I have been recently using CyteType while re-annotating a merged public dataset together with some of our own data, and it’s honestly one of the most interesting things I’ve tried in a while.

It looks at both supporting and conflicting markers, checks pathway context, and even rejects a label if the signal doesn’t make sense. In my case it caught a small cluster that turned out to be ambient RNA and low-quality debris, which every other tool labeled as microglia.

What’s neat is that it shows the reasoning behind each annotation, so you can see exactly how it reached a call. They’ve just put out a preprint explaining how their multi-agent system works across datasets. If you spend time cleaning or re-annotating complex single-cell data, it’s definitely worth a look.