaccination and post infection measurements (as it is not experimentally feasible). If correlation is not the best method, are there other ways to analyze the patterns - such as strength of association between T cell response in BCG vaccinated models versus increased survival of BCG vaccinated models (both measurements are from different batches)? We have several groups like that, with a variety of parameters measured per group in different sets of experiments.
Thanks for your responses and help.
To make it a bit simpler:
say you have treatments A and B, and your experiment is done in two batches 1 and 2.
If treatment A is analyzed in batch 1 and B in batch 2, then treatment and batch are perfectky confounded you you have no chance to distangle batch-effects from treatment effects.
If samples with treatment A are measured in batch 1 and 2, and also samples with treatment B are measured in both batches, then one can model the batch effect and reveal the treatment effect.
If you have treatments A+C in batch 1 and treatments B+C in batch 2, you might estimate the batch effect from treatment C and apply it to correct A and B as well (dangerous, if the batch effect also depends on the treatment, but better than nothing).
Thanks very much Jochen Wilhelm . In our case, we have 8 groups (based on different TB vaccine candidates and sex) - always like that. But we cannot measure all parameters in one batch of 8 groups. For example, in one experiment round, we would be looking at some of the immunological parameters (the same for all 8 groups, together), and in another experiment, their clinical disease progression. Now, the mice we used for the immunological evaluation are different from the one we follow up to understand the clinical disease progression. Looking at the data, we can separately observe that a Vaccine A has, say, increased CD 8 T cell response, and in the separate follow up experiment - vaccine A increases survival. Yet, in another experiment, we model the cytokine profiles in all these groups. So visually, if we say Vaccine A is the best performing one in clinical scores - we can also see from other experiments that vaccine A is, for example, associated with a better CD 8 T cell response, and a different cytokine profile and other things. Now, is it possible to look at this data mathematically, to find a pattern - in all the vaccines, the survival data is strongly associated with CD 8 T cell response, or it is true just in the case of vaccine A, for example. of course, these are things, one could see visually from different plots - but my question is, can we use a mathematical modelling or pattern analysis to sort of say this in an unbiased way?
Thanks really a lot for your very detailed answer. Much much appreciated.
In other words : if you have Treatment A and B, always, but in one experiment you look at the CD8 T cell response, in another, their cytokine profile, yet in another round their B cell response - so and and so forth, and lastly, also the clinical progression following treatment A and B. Now you can, maybe, visually already see that treatment A gives better survival outcome, and treatment A has increased CD 8 T cell response. Now, my question is, how do you find this patterns or associations in an unbiased mathematically robust way (rather than a visual conclusion)? Are the some modelling systems or pattern recognition algorithms that we could use? This is question I am looking into.
Thanks a huge lot for your most kind reply @Jochen Wilhelm
As I understood is each treatment (effect) estimated within a "batch" (or experiment). In this case you can correlate the estimated effects between different experiments (e.g. correlate the change in CD8+ cells to survival time).
Indeed. Thanks very very much. That's exactly the case. Each effect is estimated within a batch. So then, as I understand it - it doesn't matter that the CD8 T cell response was estimated in a different batch of experiment involving both treatment A and B groups, and survival in a separate experiment (batch) involving treatment A and B groups. We can correlate the estimated effects between these parameters (CD 8 T cell response and survival) obtained from different experiments, as long as, in each of the experiments the treatment groups (and other conditions) were the same. I hope I have understood you right. Thanks once again. You have been very helpful and kind in helping me out.
Yes, I think so. I hope other experts reading this will correct us if this was not the case.
Thanks very very much. Wish you a very good day ahead @Jochen Wilhelm
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Infectious Diseases
- Immunization
- Vaccination