I had two queries related to the LMM model equation where the fixed and random effects can be provided.
1. I have a RNA-Seq experiment where the samples have been isolated from multiple tissues, where in some patients multiple tissues represent same patient. I have attached an example metadata sheet below for your understanding. And from the same patient multiple timepoints of samples have been isolated. So here both the PatientID (multiple samples have been isolated from same patient at different timepoints) and Tissue ID (multiple Tissues are present representing same patient ), should be the random effect? While factors like Age and gender as the fixed effect? Kindly correct me if my formula or understanding is wrong.
modelFormula = ~ Sample.Condition + (1| Patient_ID) + (1|Tissue_Number)
The problem is, its not that all patients are represented by multiple tissues, only selected patients are represented by multiple tissues. In such cases how do we provide the random effect factor, within the equation?
2. When comparing samples that comes from same tissue (e.g. Timepoint0 vs Timepoint24hr in same tissue), a random slope is included in the LMM model. I have the case where I compare control vs test, where some of the patient tissues contain both timepoint0 & Timepoint24hr while certain patients contain only timepoint0 (like P4 in above table having only Control). In such cases how do we provide the Random slope? Because I get the following error, were it clearly mentions that :
Error in checkForRemoteErrors(val) :
16 nodes produced errors; first error: number of observations (=6) <= number of random effects (=8) for term (1 + testVariable | Patient_ID); the random-effects parameters and the residual variance (or scale parameter) are probably unidentifiable.
I think it would be helpful to include additional information about your study design, sample size, and what your specific research question is. It will be difficult for anyone to provide meaningful feedback without this basic information.
Thanks Chris for your suggestion. The study design is basically transcriptomic sequencing of carcinoma patients. The overall sample size is 28, but this includes multiple timepoint data included from the same patient. So for example purpose we can keep sample size to 10 as I have provided in above attached table. I would like to compare Test vs Control groups in the above table but I am stuck with certain covariates in the data like PatientID and TissueID, where I want them to be added as random effect factors to equation in Lmm model (lmer) R package. Hope I have explained the background of the study which will now help you to answer my question.