Simplify delivery and interfacing

Hi guys,
congratulations for the Nature methods paper on this.

I have a couple of suggestions:

- support running the tool from more standard input files rather than .biom (.csv, .tab. etc.), that do not need special libraries to handle.
- specify versions of all python libraries needed. Currently, this does not install, as you appear to be using an old version of tensorflow. Even better, provide a Docker image.
- minimize dependencies as far as possible. Even better, provide a Docker image.

When other labs will try to publish similar methods, they will be expected to compare the results to your method and these will make peoples life easier.

Kind regards,
Martin

No files, tests, or entry points are named. Start by locating the input handling and Python dependency configuration, then determine the scope of supporting CSV/tabular inputs, specifying library versions, reducing dependencies, and optionally providing a Docker image; done means these usability and installation concerns are addressed.

Written by the indexing model from the issue text.