RMSD: a number that means nothing without context

We've all seen it in results sections: "RMSD = 1.8". But that single number hides several assumptions worth stating explicitly.

1. Units. Most tools report in Ångströms, but not all — GROMACS, for instance, outputs in nanometres. If you don't specify, "1.8" is ambiguous. State the unit.

2. Which atoms? When superposing protein structures, RMSD changes with the atom set: Cα only, backbone (N, Cα, C), main chain (N, Cα, C, O), or all atoms. The same two structures give very different values across these. PyMOL's own docs put it plainly: without saying which atoms you used, an RMSD value is meaningless.

3. Length matters. RMSD between two 50-residue proteins isn't comparable to RMSD between two 300-residue ones — the metric is sensitive to the number of atoms compared. This is why size-normalised measures (Carugo & Pongor's RMSD100; Maiorov & Crippen's ρ) were proposed, so values across proteins of different lengths can be compared.

4. Beware silent outlier rejection. This one bites people. Many alignment tools don't use all your atoms — they iteratively prune pairs that sit too far apart and report the RMSD over what survives. Chimera/ChimeraX MatchMaker, by default, removes residue pairs until none exceed a cutoff (2 Å by default), and PyMOL's align rejects outliers over several refinement cycles unless you set cycles=0. The more you prune, the lower the RMSD looks — so a tiny RMSD over a "structurally similar core" can hide a much larger deviation across the full structure. Always report how many atoms were actually used versus the total.

5. Small-molecule poses carry the same traps, plus more. A 10-atom ligand and a 50-atom one don't weigh equally in the average. Two extra pitfalls: symmetry — topologically equivalent atoms can artificially inflate RMSD unless symmetry-corrected RMSD is used — and superposition vs in-place: docking pose RMSD is usually computed in the receptor frame without re-fitting, which is not the same number as an optimally-superposed RMSD.

The takeaway: RMSD is only meaningful when you report what it was calculated on — unit, atom selection, normalisation, pruning, and whether superposition was applied. A one-line caption fixes almost all the ambiguity.

What do you think? Am I missing something?