Problems with host-guest calculations -- paprika

[jeff231li]

I'm having some issues when running HG calculations with the paprika_integration branch separate from issue #224 and I don't think I have the necessary expertise to fix these bugs (I don't understand dask).

Issue 1

With commit aef4e06 the calculation for releasing the host restraints is only performed once. Before this, the release calculations were repeated for each HG system (redundant). Basically, when I run the calculations on a cluster evaluator is only successful in calculating one HG system while the rest fails to run. I also tried running this locally, and the log file shows that it is failing when evaluator is trying to add the energies from the different phases (add_per_orientation_free_energies). evaluator spits out the following error:

13:02:34.874 INFO     Exceptions were raised while executing batch d553ea07ca164dd79b3030e6f60fb860
13:02:34.874 INFO     e4ed6f0a0fa44553901012448561208e|add_per_orientation_free_energies_0 failed to execute.

Traceback (most recent call last):
  File "/home/openff-evaluator/openff/evaluator/workflow/protocols.py", line 1245, in _execute_protocol
    protocol.execute(directory, available_resources)
  File "/home/openff-evaluator/openff/evaluator/workflow/protocols.py", line 704, in execute
    self._execute(directory, available_resources)
  File "/home/openff-evaluator/openff/evaluator/protocols/miscellaneous.py", line 49, in _execute
    self.result += value
TypeError: unsupported operand type(s) for +=: 'UndefinedAttribute' and 'UndefinedAttribute'

So... the directory for the host-only calculation is not funneled properly to the workers, hence the add_per_orientation_free_energies protocol is failing?

Issue 2

This problem is more subtle as there is no error printed by evaluator. When I run the HG calculations without the commit mentioned above (i.e. aef4e06) the program will run as expected. However, when analyzing the free-energies I noticed that the results (off-v1.2.0_bcd.zip) for bcd-m4t and bcd-m4c are exactly the same (-5.710330040754208 +- 0.6205644212487271 kcal/mol). I tried running the commands below inside the worker-logs directory

• grep m4t *
• grep m4c *

and found that nothing was returned for the m4c case. Thus, evaluator is only calculating the bcd-m4t system but is assigning the results it calculated for this system to bcd-m4c. Is this a bug with the task graph in evaluator?
(Note: I did not observe any duplicate results with calculations I did at the start of 2020)

[SimonBoothroyd]

Hi @jeff231li,

Thanks for raising these issues! I'll try and take a look later today or if not early next week.

[dotsdl]

I will take this on. Working with @jeff231li on this now.

[SimonBoothroyd]

Issue 2 should now be fixed by #254 once merged into the paprika_integration branch - a big thanks for reporting this! Luckily it doesn't seem to have affected things other than the paprika binding affinity calculations.

I'm looking into Issue 1 now.

[SimonBoothroyd]

It is possible that Issue 1 was actually caused by the same bug as Issue 2 - I would recommend updating the paprika branch to master and testing again.

[jeff231li]

@SimonBoothroyd thank you for looking into this. I'll do the update to the paprika branch and run another set of calculations. Will get back to you shortly.

[jeff231li]

So far so good, all of the HG systems are running as well as the host-only calculations. Hopefully, evaluator will report the results for the bcd-m4t and bcd-m4c systems properly. Will report back when the calculations are done.

[jeff231li]

I checked the results and all of the values (attach-pull-release) summed up correctly. Also, evaluator now calculates the bcd-m4t and bcd-m4c systems properly. Thanks @SimonBoothroyd for fixing this bug.

host	guest	fe (kcal/mol)	sem (kcal/mol)
bcd	ben	-1.54	0.67
bcd	cbu	-2.34	0.68
bcd	chp	-5.14	0.67
bcd	coc	-6.57	0.81
bcd	cpe	-2.59	0.62
bcd	ham	-2.70	0.68
bcd	hep	-3.37	0.78
bcd	hex	-2.13	0.66
bcd	m4c	-4.28	0.62
bcd	m4t	-4.72	0.62
bcd	mch	-5.01	0.64
bcd	mha	-2.90	0.64
bcd	mo3	-2.52	0.74
bcd	mo4	-3.90	0.76
bcd	mp3	-2.31	0.71
bcd	mp4	-1.97	0.67
bcd	oam	-4.01	0.64
bcd	pb3	-4.16	0.78
bcd	pb4	-4.80	0.79
bcd	pha	-3.50	0.70
bcd	pnt	-2.15	0.68

[SimonBoothroyd]

Great to hear! And thanks for raising the issue - this could have caused some nasty scientific issues in a number of different places had it not been caught!

[SimonBoothroyd]

Feel free to re-open this issue if any more problems pop up.

[dotsdl]

Awesome, thank you @SimonBoothroyd and @jeff231li! I'm happy that this was resolvable!