Deseq2 opposite pattern than raw files in only a subset of the results

Hello, I have an experiment where we performed 16s analysis at three-time points on human samples. On the ASV count tables, we performed deseq2 to see differential genera between time points. The deseq2 results matched the ASV counts (meaning if a genus showed higher counts in time point 1, it showed positive log fold change as well in deseq2), with the exception of two genera that displayed opposite patterns comparing the raw count and deseq2logFC. Any possible explanation for this observation, which occurs only on a subset of the results and the majority where consistent with raw files.
Thanks in advance