When is trajectory preprocessing (whole/cluster/nojump/center) actually required for analysis?

User discussions
, ,
1

GROMACS version: 2024.4
GROMACS modification: No

When performing analysis on MD trajectories of a peptide condensate (cubic box), I am trying to understand which analyses strictly require the full GROMACS preprocessing pipeline:

whole → cluster → nojump → center

From my understanding so far:

  • gmx rdf — uses pairwise distances with minimum image convention internally, so raw trajectory is sufficient
  • gmx hbond — same reasoning, raw trajectory is sufficient
  • π–π stacking (custom script using ring centroid distances) — same reasoning, raw trajectory is sufficient
  • gmx sasa — requires processed trajectory because the rolling probe sphere depends on correct neighbor identification across PBC
  • gmx gyrate / gmx rg — requires processed trajectory because Rg depends on absolute distances to COM
  • gmx dssp — raw trajectory is sufficient, local geometry only
  • gmx clustsize — requires at minimum -pbc whole, since broken molecules across PBC will be miscounted as separate clusters

Questions:

  1. Is this reasoning correct?
  2. Are there any edge cases where RDF/hbond calculations would fail on an unprocessed trajectory (e.g. very large box, molecules diffusing far from the condensate)?
  3. For SASA specifically, is -pbc whole alone sufficient, or is the full pipeline (including cluster and center) needed?
  4. When processing a trajectory of a large multi-peptide condensate, a small number of frames (~10%) show the condensate apparently split inside the box, likely because a transiently detached region is treated as a separate cluster by -pbc cluster. The remaining ~90% of frames look correct. What is the recommended approach: exclude those frames, accept them as physically real detachment events, or is there a more robust preprocessing strategy for loosely associated condensates?
  5. When using MDTraj’s image_molecules() function, the condensate appears perfectly centered and whole across all frames, including the problematic ones. This seems analogous to adding -fit rot+trans after the last centering step I mention above in the GROMACS pipeline. However, fitting suppresses real conformational fluctuations and shape changes of the condensate, so it seems only appropriate for visualization and not for quantitative analysis (Rg, SASA). Is this reasoning correct, i.e., should fitting/imaging-based centering be restricted to visualization only, and all quantitative analysis performed on the unprocessed or minimally processed (whole-only) trajectory?
2
  1. Yes.
  2. no.
  3. If you want the SASA of the cluster, you need to make the cluster whole. If the cluster doesn’t break up, one pass with -pbc nojump should do the trick, if I am not mistaken.
  4. That is up to you to decide and depends on your goals.
  5. I don’t understand how fitting would suppress fluctuations, apart from the overall COM position and rotation. But fitting is problematic for any analysis involving the solvent.
1 Like
3

Thank you!