What to do with selection tuples?

[cgreenberg]

So in PMI1 a restraint is typically set up with:

x = Restraint(representation = r,
            selection_tuples=[('A',100,150),('B',200,235)],
             resolution=10)

Then the restraint internally does the selection. But now in PMI2 we have nice "handles" and also IMP Selection works great with all resolutions (including densities). We could now just have all restraints be set up like so:

x = Restraint(hiers = [molA[100:150],molB[200:235]],
            resolution=10)

So we'd get rid of selection tuples altogether. What do you think? @Pellarin ?

[Pellarin]

Yes, but we still need backcompatibility. How can we prevent to broke previous code?

[benmwebb]

Why not just pass a Selection object to each PMI restraint, rather than passing Selection-like parameters (e.g. resolution, residue indices) to each one?

[cgreenberg]

I'm all for that. Only problem is that many PMI restraints do the selecting for you - they read in data and select within the passed hierarchy. E.g. crosslink data or elastic networks on SSE elements. Maybe the way to do it is have PMI helper functions which create a selection from input data?

[benmwebb]

Perhaps... or we make Selection itself more flexible.

[cgreenberg]

Proposal after talking with @saltzberg : have all data parsers in IO (including CrosslinkDatabase, DSSPParser, ...) have a select(hier,**kwargs) function which: 1) selects particles in hier according to the underlying data (e.g. gives lists of lists of particle pairs for crosslinks) and 2) admits additional selection keywords. So you could, for example, create a customized XL selection with:

xldb = IMP.pmi.io.crosslink.CrossLinkDataBase()
xldb.create_set_from_file("xlinks.csv")
pps = xldb.select(hier,resolution=0,atom_type=IMP.atom.AtomType("NZ"))
r = IMP.pmi.restraints.CrossLinkingMassSpectrometryRestraint(particle_pair_lists=pps,
                                                             slope=0.0,
                                                             length=21.0,
                                                             label="XL")

Thoughts?

[benmwebb]

1) looks fine but 2) duplicates the existing Selection framework somewhat, so you'd end up writing a lot of duplicated code.

[cgreenberg]

I'd have it internally call IMP Selection and just pass the kwargs on.