id	summary	reporter	owner	description	type	status	priority	milestone	component	version	resolution	keywords	cc	blockedby	blocking	notify_on_close	platform	project
6959	AlphaFold prediction energy OpenMM minimization always fails	Tom Goddard	Tom Goddard	"Tomas Fernandez noted in a YouTube comment of a ChimeraX AlphaFold video that AlphaFold runs that used to work are now failing in energy minimization.

Tomas writes:

""Nevertheless, and in order to confirm that this error is dependent on the predicted structure, I tried to predict the structure of a 70 amino acid protein whose model has been calculated several times through AlphaFold in classes by many students, and the error still pops up: ValueError ­ Traceback (most recent call last) in () 692 seq_list = seq_list[1:] 693 --> 694 run_prediction(seq_list, energy_minimize = not dont_minimize) 5 frames /usr/local/lib/python3.7/dist-packages/alphafold/relax/amber_minimize.py in _run_one_iteration(pdb_string, max_iterations, tolerance, stiffness, restraint_set, max_attempts, use_gpu, exclude_residues) 417 logging.info(e) 418 if not minimized: --> 419 raise ValueError(f""Minimization failed after {max_attempts} attempts."") 420 ret[""opt_time""] = time.time() - start 421 ret[""min_attempts""] = attempts ValueError: Minimization failed after 100 attempts.""

I also tested on 128 amino acid 7mrx chain A and it failed in the same way although it works correctly with full alphafold on minsky.

So it looks like OpenMM is somehow broken now on Google Colab.  ChimeraX installs fixed versions of AlphaFold and OpenMM and all dependencies but possibly some implicit dependency without a specified version number updated and broke things.  Another possibility is that Google Colab changed, possibly updating their CUDA version and that broke OpenMM.
"	defect	closed	moderate		Structure Prediction		fixed						all	ChimeraX