In theory, most integration methods should keep biological distinct cells, well, distinct from each other, only accounting for technical variation between highly biologically similar entities (e.g. T cells from two different people).

In reality, many of them suffer from a heavy hand and force together cells that are "similar" even if they have meaningful biological differences. In this case, since you already have a good idea of what's happening and what should be different, you could try the reciprocal PCA method from Seurat, which allows you to modulate the strength of integration. This would allow you to ensure your unaffected cell compartments show good overlap while your myeloid compartment still retains differences between the conditions.