Decreased gene expression in feature plots depending on samples integrated

[hannahvanm]

Hi there, I'm hoping to get an explanation for what I'm seeing and if it is normal / to be expected with the way that Seurat handles integration.

I have two main multiomic datasets I'm working on (T20_CNC and T24_TNC), for which gene expression is integrated with rpca. Shown here is a feature plot of a gene, split by sample.

With the same data (T20_CNC and T24_TNC), I added another sample (T24_arches, scRNAseq) and found that some genes showed very different expression patterns in the feature plots. They are integrated with the exact same pipeline and the only thing that differs is the addition of the third dataset.

When all three datasets are used, the gene shows significant overlap with marker genes I'm interested in, but without the third dataset this expression and overlap entirely disappears.

So, can someone explain if this is because with the addition of the third dataset, scaling is change dramatically? Or if this is due to which variable features are chosen in the integration steps due to the inclusion of the third data set? Or if anyone has other explanations, that would be appreciated

Also, any suggestions to get the expression shown in the 3 sample integration to show up in the two samples on their own would be much appreciated, as I'm sure there are many genes that aren't showing up for a similar reason. Thanks!