Add scoring to fragmenstein multiplex

[phraenquex]

Check whether fits in ALC2 budget. @tdudgeon?

Which of these scores can be straightforwardly added?:
SuCOS
PointVS
CoPriNet
Interaction preservation
Ro5 (?)
...

Discuss / finalise based on effort.

Required for ALC2 milestone.

[phraenquex]

Conclusion from chat (this morning):
The following scores should be computed for EVERY compound generated by the existing Fragmenstein job, and added to the returned SDF.

• SuCOS
• Open3Dalign - quality of alignment to fragments
• simple SA score

Out of scope, and requiring a new ticket (action: @tdudgeon):

• No DL-based scores, as they need to solve two technical problems:
  • Set up Docker container framework - right libraries and stuff
  • Add GPUs to the K8 cluster - configuring K8 (well-documented but potential faff)
  • Squonk job definitions need to specify GPU (maybe just an extra label)
  • Unknown time to containerise PointVS and CoPriNet.
• No FE options to configure the scoring, as that is for Implement generic squonk job configuration as backend feature #944 .
  • As part of that work, the scorings in this ticket will be "unwound" (Tim's words) from the existing Fragmentstein jobs.

[tdudgeon]

#1026 created to handle the first step - making GPUs available to K8S

[phraenquex]

@tdudgeon estimates ~1 day, plus minor FE work to update the configuration.

[phraenquex]

Needs hacking (something about gitlab allowing private registries, because of PyRosetta licensing.)

Options:

• PlanA: create a new repo, combining the artefacts (fragmenstein + simple agreed-on scorings). Will be discarded later when workflows and stuff properly implemented as part of ALC3.
• PlanB: @matteoferla create a "dumbed down" Fragmenstein to get rid of PyRosetta (but doesn't seem like a good idea).

[matteoferla]

X-linking to Fragmenstein PyRosetta > OpenMM shift enhancement: matteoferla/Fragmenstein#33

Regarding Fragmenstein without the in protein minimisation, this is the code (xposted):

from fragmenstein import Victor
from fragmenstein import __version__ as fragn_version
from fragmenstein.demo import Mac1
# RuntimeWarning: PyRosetta is not installed. A mock object is loaded. Any calls will fail.
print(fragn_version)
# 0.9.12.6
from rdkit import Chem

class Wictor(Victor):
    """
    This Victor does not call Igor
    """
    def _calculate_combination_thermo(self):
        # override igor.
        pass
    
    
    def _calculate_placement_thermo(self):
        # override igor.
        pass
    
    def post_monster_step(self):
        # this is a black overridable methods that will be the last thing called
        self.minimized_mol: Chem.Mol = self.monster.positioned_mol
        self.minimized_pdbblock: str = self.unminimized_pdbblock

Wictor.capture_rdkit_log()
#Wictor.enable_stdout()
Wictor.error_to_catch = ()
wicky = Wictor(hits=Mac1.get_n_filtered_mols(2),
               pdb_block=Mac1.get_template(),
              )
wicky.combine()

NB. odd behaviour is seen however with wicky.to_nglview() as the data is incomplete. In particular there is no ∆∆G.
The rdkit FF score is however available.