Reproducibility is huge problem and most distribution systems like GEO are not well suited for distributing data other than a few simple and known types like fastq, fasta etc.

Here is what you can do:

1. Check the supplementary data in the paper - these typically are notoriously obtuse and convoluted to deal with. Often are not described in sufficient detail hence you will need to download and open each to see what is inside them.
2. Your actual best bet is to contact the authors and ask for this data (or have them point out where this data is).