Hello all. I have several batches of microbiome data derived from unique users using 16s with Illumina. The fastq files from a part of all the data has a slightly higher read count but slightly lower average quality score. These differences are probably caused by a change of protocol, they appear after a certain date from which the read count has increased but quality decreased. Using an unsupervised embeddings method the microbiome data separates perfectly into two clusters based on when it was clustered, before or after a specific date.
The lab part of the hospital is now going to find out what happened in the protocol. But regardless of their answer I will need to correct for this batch effect. Since it clearly overshadows the signal of interest. I am familiar with scRNA-seq batch effect correction methods like mutual nearest neighbours. But those are not suited for this data since there is no overlap of samples between batches as each microbiome has been sequenced from a unique patient.
So my question is does anyone know how I could correct for these batch effects while preserving the unique signals of each patient? And how can batch effect of unique samples remain after normalization of the data?
I think this discussion should be useful: Batch correction in DESeq2 I don't know how you are performing the statistical analysis but if it's a linear model of some sort the batch can be added to the design formula.