Batch-specific contaminant detection or not?

[zztin]

Hi Benjamin,

Thank you for developing this package. We are using it to exclude contaminants for the purpose of identifying sepsis-causative bacteria in newborn foals.
Can we ask your advice on whether to use a batched analysis or not in our analysis?

We have 32 foals in total, across 6 isolation batches, and 3 library prep batches (or xxx batches per library prep kit). We prefer to do a non-batched analysis AND a batched analysis and exclude all contaminants that were found in any of the comparisons (frequency-based method). However, some batches are too small (n = 3), thus we excluded these batches and performed the isContaminants with batch parameter on the remaining batches. The batch sizes are still quite small (n=4-11). On the other hand, we don't want to miss batch-specific contaminants.

1. Would you recommend a combined approach like we proposed in this case? Or a batched approach/non-batch specific approach?

Attached is a figure with the histogram distribution of p-values of all species that was present in at least 6 samples. On the left, p-value distribution without batch-specific detection. On the right, p-value distribution with batch-specific detection ('minimum').

2. It is hard to find a trough in the p-value distribution. We suspect that it is because we are working with low-biomass sample (with little bacterial DNA in the samples). To be lenient, we proposed to use 0.25 as a p-value cutoff to exclude contaminants. However, we would also like to hear your input.