version 3.0: adjust regularization weights, add examples

[jbloom]

This version makes the uniqueness and spatial regularization part of the standard package. Changes regularization defaults and a few other non-backward-compatible changes.

[jbloom]

@Bernadetadad @caelanradford @Caleb-Carr @fwelsh @timcyu @matsen, I just merged this pull request which creates version 3.0.

The main changes are summarized in the CHANGELOG, and include everything in both version 2.6 and 3.0: https://github.com/jbloomlab/polyclonal/blob/main/CHANGELOG.rst

Documentation is here: https://jbloomlab.github.io/polyclonal/index.html

Key things:

• You can now add spatial distances and do spatial regularization. @caelanradford, hopefully this can address some of what you were discussing in in #132. Note that this requires you to add spatial distances when initializing a Polyclonal object, and a function was added to help get those distances in the correct format (https://jbloomlab.github.io/polyclonal/polyclonal.pdb_utils.html#polyclonal.pdb_utils.inter_residue_distances). Note you should think carefully about which chains to include when calculating distances: for a trimeric protein, you want to use all chains in that trimer in case there are inter-monomer epitopes. See https://jbloomlab.github.io/polyclonal/real_serum.html for use of this regularization for HIV Env.
• The "similarity" regularization @timcyu added is now called "uniqueness" and there are weights for both the uniqueness and squared uniqueness.
• I've changed the activity regularization in a way that should be better and should make the regularization mostly insensitive to the units used for concentration. This may actually help quite a bit in fitting multiple epitopes: in particularly, it helps prevent secondary epitopes from immediately being fit to highly negative activities. If you are having problems with multiple epitopes, you could even try increasing reg_activity_weight more.
• The regularization is now more conceptually summarized here: https://jbloomlab.github.io/polyclonal/regularization.html

I have not yet integrated this into dms-vep-pipeline, but can do so. But first, can you take a look over above and see if there are any relevant issues you spot that should be addressed before integrating into pipeline? If you find any, just post them here.

[fwelsh]

Haven't gotten to testing the update yet, but minor note for the docs. My last seq run returned much better antibody cocktail data now that reflects updated selection conditions. Here's the prob_escape file:

libA_221108_1_1C04-5G04_1_prob_escape.csv.zip

Would suggest updating the antibody cocktail example in the docs with this data instead!