Correct Type For Operation: DataAdapter Vs. DataReader

I am reading multiple databases into a single dataset, putting each read into a new datatable in the set. Ultimately, I want to merge the multiple tables (they are all from different DB tables in different locations using the same schema). I am also combining this with data from an Oracle database which is in yet another table in the same dataset.

Here is my question:

I will not be propagating any changes to this data back to its original source, but I will be populating it out to Active Directory. (Don't worry; I'm not asking any AD questions here.) But I do want to run SELECT, DELETE, and UPDATE queries against the data in the dataset in memory without writing it out to a separate file. Based on that type of usage, would I be better to use a DataReader or a DataAdapter to do the read from the original database? If I use a DataReader, can I still update the data in memory? In either case, what's the correct methodology to update data in a dataset in memory?

Thanks!