Industry benchmark failure cases

[hannahbaumann]

Comparison of old vs new Kartograf mappings. Main difference are the bond break filter and the hydrogen-to-hydrogen only mapping filter in the new (kartograf 2.0) version.
mapping_comparison_3d.html

charge_annihilation_set, jnk1

• edge 19a_charg - 19b_charg
• charge change of 2, not supported by the protocol

jacs set, ptp1b

• edge 23480 - 20670
• From 5 repeats, one complex leg failed, the rest and all solvent legs completed successfully
• Could be an equilibration issue

jannsen bace, bace_p3_arg368_in

• edges 17f - 17g; 27f - 27g; 8f - 8g;
• All three transformations failed in most attempt (NaN)
• likely due to hybridization change in the mapping (see html link above)
• simulations sometimes ran successfully after many restart attempts

merck, shp2

• edges SHP836-2 - 6; SHP836-2 - Example 27
• No failures
• Potentially these edges were not included in the old ligand network
• SHP836-2 --> 6: High MBAR error

merck, cmet

• edge CHEMBL3402750_400 - CHEMBL3402748_5300
• No failures
• New Kartograf filter avoids bond break

merck, syk

• edge CHEMBL3265009 - CHEMBL3265010
• No failures
• New Kartograf filter avoids bond break

misc set, faah

• edge 6 - 31
• No failures

misc set, galectin

• edge 17a - 17b
• No Failures
• But some complex simulations needed multiple attempts, potentially also a case that could benefit from improved equilibration?

misc set, hiv1 protease

• edge 13 - 20
• No failures

[hannahbaumann]

For the galectin system, there are water molecules in close proximity to the ligand which could have lead to random crashes.

[jthorton]

Thanks @hannahbaumann this looks great! For the Jannsen bace, bace_p3_arg368_in how about grouping the inputs and outputs for each transformation into a folder ? Its not vital but I can see it getting confusing when a user submits a large number of errors.

[hannahbaumann]

The crash in the one ptp1b edge is likely caused by the close proximity of this water molecule to the ligand, which is close to the area where the crash happened.

[jthorton]

Thanks @hannahbaumann this is a great start!