Hi @mkarikom, I think concatenation here is only defined for the same feature sets. In this case it seems the peaks were called separately on different samples.
Thanks @gtca, I ended up re-using the 10x cell-calling and reducing the peaks in signac using the raw ATAC data, then substituting these features for the ones generated by the per-batch peak-calling. In this case, the cell-calling previously performed for each batch was recycled.