How to use batch normalized data for scanpy analysis

I got a MNN batch normalized data (.txt) form published paper, the data contain several subjects and batch normalized by mnnpy method. Then I set up AnnData with this txt file, following the “Integrating data using ingest and BBKNN” tutorial (scale, pca, find neighbors, umap) without remove batch effect by ingest and BBKNN, but the umap plot still showed an obvious batch effect. Does anyone know how to figure out this problem?