[us-commits] [ehb54/ultrascan3] d57f01: Remove the psv/molvol estimate from generated entries

[emre brookes]

  Branch: refs/heads/ehb54-issue-978
  Home:   https://github.com/ehb54/ultrascan3
  Commit: d57f0171950500cb1161cc86fdfac4738c9574b4
      https://github.com/ehb54/ultrascan3/commit/d57f0171950500cb1161cc86fdfac4738c9574b4
  Author: ehb54 <brookes at uthscsa.edu>
  Date:   2026-07-21 (Tue, 21 Jul 2026)

  Changed paths:
    M us_somo/develop/perceiver/DECISIONS.md
    M us_somo/develop/src/us_hydrodyn_perceive.cpp

  Log Message:
  -----------
  Remove the psv/molvol estimate from generated entries

The tentative entry emitted a vbar (and molvol) from a sum of vdW spheres.
Fitting a group-contribution model over the ~100 polyatomic residues
somo.residue codes, scored leave-one-out (perceiver/tools/psv_model.py),
shows that estimate is the WORST option available: mean absolute error 0.193
in vbar, against 0.110 for simply assuming the global mean. Summing spheres
ignores bond overlap and runs systematically high.

vbar enters hydrodynamics through the buoyancy term (1 - vbar*rho), so
emitting a number worse than a guess is actively misleading. Both fields are
now left at 0 = unset, and the entry says why and points at
ehb54/ultrascan-tickets#980, where the tiered replacement is tracked
(group-contribution for organic residues, MAE 0.036; residue-type group
average for out-of-domain cases such as metal-containing residues, where the
fit degrades to 0.222 - worse than that average; always user-overridable).

0 rather than omission because the header must keep its 7 fields to parse,
0 reads unambiguously as unset (the convention somo.saxs_atoms already uses
for F and SE), and it fails loudly: a 0 vbar is obviously wrong if used,
whereas a plausible-looking 0.80 is not.

The entry now reports the residue mass instead, since what decides whether an
imprecise psv matters is its mass fraction of the model - molecular vbar is
mass-weighted, so a 616 Da heme in a 50 kDa protein shifts it ~0.5% even with
a 0.34 vbar error.

Also tightens the REVIEW block to genuinely uncertain atoms: a confident
classification carrying an informational note was being listed as uncertain,
disagreeing with the flagged count callers report.

Co-Authored-By: Claude Opus 4.8 <noreply at anthropic.com>



To unsubscribe from these emails, change your notification settings at https://github.com/ehb54/ultrascan3/settings/notifications