I have noticed an inconsistency in the scoring of a designed sequence on using the "aa_composition" term.
Inconsistency: A designed sequence is showing a deviation from the desired composition by one amino acid and thus a penalty must be reported by the ScoreType filter (scorefunction:ref2015 score term weighted by "aa_composition=1"). However this is not the case when the design was evaluted using the ScoreType filters after a set of design protocols, the raw_ref2015 and weighted_ref2015 scorefunctions show the same scores. Surprisingly, the same designed sequence was re-scored using the raw_ref2015, weighted_ref2015 scores separately in a different XML script (note that in this script no other design/structural change movers are present) correctly reported the penalty. For example:
The designed sequence: ALA,ALA,GLU,ASN,GLY,GLY,LEU, relax_renum_0041 showed a score of -310 REU and -310 REU (no difference in the raw_ref2015 and weighted_ref2015 scores) when scored after all the design protocols. This is wrong (not expected ?).
The same sequence (relax_renum_0041_0001) showed a score of -307 REU, -297 REU ( a difference of 10 REU in the raw_ref2015 and weighted_ref2015 scores) when re-scored separately in the absence of other movers. Since we have three amino acids (one less than four amino acids) from the specifed list in the designed sequnce, we expect an penalty of 10 REU (weight of aa_composition is 1).
PENALTY_DEFINITION TYPE ALA DAL PHE DPH ILE DIL LEU DLE MET DME PRO DPR VAL DVA TRP DTR DELTA_START -3 DELTA_END 1 PENALTIES 100 25 10 0 0 ABSOLUTE 4 BEFORE_FUNCTION QUADRATIC AFTER_FUNCTION CONSTANT END_PENALTY_DEFINITION
I am perplexed to why the exact same ScoreType filter (and scorefunction) is unable to report the scores correctly when it appears after some design/structural change protocols but it works correctly in their absence.
Any insights would be very helpful. I can also share the XML scripts for better clarity if needed.