18 months ago by
It's hypothetically feasible, to a certain extent. The GEOmetadb package will allow you to find all arrays by species. Given that list you could use GEOquery to download the GPL file for each of those arrays and search for a gene of interest (or if you really want to hang with the cool kids, you could dump the GPL data into a SQLite DB, so you can make arbitrary searches in future).
Once you know all the GPL IDs that contain the gene of interest, you could use the gse_gpl table in GEOmeta.db to get all the series that are on the arrays you care about, and then use gse_gsm to map to the actual samples.
You won't be able to do anything with a range of p-values, as GEO is intended as a repository of data, not results. Submissions only contain information about the underlying experiment and the data, but no statistics.